Merge remote-tracking branch 'origin/main' into codex/fix-text-input-bounds

fix(ci): restore eval dispatch PR comments

.github/CODEOWNERS

@@ -1,4 +0,0 @@
-# Default code owners for all files
-# These users will be automatically requested for review on PRs.
-
-* @DingTalk-Real-AI/cli-maintainers

.github/PULL_REQUEST_TEMPLATE.md

@@ -3,15 +3,41 @@
 - What changed?
 - Why is this change needed?
 
+## Risk tier
+
+- [ ] Documentation-only: prose/assets only; no executable, generated, workflow,
+  packaging, or interface behavior changed
+- [ ] Standard: ordinary implementation change with a stable package graph
+- [ ] High-risk: workflow/policy, package graph, generated Schema/registry,
+  platform, auth/keychain, installer, packaging, release, transport, recovery,
+  or another fail-closed infrastructure change
+
 ## Verification
 
-- [ ] `make build`
-- [ ] `make lint`
-- [ ] `make test`
-- [ ] `make policy`
+Record the smallest targeted evidence that proves the changed behavior. Do not
+repeat the entire CI suite locally only to fill this checklist: CI expands the
+selected tier from documentation checks, through affected-package tests, to
+the complete high-risk suite.
+
+- [ ] Exact in-place `CHANGELOG.md`-only check (otherwise `N/A`):
+  `./scripts/policy/check-changelog-pr.sh --fast-path "$(git merge-base HEAD origin/main)" HEAD`
+- [ ] Targeted test/check commands and results:
+- [ ] Behavior evidence (test name, CLI output shape, or before/after result):
+- [ ] Documentation links/content/rendering checked (documentation-only, otherwise
+  `N/A`)
+- [ ] Full local suite run because the change is high-risk (optional for other
+  tiers; record command/result or `N/A`)
 - [ ] `./scripts/policy/check-generated-drift.sh`
+  (when generator inputs or generated artifacts may change)
 - [ ] `./scripts/policy/check-command-surface.sh --strict` (if command surface changed)
+- [ ] `./scripts/release/verify-package-managers.sh`
+  (after `make package`, if packaging or installer surfaces changed)
 
 ## Notes
 
 - Any risks, follow-up work, or intentional scope cuts
+
+The repository automatically requests one eligible peer reviewer, including
+after a new head push when another review is needed. Once the latest push has
+peer approval and all nine required checks are current and green, auto-merge
+completes the PR; authors do not need to coordinate a separate routine merge.

.github/actionlint.yaml

@@ -0,0 +1,6 @@
+paths:
+  .github/workflows/release.yml:
+    ignore:
+      # GitHub Actions added concurrency.queue in 2026. actionlint v1.7.12's
+      # bundled workflow schema has not caught up with the platform syntax.
+      - 'unexpected key "queue" for "concurrency" section'

.github/badges/coverage.svg

@@ -1 +1 @@
-<svg xmlns="http://www.w3.org/2000/svg" width="108" height="20" role="img" aria-label="coverage: 54.2%"><title>coverage: 54.2%</title><filter id="blur"><feGaussianBlur in="SourceGraphic" stdDeviation="16"/></filter><linearGradient id="s" x2="0" y2="100%"><stop offset="0" stop-color="#bbb" stop-opacity=".1"/><stop offset="1" stop-opacity=".1"/></linearGradient><clipPath id="r"><rect width="108" height="20" rx="3" fill="#fff"/></clipPath><g clip-path="url(#r)"><rect width="61" height="20" fill="#555"/><rect x="61" width="47" height="20" fill="#dd4343"/><rect width="108" height="20" fill="url(#s)"/></g><g fill="#fff" text-anchor="middle" font-family="Verdana,Geneva,DejaVu Sans,sans-serif" text-rendering="geometricPrecision" font-size="110"><text aria-hidden="true" x="315" y="150" fill="#010101" fill-opacity=".80" filter="url(#blur)" transform="scale(.1)" textLength="510">coverage</text><text aria-hidden="true" x="315" y="150" fill="#010101" fill-opacity=".3" transform="scale(.1)" textLength="510">coverage</text><text x="315" y="140" transform="scale(.1)" fill="#fff" textLength="510">coverage</text><text aria-hidden="true" x="835" y="150" fill="#010101" fill-opacity=".80" filter="url(#blur)" transform="scale(.1)" textLength="370">54.2%</text><text aria-hidden="true" x="835" y="150" fill="#010101" fill-opacity=".3" transform="scale(.1)" textLength="370">54.2%</text><text x="835" y="140" transform="scale(.1)" fill="#fff" textLength="370">54.2%</text></g></svg>
+<svg xmlns="http://www.w3.org/2000/svg" width="114" height="20" role="img" aria-label="coverage: 100.0%"><title>coverage: 100.0%</title><filter id="blur"><feGaussianBlur stdDeviation="16"/></filter><linearGradient id="s" x2="0" y2="100%"><stop offset="0" stop-color="#bbb" stop-opacity=".1"/><stop offset="1" stop-opacity=".1"/></linearGradient><clipPath id="r"><rect width="114" height="20" rx="3"/></clipPath><g clip-path="url(#r)"><rect width="61" height="20" fill="#555"/><rect x="61" width="53" height="20" fill="#4b0"/><rect width="114" height="20" fill="url(#s)"/></g><g fill="#fff" text-anchor="middle" font-family="Verdana,Geneva,DejaVu Sans,sans-serif" text-rendering="geometricPrecision" font-size="110"><g transform="scale(.1)"><g aria-hidden="true" fill="#010101"><text x="315" y="150" fill-opacity=".8" filter="url(#blur)" textLength="510">coverage</text><text x="315" y="150" fill-opacity=".3" textLength="510">coverage</text></g><text x="315" y="140" textLength="510">coverage</text></g><g transform="scale(.1)"><g aria-hidden="true" fill="#010101"><text x="865" y="150" fill-opacity=".8" filter="url(#blur)" textLength="430">100.0%</text><text x="865" y="150" fill-opacity=".3" textLength="430">100.0%</text></g><text x="865" y="140" textLength="430">100.0%</text></g></g></svg>

.github/eval-allowlist.txt

@@ -0,0 +1,61 @@
+# /eval 自助触发允许名单
+#
+# 名单内的 GitHub 登录名可对【自己创建的 PR】触发 /eval 评测；
+# 对任意 PR 触发仍需仓库 write/maintain/admin 权限（维护者背书）。
+# 授权读取的始终是默认分支上的本文件，PR 无法修改自身授权。
+#
+# 变更本文件必须走 PR 评审。每行一个 GitHub login，# 开头为注释。
+
+aftersss
+notable-open
+EdgarWang0925
+ayunya
+yutongshe
+qingyang1014
+caiTriumph
+xlb1130
+Anonymity-0
+FuShu-Yang
+guimingyue
+AlwaysLee
+TaoJikun
+zengyoulingzyl-stack
+liyuan333
+huangyoo
+lifeihong
+nitonitori
+cywan1998
+gangwn
+junlonghuo2
+aqruan
+Freda0909
+ShawnWhite777
+PeterGuy326
+abucraft
+pengzhihan47-star
+rainyak8
+gongrongyun
+huangyuanzhuo-coder
+ybcstudy
+bigqy
+liwang-ai
+meng93
+wxianfeng
+Patrick-Star-CN
+rossluo28-hz
+dxy704330469
+gtezg30062
+Neige-Premaire
+zhuoyu20
+avicii-chen
+typefield
+Haofeng0705
+Huwenjiao
+liuzeyang
+maoqxxmm
+FloralTide
+lingyun9833
+dxb121
+C0922
+xiaoji121
+H3java

.github/reviewer-routing.js

@@ -0,0 +1,285 @@
+'use strict';
+
+// 评审归属是受保护分支上的声明式规则；未知路径不猜测，交给工作流负载均衡兜底。
+const REVIEWER_POOL = ['wxianfeng', 'typefield', 'haofeng0705', 'hlzjsong'];
+
+const PRODUCT_GROUPS = [
+  {
+    primary: 'wxianfeng',
+    backup: 'typefield',
+    products: ['chat', 'contact', 'ding', 'event', 'mail', 'live', 'conference', 'dev', 'devapp', 'mcp', 'aiapp'],
+  },
+  {
+    primary: 'typefield',
+    backup: 'wxianfeng',
+    products: ['doc', 'drive', 'wiki', 'markdown', 'docparse', 'aidesign', 'devdoc', 'blackboard', 'finance', 'law', 'credit'],
+  },
+  {
+    primary: 'haofeng0705',
+    backup: 'typefield',
+    products: ['minutes', 'sheet', 'aitable', 'calendar', 'todo', 'oa', 'attendance', 'report', 'agoal', 'aisearch', 'yida', 'hrbrain'],
+  },
+];
+
+const pathStartsWith = (prefixes) => (path) => prefixes.some((prefix) => path.startsWith(prefix));
+
+const MODULES = [
+  {
+    id: 'security',
+    label: '登录、认证、权限、安全',
+    primary: 'hlzjsong',
+    backup: 'typefield',
+    requiresSecondary: true,
+    matches: pathStartsWith([
+      'internal/auth/',
+      'internal/keychain/',
+      'internal/audit/',
+      'internal/pat/',
+      'internal/security/',
+      'internal/safety/',
+      'pkg/edition/',
+    ]),
+  },
+  {
+    id: 'delivery',
+    label: 'CI、测试、发布、安装',
+    primary: 'haofeng0705',
+    backup: 'wxianfeng',
+    requiresSecondary: true,
+    matches: (path) =>
+      path.startsWith('.github/') ||
+      path.startsWith('scripts/release/') ||
+      path.startsWith('scripts/policy/') ||
+      path.startsWith('scripts/dev/') ||
+      path.startsWith('scripts/install') ||
+      path.startsWith('Formula/') ||
+      path.startsWith('build/') ||
+      path.startsWith('internal/upgrade/') ||
+      path.startsWith('internal/app/upgrade') ||
+      path.startsWith('test/') ||
+      path.startsWith('verify/') ||
+      path.startsWith('.workflow/') ||
+      path === 'coverage.txt' ||
+      path === 'coverage-base.txt' ||
+      path === '.goreleaser.yaml' ||
+      path === 'package.json' ||
+      path === 'package-lock.json' ||
+      path === 'docs/releasing.md',
+  },
+  {
+    id: 'architecture',
+    label: 'DWS 架构、公共内核',
+    primary: 'wxianfeng',
+    backup: 'typefield',
+    requiresSecondary: true,
+    matches: pathStartsWith([
+      'cmd/',
+      'internal/apiclient/',
+      'internal/app/',
+      'internal/cli/',
+      'internal/cobracmd/',
+      'internal/corecmd/',
+      'internal/errors/',
+      'internal/executor/',
+      'internal/generator/',
+      'internal/i18n/',
+      'internal/interfacesnapshot/',
+      'internal/jsonutil/',
+      'internal/localio/',
+      'internal/logging/',
+      'internal/output/',
+      'internal/pipeline/',
+      'internal/plugin/',
+      'internal/profilectx/',
+      'internal/registry/',
+      'internal/syncdata/',
+      'internal/testseam/',
+      'internal/transport/',
+      'pkg/',
+    ]),
+  },
+  {
+    id: 'compatibility',
+    label: '兼容性',
+    primary: 'wxianfeng',
+    backup: 'typefield',
+    requiresSecondary: true,
+    matches: (path) =>
+      /(?:^|[/_.-])compat(?:ibility)?(?=$|[/_.-])/.test(path) ||
+      path.includes('schema_compat'),
+  },
+];
+
+function productMatches(path, product) {
+  const aliases = product === 'blackboard' ? ['blackboard', 'whiteboard'] : [product];
+  return aliases.some((alias) => new RegExp(`(?:^|[/_.-])${alias}(?=$|[/_.-])`).test(path));
+}
+
+const PRODUCT_MODULES = PRODUCT_GROUPS.flatMap((group) =>
+  group.products.map((product) => ({
+    id: `product:${product}`,
+    label: `产品：${product}`,
+    primary: group.primary,
+    backup: group.backup,
+    requiresSecondary: false,
+    matches: (path) => productMatches(path, product),
+  })),
+);
+
+const ALL_MODULES = [MODULES[0], MODULES[1], ...PRODUCT_MODULES, MODULES[2], MODULES[3]];
+
+function normalizedPaths(file) {
+  return [file?.filename, file?.previous_filename]
+    .filter((path) => typeof path === 'string' && path !== '')
+    .map((path) => path.toLowerCase());
+}
+
+function compareStats(left, right) {
+  return right.files - left.files || left.module.order - right.module.order || left.module.id.localeCompare(right.module.id);
+}
+
+function classifyFiles(files) {
+  const counts = new Map();
+  for (const file of files || []) {
+    const matchingModules = new Set();
+    for (const path of normalizedPaths(file)) {
+      const matches = ALL_MODULES.filter((module) => module.matches(path));
+      const securityOrDelivery = matches.filter(
+        (module) => module.id === 'security' || module.id === 'delivery',
+      );
+      const effectiveMatches = securityOrDelivery.length > 0
+        ? [...securityOrDelivery, ...matches.filter((module) => module.id === 'compatibility')]
+        : matches;
+      for (const match of effectiveMatches) {
+        matchingModules.add(match.id);
+      }
+      if (
+        effectiveMatches.length === 0 &&
+        (path.startsWith('internal/helpers/') || path.startsWith('internal/shortcut/'))
+      ) {
+        matchingModules.add('architecture');
+      }
+    }
+    for (const moduleID of matchingModules) {
+      counts.set(moduleID, (counts.get(moduleID) || 0) + 1);
+    }
+  }
+
+  return [...counts.entries()]
+    .map(([id, files]) => {
+      const index = ALL_MODULES.findIndex((module) => module.id === id);
+      return {module: {...ALL_MODULES[index], order: index}, files};
+    })
+    .sort(compareStats);
+}
+
+function chooseModuleReviewer(module, unavailable) {
+  return [module.primary, module.backup].find(
+    (reviewer) => REVIEWER_POOL.includes(reviewer) && !unavailable.has(reviewer),
+  );
+}
+
+function addReviewer(reviewers, reviewer) {
+  if (reviewer && !reviewers.includes(reviewer)) {
+    reviewers.push(reviewer);
+  }
+}
+
+function reviewerCandidates({preferredReviewers, fallbackReviewers, eligibleReviewers}) {
+  const eligible = new Set(eligibleReviewers.map((reviewer) => reviewer.toLowerCase()));
+  const candidates = [];
+  for (const reviewer of [...preferredReviewers, ...fallbackReviewers]) {
+    if (
+      eligible.has(reviewer.toLowerCase()) &&
+      !candidates.some((candidate) => candidate.toLowerCase() === reviewer.toLowerCase())
+    ) {
+      candidates.push(reviewer);
+    }
+  }
+  return candidates;
+}
+
+async function requestReviewersWithFallback({
+  candidates,
+  requiredReviewers,
+  satisfiedReviewers = [],
+  requestReviewer,
+  onFailure = () => {},
+}) {
+  const alreadySatisfied = new Set(
+    satisfiedReviewers.map((reviewer) => reviewer.toLowerCase()),
+  );
+  const satisfied = new Set();
+  const requested = [];
+
+  for (const reviewer of candidates) {
+    if (satisfied.size >= requiredReviewers) {
+      break;
+    }
+    const normalizedReviewer = reviewer.toLowerCase();
+    if (alreadySatisfied.has(normalizedReviewer)) {
+      satisfied.add(normalizedReviewer);
+      continue;
+    }
+
+    try {
+      const shouldContinue = await requestReviewer(reviewer);
+      if (shouldContinue === false) {
+        return {requested, satisfiedReviewers: [...satisfied], aborted: true};
+      }
+      requested.push(reviewer);
+      satisfied.add(normalizedReviewer);
+    } catch (error) {
+      onFailure(reviewer, error);
+    }
+  }
+
+  return {requested, satisfiedReviewers: [...satisfied], aborted: false};
+}
+
+function resolveReviewRouting({files, author, latestPusher, fallbackReviewers = REVIEWER_POOL}) {
+  const modules = classifyFiles(files);
+  const unavailable = new Set([author, latestPusher].filter(Boolean).map((login) => login.toLowerCase()));
+  const reviewers = [];
+  const primaryModule = modules[0];
+
+  if (!primaryModule) {
+    return {modules: [], reviewers, requiredReviewers: 1, reason: 'unknown_paths'};
+  }
+
+  addReviewer(reviewers, chooseModuleReviewer(primaryModule.module, unavailable));
+
+  const requiresSecondary =
+    modules.length > 1 || modules.some(({module}) => module.requiresSecondary);
+  const secondaryModule = modules.find(({module}) => module.id !== primaryModule.module.id) || primaryModule;
+  if (requiresSecondary) {
+    addReviewer(
+      reviewers,
+      chooseModuleReviewer(secondaryModule.module, new Set([...unavailable, ...reviewers])),
+    );
+  }
+
+  for (const reviewer of fallbackReviewers) {
+    if (reviewers.length >= (requiresSecondary ? 2 : 1)) {
+      break;
+    }
+    if (REVIEWER_POOL.includes(reviewer) && !unavailable.has(reviewer)) {
+      addReviewer(reviewers, reviewer);
+    }
+  }
+
+  return {
+    modules: modules.map(({module, files}) => ({id: module.id, label: module.label, files})),
+    reviewers,
+    requiredReviewers: requiresSecondary ? 2 : 1,
+    reason: requiresSecondary ? 'cross_or_sensitive' : 'single_module',
+  };
+}
+
+module.exports = {
+  REVIEWER_POOL,
+  classifyFiles,
+  requestReviewersWithFallback,
+  resolveReviewRouting,
+  reviewerCandidates,
+};

.github/reviewer-routing.test.js

@@ -0,0 +1,148 @@
+'use strict';
+
+const assert = require('node:assert/strict');
+const {
+  requestReviewersWithFallback,
+  resolveReviewRouting,
+  reviewerCandidates,
+} = require('./reviewer-routing');
+
+function route(files, author = 'author', latestPusher = author) {
+  return resolveReviewRouting({files: files.map((filename) => ({filename})), author, latestPusher});
+}
+
+{
+  const result = route(['internal/helpers/chat_toolbar.go']);
+  assert.deepEqual(result.reviewers, ['wxianfeng']);
+  assert.equal(result.requiredReviewers, 1);
+  assert.deepEqual(result.modules.map((module) => module.id), ['product:chat']);
+}
+
+{
+  const result = route(['internal/helpers/chat_toolbar.go', 'internal/helpers/doc_style.go']);
+  assert.deepEqual(result.reviewers, ['wxianfeng', 'typefield']);
+  assert.equal(result.requiredReviewers, 2);
+}
+
+{
+  const result = route(['.github/workflows/ci.yml']);
+  assert.deepEqual(result.reviewers, ['haofeng0705', 'wxianfeng']);
+  assert.equal(result.requiredReviewers, 2);
+  assert.equal(result.reason, 'cross_or_sensitive');
+}
+
+{
+  const result = route(['internal/auth/login.go'], 'hlzjsong');
+  assert.deepEqual(result.reviewers, ['typefield', 'wxianfeng']);
+  assert.equal(result.requiredReviewers, 2);
+}
+
+{
+  const result = route(['internal/upgrade/downloader.go']);
+  assert.deepEqual(result.reviewers, ['haofeng0705', 'wxianfeng']);
+  assert.equal(result.requiredReviewers, 2);
+}
+
+{
+  const result = route(['internal/app/upgrade.go', 'scripts/dev/test-release.sh']);
+  assert.deepEqual(result.reviewers, ['haofeng0705', 'wxianfeng']);
+  assert.equal(result.requiredReviewers, 2);
+}
+
+{
+  const result = route(['pkg/edition/edition.go']);
+  assert.deepEqual(result.reviewers, ['hlzjsong', 'typefield']);
+  assert.equal(result.requiredReviewers, 2);
+}
+
+{
+  const result = route(['internal/shortcut/chat/compatibility_coverage_test.go']);
+  assert.deepEqual(result.reviewers, ['wxianfeng', 'typefield']);
+  assert.equal(result.requiredReviewers, 2);
+  assert.deepEqual(result.modules.map((module) => module.id), ['product:chat', 'compatibility']);
+}
+
+{
+  const result = route(['internal/helpers/leaf_dispatch.go']);
+  assert.deepEqual(result.reviewers, ['wxianfeng', 'typefield']);
+  assert.equal(result.requiredReviewers, 2);
+}
+
+{
+  const result = route(['docs/unknown-area.md']);
+  assert.deepEqual(result.reviewers, []);
+  assert.equal(result.reason, 'unknown_paths');
+}
+
+async function testSingleReviewerFallback() {
+  const candidates = reviewerCandidates({
+    preferredReviewers: ['wxianfeng'],
+    fallbackReviewers: ['wxianfeng', 'typefield', 'haofeng0705'],
+    eligibleReviewers: ['wxianfeng', 'typefield', 'haofeng0705'],
+  });
+  const attempts = [];
+  const result = await requestReviewersWithFallback({
+    candidates,
+    requiredReviewers: 1,
+    requestReviewer: async (reviewer) => {
+      attempts.push(reviewer);
+      if (reviewer === 'wxianfeng') {
+        throw Object.assign(new Error('cannot request primary'), {status: 422});
+      }
+      return true;
+    },
+  });
+  assert.deepEqual(attempts, ['wxianfeng', 'typefield']);
+  assert.deepEqual(result.requested, ['typefield']);
+  assert.equal(result.satisfiedReviewers.length, 1);
+}
+
+async function testTwoReviewerFallback() {
+  const candidates = reviewerCandidates({
+    preferredReviewers: ['haofeng0705', 'wxianfeng'],
+    fallbackReviewers: ['haofeng0705', 'wxianfeng', 'typefield', 'hlzjsong'],
+    eligibleReviewers: ['haofeng0705', 'wxianfeng', 'typefield', 'hlzjsong'],
+  });
+  const attempts = [];
+  const result = await requestReviewersWithFallback({
+    candidates,
+    requiredReviewers: 2,
+    requestReviewer: async (reviewer) => {
+      attempts.push(reviewer);
+      if (reviewer === 'wxianfeng') {
+        throw Object.assign(new Error('temporary failure'), {status: 503});
+      }
+      return true;
+    },
+  });
+  assert.deepEqual(attempts, ['haofeng0705', 'wxianfeng', 'typefield']);
+  assert.deepEqual(result.requested, ['haofeng0705', 'typefield']);
+  assert.equal(result.satisfiedReviewers.length, 2);
+}
+
+async function testLowerPriorityExistingRequestDoesNotReplaceOwner() {
+  const attempts = [];
+  const result = await requestReviewersWithFallback({
+    candidates: ['wxianfeng', 'typefield'],
+    requiredReviewers: 1,
+    satisfiedReviewers: ['typefield'],
+    requestReviewer: async (reviewer) => {
+      attempts.push(reviewer);
+      return true;
+    },
+  });
+  assert.deepEqual(attempts, ['wxianfeng']);
+  assert.deepEqual(result.requested, ['wxianfeng']);
+  assert.deepEqual(result.satisfiedReviewers, ['wxianfeng']);
+}
+
+Promise.all([
+  testSingleReviewerFallback(),
+  testTwoReviewerFallback(),
+  testLowerPriorityExistingRequestDoesNotReplaceOwner(),
+])
+  .then(() => console.log('reviewer routing policy tests passed'))
+  .catch((error) => {
+    console.error(error);
+    process.exitCode = 1;
+  });

.github/workflows/ai-behavior-check.yml

@@ -0,0 +1,131 @@
+name: Code Admission — AI Behavior
+
+on:
+  pull_request_target:
+    types: [opened, synchronize, reopened, labeled, unlabeled]
+  push:
+    branches:
+      - main
+
+permissions:
+  contents: read
+  pull-requests: read
+  statuses: write
+
+jobs:
+  ai-behavior-check:
+    name: AI Behavior
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    steps:
+      # Deliberately do not check out or execute pull-request code here.
+      # pull_request_target keeps this policy anchored to the base branch.
+      - name: Check AI-generated PR boundaries
+        uses: actions/github-script@v7
+        with:
+          script: |
+            const pullRequest = context.payload.pull_request;
+            const sha = context.eventName === 'push' ? context.sha : pullRequest.head.sha;
+            const setStatus = (state, description) =>
+              github.rest.repos.createCommitStatus({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                sha,
+                state,
+                context: 'AI Behavior',
+                description,
+              });
+
+            await setStatus('pending', 'Evaluating AI-generated PR boundaries');
+
+            if (context.eventName === 'push') {
+              await setStatus('success', 'Not applicable to the protected main push');
+              core.notice('AI Behavior is a PR policy; the main push context is sealed.');
+              return;
+            }
+
+            try {
+              const expectedHead = pullRequest.head.sha;
+              const expectedBase = pullRequest.base.sha;
+              const currentPull = async (phase) => {
+                const { data: pull } = await github.rest.pulls.get({
+                  owner: context.repo.owner,
+                  repo: context.repo.repo,
+                  pull_number: context.issue.number,
+                });
+                if (pull.head.sha !== expectedHead || pull.base.sha !== expectedBase) {
+                  throw new Error(
+                    `Pull request revision changed during ${phase}: ` +
+                    `expected base/head ${expectedBase}/${expectedHead}, ` +
+                    `got ${pull.base.sha}/${pull.head.sha}`
+                  );
+                }
+                return pull;
+              };
+
+              const before = await currentPull('pre-policy check');
+              const labels = before.labels.map(({ name }) => name);
+              if (!labels.includes('ai-generated')) {
+                await setStatus('success', 'Not labeled ai-generated');
+                core.notice('Not an ai-generated PR; no AI-only policy applied.');
+                return;
+              }
+
+              const files = await github.paginate(github.rest.pulls.listFiles, {
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                pull_number: context.issue.number,
+                per_page: 100,
+              });
+              await currentPull('post-policy check');
+
+              const maxChangedFiles = 30;
+              if (files.length > maxChangedFiles) {
+                await setStatus(
+                  'failure',
+                  `Changes ${files.length} files; limit is ${maxChangedFiles}`
+                );
+                core.setFailed(
+                  `AI-generated PR changes ${files.length} files; limit is ${maxChangedFiles}.`
+                );
+                return;
+              }
+
+              const isProtectedPath = (filename) =>
+                typeof filename === 'string' &&
+                (
+                  filename.startsWith('.github/workflows/') ||
+                  filename.startsWith('scripts/ci/') ||
+                  filename.startsWith('scripts/policy/') ||
+                  filename.startsWith('scripts/release/') ||
+                  filename === 'test/fixtures/cli-interface-baseline.txt' ||
+                  filename === '.goreleaser.yaml' ||
+                  filename === 'Makefile'
+                );
+              const protectedPaths = [...new Set(
+                files
+                  .flatMap(({ filename, previous_filename }) => [filename, previous_filename])
+                  .filter(isProtectedPath)
+              )];
+
+              if (protectedPaths.length > 0) {
+                await setStatus('failure', 'Modifies protected release/CI infrastructure');
+                core.setFailed(
+                  'AI-generated PR modifies protected release/CI infrastructure:\n' +
+                  protectedPaths.map((filename) => `  - ${filename}`).join('\n') +
+                  '\nSplit these changes into a human-owned PR with explicit review.'
+                );
+                return;
+              }
+
+              await setStatus(
+                'success',
+                `Passed with ${files.length} changed files (limit ${maxChangedFiles})`
+              );
+              core.notice(
+                `AI behavior check passed (${files.length} changed files; limit ${maxChangedFiles}).`
+              );
+            } catch (error) {
+              await setStatus('error', 'Could not evaluate the exact pull request revision');
+              throw error;
+            }

.github/workflows/eval-dispatch.yml

@@ -0,0 +1,296 @@
+name: PR Eval Dispatch
+
+# `/eval <products> [sha=<full-head-sha>] [cases=<ref>]` PR 评论 → 生成可验证的评测请求，报告由 bot 回贴。
+# 本 workflow 只在默认分支上下文运行，不 checkout、不执行 PR 代码。
+# 审核 SHA 规则：评测他人 PR 必须显式携带 sha=（审阅背书凭据，验证
+# 其恰为当前 open head）；评测自己创建的 PR 可省略，自动钉住派发时刻
+# 的当前 head（作者自背书，无第三方偷换窗口）；受控评测执行端另以
+# FETCH_HEAD 校验兜底派发后的变更。
+# 授权两级：仓库 write/maintain/admin 可派发任意 PR；默认分支
+# .github/eval-allowlist.txt 名单内的用户仅可派发自己创建的 PR。
+# 触发通道：workflow 先创建占位评论，再上传与本次 run/comment 绑定的
+# 不可变 manifest artifact，最后把 artifact 指针写回同一评论。评论仅是
+# 不可信通知；受控评测服务必须验证成功 run、artifact 与 manifest，并在
+# 触发评测前原子占用 manifest.idempotency_key，重复占用只能 no-op。
+
+on:
+  issue_comment:
+    types:
+      - created
+
+permissions: {}
+
+concurrency:
+  group: eval-dispatch-${{ github.event.issue.number }}
+  cancel-in-progress: false
+
+jobs:
+  dispatch:
+    name: Dispatch internal evaluation
+    if: >-
+      github.event.issue.pull_request &&
+      startsWith(github.event.comment.body, '/eval')
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    permissions:
+      contents: read
+      # 该 job 仅处理 PR；评论写入也限定在 PR Conversation 这一权限域。
+      pull-requests: write
+    steps:
+      - name: Check out default branch tooling
+        uses: actions/checkout@v4
+
+      - name: Verify commenter dispatch authorization
+        env:
+          GH_TOKEN: ${{ github.token }}
+          COMMENTER: ${{ github.event.comment.user.login }}
+          PR_AUTHOR: ${{ github.event.issue.user.login }}
+          EVAL_ALLOWLIST_PATH: .github/eval-allowlist.txt
+        run: |
+          # 不用 --fail：非协作者查权限返回 404 错误体，交由 guard 走名单分支；硬网络错误降级为空对象同样 fail-closed
+          permission_json="$(curl --silent --show-error \
+            -H "Authorization: Bearer ${GH_TOKEN}" \
+            -H "Accept: application/vnd.github+json" \
+            "https://api.github.com/repos/${GITHUB_REPOSITORY}/collaborators/${COMMENTER}/permission")" || permission_json='{}'
+          printf '%s' "$permission_json" | python3 scripts/ci/eval_dispatch_guard.py permission
+
+      - name: Parse /eval command
+        id: parse
+        continue-on-error: true
+        env:
+          COMMENT_BODY: ${{ github.event.comment.body }}
+        run: python3 scripts/ci/eval_comment_parse.py
+
+      - name: Reply usage on parse failure
+        if: steps.parse.outcome == 'failure'
+        env:
+          GH_TOKEN: ${{ github.token }}
+          PR_NUMBER: ${{ github.event.issue.number }}
+          PARSE_ERROR: ${{ steps.parse.outputs.error }}
+        run: |
+          body="❌ /eval 命令解析失败：${PARSE_ERROR}"
+          gh api --method POST \
+            "repos/${GITHUB_REPOSITORY}/issues/${PR_NUMBER}/comments" \
+            --raw-field body="$body" \
+            > /dev/null
+          exit 1
+
+      - name: Verify reviewed PR head
+        id: pr
+        env:
+          GH_TOKEN: ${{ github.token }}
+          PR_NUMBER: ${{ github.event.issue.number }}
+          EXPECTED_PR_NUMBER: ${{ github.event.issue.number }}
+          REVIEWED_SHA: ${{ steps.parse.outputs.reviewed_sha }}
+          COMMENTER: ${{ github.event.comment.user.login }}
+        run: |
+          pr_json="$(curl --fail --silent --show-error \
+            -H "Authorization: Bearer ${GH_TOKEN}" \
+            -H "Accept: application/vnd.github+json" \
+            "https://api.github.com/repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}")"
+          printf '%s' "$pr_json" \
+            | python3 scripts/ci/eval_dispatch_guard.py head \
+            >> "$GITHUB_OUTPUT"
+
+      - name: Create dispatch placeholder
+        id: placeholder
+        env:
+          GH_TOKEN: ${{ github.token }}
+          PR_NUMBER: ${{ github.event.issue.number }}
+        run: |
+          set -euo pipefail
+          placeholder_body="🛰️ /eval 请求已通过权限与版本校验，正在生成可验证的评测请求。"
+          response="$(
+            gh api --method POST \
+              "repos/${GITHUB_REPOSITORY}/issues/${PR_NUMBER}/comments" \
+              --raw-field body="$placeholder_body"
+          )"
+          comment_id="$(
+            printf '%s' "$response" \
+              | jq -er \
+                --arg issue_url "https://api.github.com/repos/${GITHUB_REPOSITORY}/issues/${PR_NUMBER}" \
+                'select(.issue_url == $issue_url) | .id | tostring | select(test("^[1-9][0-9]*$"))'
+          )"
+          printf 'comment_id=%s\n' "$comment_id" >> "$GITHUB_OUTPUT"
+
+      - name: Build dispatch request manifest
+        env:
+          REPOSITORY_ID: '1187709537'
+          REPOSITORY: ${{ github.repository }}
+          WORKFLOW_ID: '331725458'
+          WORKFLOW_PATH: .github/workflows/eval-dispatch.yml
+          RUN_ID: ${{ github.run_id }}
+          RUN_ATTEMPT: ${{ github.run_attempt }}
+          SOURCE_COMMENT_ID: ${{ github.event.comment.id }}
+          DISPATCH_COMMENT_ID: ${{ steps.placeholder.outputs.comment_id }}
+          ACTOR_ID: ${{ github.event.comment.user.id }}
+          ACTOR_LOGIN: ${{ github.event.comment.user.login }}
+          PR_NUMBER: ${{ github.event.issue.number }}
+          PR_HEAD_SHA: ${{ steps.pr.outputs.head_sha }}
+          PRODUCTS: ${{ steps.parse.outputs.products }}
+          CASES_REF: ${{ steps.parse.outputs.cases_ref }}
+          SOURCE_BODY: ${{ github.event.comment.body }}
+          MANIFEST_PATH: ${{ runner.temp }}/eval-dispatch-request.json
+        run: |
+          set -euo pipefail
+          if [ "$REPOSITORY" != "DingTalk-Real-AI/dingtalk-workspace-cli" ]; then
+            echo "unexpected repository: ${REPOSITORY}" >&2
+            exit 1
+          fi
+          for value in \
+            "$REPOSITORY_ID" \
+            "$WORKFLOW_ID" \
+            "$RUN_ID" \
+            "$RUN_ATTEMPT" \
+            "$SOURCE_COMMENT_ID" \
+            "$DISPATCH_COMMENT_ID" \
+            "$ACTOR_ID" \
+            "$PR_NUMBER"; do
+            if [[ ! "$value" =~ ^[1-9][0-9]*$ ]]; then
+              echo "dispatch manifest contains a non-canonical identifier" >&2
+              exit 1
+            fi
+          done
+          if [[ ! "$PR_HEAD_SHA" =~ ^[0-9a-f]{40}$ ]]; then
+            echo "dispatch manifest contains an invalid PR head SHA" >&2
+            exit 1
+          fi
+
+          hash_output="$(printf '%s' "$SOURCE_BODY" | sha256sum)"
+          source_body_sha256="${hash_output%% *}"
+          if [[ ! "$source_body_sha256" =~ ^[0-9a-f]{64}$ ]]; then
+            echo "failed to hash source comment" >&2
+            exit 1
+          fi
+          idempotency_key="${REPOSITORY_ID}:${SOURCE_COMMENT_ID}"
+
+          umask 077
+          jq -n \
+            --arg repository_id "$REPOSITORY_ID" \
+            --arg repository "$REPOSITORY" \
+            --arg workflow_id "$WORKFLOW_ID" \
+            --arg workflow_path "$WORKFLOW_PATH" \
+            --arg run_id "$RUN_ID" \
+            --arg run_attempt "$RUN_ATTEMPT" \
+            --arg source_comment_id "$SOURCE_COMMENT_ID" \
+            --arg dispatch_comment_id "$DISPATCH_COMMENT_ID" \
+            --arg actor_id "$ACTOR_ID" \
+            --arg actor_login "$ACTOR_LOGIN" \
+            --arg pr_number "$PR_NUMBER" \
+            --arg pr_head_sha "$PR_HEAD_SHA" \
+            --arg products "$PRODUCTS" \
+            --arg cases_ref "$CASES_REF" \
+            --arg source_body_sha256 "$source_body_sha256" \
+            --arg idempotency_key "$idempotency_key" \
+            '{
+              schema_version: 1,
+              repository_id: $repository_id,
+              repository: $repository,
+              workflow_id: $workflow_id,
+              workflow_path: $workflow_path,
+              run_id: $run_id,
+              run_attempt: $run_attempt,
+              source_comment_id: $source_comment_id,
+              dispatch_comment_id: $dispatch_comment_id,
+              actor_id: $actor_id,
+              actor_login: $actor_login,
+              pr_number: $pr_number,
+              pr_head_sha: $pr_head_sha,
+              products: $products,
+              cases_ref: $cases_ref,
+              source_body_sha256: $source_body_sha256,
+              idempotency_key: $idempotency_key
+            }' > "$MANIFEST_PATH"
+
+      - name: Upload dispatch request manifest
+        id: artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: eval-dispatch-request-${{ github.run_id }}-${{ github.run_attempt }}-${{ steps.placeholder.outputs.comment_id }}
+          path: ${{ runner.temp }}/eval-dispatch-request.json
+          if-no-files-found: error
+          retention-days: 1
+          overwrite: false
+
+      - name: Finalize dispatch marker
+        env:
+          GH_TOKEN: ${{ github.token }}
+          DISPATCH_COMMENT_ID: ${{ steps.placeholder.outputs.comment_id }}
+          REPOSITORY_ID: '1187709537'
+          WORKFLOW_ID: '331725458'
+          WORKFLOW_PATH: .github/workflows/eval-dispatch.yml
+          RUN_ID: ${{ github.run_id }}
+          RUN_ATTEMPT: ${{ github.run_attempt }}
+          ARTIFACT_ID: ${{ steps.artifact.outputs.artifact-id }}
+          ARTIFACT_DIGEST: ${{ steps.artifact.outputs.artifact-digest }}
+          PR_HEAD_SHA: ${{ steps.pr.outputs.head_sha }}
+          PRODUCTS: ${{ steps.parse.outputs.products }}
+          CASES_REF: ${{ steps.parse.outputs.cases_ref }}
+        run: |
+          set -euo pipefail
+          if [[ ! "$DISPATCH_COMMENT_ID" =~ ^[1-9][0-9]*$ ]] || \
+             [[ ! "$ARTIFACT_ID" =~ ^[1-9][0-9]*$ ]]; then
+            echo "artifact marker contains a non-canonical identifier" >&2
+            exit 1
+          fi
+
+          artifact_digest="${ARTIFACT_DIGEST,,}"
+          if [[ "$artifact_digest" != sha256:* ]]; then
+            artifact_digest="sha256:${artifact_digest}"
+          fi
+          if [[ ! "$artifact_digest" =~ ^sha256:[0-9a-f]{64}$ ]]; then
+            echo "artifact marker contains an invalid digest" >&2
+            exit 1
+          fi
+
+          marker_json="$(
+            jq -nc \
+              --arg repository_id "$REPOSITORY_ID" \
+              --arg workflow_id "$WORKFLOW_ID" \
+              --arg workflow_path "$WORKFLOW_PATH" \
+              --arg run_id "$RUN_ID" \
+              --arg run_attempt "$RUN_ATTEMPT" \
+              --arg dispatch_comment_id "$DISPATCH_COMMENT_ID" \
+              --arg artifact_id "$ARTIFACT_ID" \
+              --arg artifact_digest "$artifact_digest" \
+              '{
+                schema_version: 1,
+                repository_id: $repository_id,
+                workflow_id: $workflow_id,
+                workflow_path: $workflow_path,
+                run_id: $run_id,
+                run_attempt: $run_attempt,
+                dispatch_comment_id: $dispatch_comment_id,
+                artifact_id: $artifact_id,
+                artifact_digest: $artifact_digest
+              }'
+          )"
+          cases_note=""
+          if [ -n "$CASES_REF" ]; then
+            cases_note="，用例版本 \`${CASES_REF}\`"
+          fi
+          body="<!-- eval-dispatch: ${marker_json} -->"$'\n'"🛰️ /eval 已受理：产品集 \`${PRODUCTS}\`${cases_note}，评测对象 \`${PR_HEAD_SHA}\`。"$'\n'"受控评测服务将在数分钟内处理，完成后由 bot 回贴报告。"
+          response="$(
+            gh api --method PATCH \
+              "repos/${GITHUB_REPOSITORY}/issues/comments/${DISPATCH_COMMENT_ID}" \
+              --raw-field body="$body"
+          )"
+          printf '%s' "$response" \
+            | jq -e \
+              --arg comment_id "$DISPATCH_COMMENT_ID" \
+              --arg body "$body" \
+              '((.id | tostring) == $comment_id) and (.body == $body)' \
+              > /dev/null
+
+      - name: Mark dispatch preparation failure
+        if: ${{ failure() && steps.placeholder.outputs.comment_id != '' }}
+        env:
+          GH_TOKEN: ${{ github.token }}
+          DISPATCH_COMMENT_ID: ${{ steps.placeholder.outputs.comment_id }}
+        run: |
+          failure_body="❌ /eval 请求准备失败，未生成可消费的评测请求。请稍后重试。"
+          gh api --method PATCH \
+            "repos/${GITHUB_REPOSITORY}/issues/comments/${DISPATCH_COMMENT_ID}" \
+            --raw-field body="$failure_body" \
+            > /dev/null \
+            || true

.github/workflows/mirror-to-gitee.yml

@@ -1,4 +1,5 @@
-# 把本仓库代码自动镜像到 Gitee，供国内用户访问（raw 脚本入口 + tags）。
+# 把本仓库 main 代码自动镜像到 Gitee，供国内用户访问 raw 脚本入口。
+# Release tag 与附件只由 release.yml 的受控 publication queue 发布。
 # 用 HTTPS + 令牌直接 git push（无需 SSH key），复用已配置的 secret：
 #   GITEE_TOKEN  —— Gitee 私人令牌（勾 projects）
 #   GITEE_USER   —— 令牌所属 Gitee 用户名（用于 https 推送鉴权）
@@ -10,11 +11,14 @@ on:
   push:
     branches:
       - main
-    tags:
-      - 'v*'
   schedule:
     - cron: '0 18 * * *'
   workflow_dispatch:
+    inputs:
+      sync_release_version:
+        description: "Sync a specific release version's assets to Gitee (e.g. v1.0.55-beta.3)"
+        required: false
+        type: string
 
 concurrency:
   group: gitee-code-mirror
@@ -23,13 +27,13 @@ concurrency:
 jobs:
   mirror:
     runs-on: ubuntu-latest
-    # GitHub Actions 不允许在 job-level if 直接引用 secrets，故先用 env 暴露再在 step 守卫。
+    if: ${{ github.ref_name == github.event.repository.default_branch && github.repository_owner == 'DingTalk-Real-AI' }}
     env:
       GITEE_TOKEN: ${{ secrets.GITEE_TOKEN }}
       GITEE_USER: ${{ secrets.GITEE_USER }}
       GITEE_REPO: ${{ secrets.GITEE_REPO }}
     steps:
-      - name: Checkout (full history + tags)
+      - name: Checkout main history
         if: env.GITEE_TOKEN != ''
         uses: actions/checkout@v4
         with:
@@ -41,15 +45,7 @@ jobs:
           set -eu
           REMOTE="https://${GITEE_USER}:${GITEE_TOKEN}@gitee.com/${GITEE_REPO}.git"
 
-          if [ "${GITHUB_REF_TYPE:-}" = "tag" ]; then
-            git fetch --force --tags origin "refs/tags/${GITHUB_REF_NAME}:refs/tags/${GITHUB_REF_NAME}"
-            git push --force "$REMOTE" "refs/tags/${GITHUB_REF_NAME}:refs/tags/${GITHUB_REF_NAME}"
-            echo "✅ 已镜像 tag ${GITHUB_REF_NAME} 到 Gitee ${GITEE_REPO}"
-            exit 0
-          fi
-
-          # 取到 main 与所有 tag（落到 origin/* 与本地 tags，避免推当前分支引用冲突）
-          git fetch --force --tags origin 'refs/heads/main:refs/remotes/origin/main'
+          git fetch --force origin 'refs/heads/main:refs/remotes/origin/main'
 
           # Gitee 专属分支：在 origin/main 之上叠加一个 README 本地化 commit。
           # GitHub 那份 README 不变；只有推往 Gitee 的副本被改写。
@@ -81,7 +77,45 @@ jobs:
           git add README.md README_zh.md 2>/dev/null || true
           git commit -m "docs(gitee): localize install commands + coverage badge for Gitee mirror" || true
 
-          # 镜像对齐（force：Gitee 始终跟随 GitHub + Gitee 专属 README 本地化）
+          # main 镜像对齐；release tag 由 release.yml 单独校验后创建，禁止在这里 force。
           git push --force "$REMOTE" 'gitee-main:refs/heads/main'
-          git push --force --tags "$REMOTE"
-          echo "✅ 已镜像 main(+Gitee README 本地化) + tags 到 Gitee ${GITEE_REPO}"
+          echo "✅ 已镜像 main（含 Gitee README 本地化）到 Gitee ${GITEE_REPO}"
+
+      - name: Download GitHub Release assets
+        if: ${{ inputs.sync_release_version != '' }}
+        env:
+          VERSION: ${{ inputs.sync_release_version }}
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          set -eu
+          echo "📥 Downloading release assets for ${VERSION}"
+          mkdir -p dist
+          gh release download "$VERSION" \
+            --repo "$GITHUB_REPOSITORY" \
+            --dir dist \
+            --pattern 'dws-*' \
+            --pattern 'checksums.txt' \
+            --clobber
+          ls -la dist/
+
+      - name: Verify release artifacts
+        if: ${{ inputs.sync_release_version != '' }}
+        env:
+          VERSION: ${{ inputs.sync_release_version }}
+        run: |
+          set -eu
+          DWS_PACKAGE_DIST_DIR="$GITHUB_WORKSPACE/dist" \
+            ./scripts/release/verify-release-artifacts.sh "$VERSION"
+
+      - name: Sync release assets to Gitee
+        if: ${{ inputs.sync_release_version != '' }}
+        env:
+          VERSION: ${{ inputs.sync_release_version }}
+          GITEE_TOKEN: ${{ secrets.GITEE_TOKEN }}
+          GITEE_USER: ${{ secrets.GITEE_USER }}
+          GITEE_REPO: ${{ secrets.GITEE_REPO }}
+          DIST_DIR: ${{ github.workspace }}/dist
+        run: |
+          set -eu
+          echo "📦 Syncing release assets for ${VERSION} to Gitee ${GITEE_REPO}"
+          ./scripts/release/sync-to-gitee.sh

.github/workflows/multi-profile-e2e.yml

@@ -1,8 +1,9 @@
-name: Multi Profile E2E
+name: Main Integration — 主干集成
 
 on:
-  pull_request:
   push:
+    branches:
+      - main
   workflow_dispatch:
 
 permissions:
@@ -14,7 +15,7 @@ concurrency:
 
 jobs:
   multi-profile-e2e:
-    name: Multi Profile E2E
+    name: Multi-profile E2E
     runs-on: ubuntu-latest
     timeout-minutes: 15
     env:
@@ -36,7 +37,7 @@ jobs:
           mkdir -p .tmp-bin
           bash scripts/dev/test-multi-profile-e2e.sh --keep-workdir | tee "$MULTI_PROFILE_E2E_LOG"
           {
-            echo "### Multi Profile E2E"
+            echo "### Multi-profile E2E"
             echo "- Command: \`bash scripts/dev/test-multi-profile-e2e.sh --keep-workdir\`"
             echo "- Scope: isolated auth/profile storage, profile switch/use, one-shot profile override, CSV multi-profile aggregation, legacy migration"
             echo "- Result: passed"
@@ -50,5 +51,6 @@ jobs:
           path: |
             .tmp-bin/multi-profile-e2e.*/out
             .tmp-bin/multi-profile-e2e.log
+          include-hidden-files: true
           if-no-files-found: ignore
           retention-days: 3

.github/workflows/notify-wukong.yml

@@ -0,0 +1,38 @@
+name: Main Integration — Wukong Overlay
+
+on:
+  workflow_run:
+    workflows:
+      - CI
+    types:
+      - completed
+
+permissions: {}
+
+jobs:
+  notify-downstream:
+    name: Notify Wukong Overlay
+    if: >-
+      github.event.workflow_run.conclusion == 'success' &&
+      github.event.workflow_run.event == 'push' &&
+      github.event.workflow_run.head_branch == 'main'
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    steps:
+      - name: Trigger downstream CI
+        env:
+          UPSTREAM_SHA: ${{ github.event.workflow_run.head_sha }}
+          WUKONG_TRIGGER_TOKEN: ${{ secrets.WUKONG_TRIGGER_TOKEN }}
+          WUKONG_TRIGGER_URL: ${{ secrets.WUKONG_TRIGGER_URL }}
+        run: |
+          if [ -n "$WUKONG_TRIGGER_TOKEN" ]; then
+            curl --fail --silent --show-error \
+              -X POST \
+              -F "token=$WUKONG_TRIGGER_TOKEN" \
+              -F "ref=main" \
+              -F "variables[UPSTREAM_SHA]=$UPSTREAM_SHA" \
+              "$WUKONG_TRIGGER_URL"
+            echo "Downstream CI triggered."
+          else
+            echo "No WUKONG_TRIGGER_TOKEN configured, skipping downstream notification."
+          fi

.github/workflows/publish-npm-release.yml

@@ -1,71 +0,0 @@
-name: Publish npm release
-
-on:
-  workflow_dispatch:
-    inputs:
-      version:
-        description: "Release tag to publish to npm (e.g. v1.0.48)"
-        required: true
-        type: string
-
-permissions:
-  contents: read
-
-jobs:
-  publish-npm:
-    runs-on: ubuntu-latest
-    timeout-minutes: 15
-    steps:
-      - name: Check out repository
-        uses: actions/checkout@v4
-
-      - name: Download GitHub release assets
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          set -eu
-          mkdir -p dist
-          gh release download "${{ inputs.version }}" \
-            --repo "${{ github.repository }}" \
-            --dir dist \
-            --pattern 'dws-*' \
-            --pattern 'checksums.txt' \
-            --clobber
-          ls -la dist
-
-      - name: Stage npm package
-        run: |
-          set -eu
-          version="${{ inputs.version }}"
-          semver="${version#v}"
-          pkg_root="dist/npm/dingtalk-workspace-cli"
-          rm -rf "$pkg_root"
-          mkdir -p "$pkg_root/assets" "$pkg_root/bin"
-          cp build/npm/install.js "$pkg_root/install.js"
-          cp build/npm/bin/dws.js "$pkg_root/bin/dws.js"
-          cp build/npm/README.md "$pkg_root/README.md"
-          sed "s|__VERSION__|${semver}|g" build/npm/package.json.tmpl > "$pkg_root/package.json"
-          cp dist/dws-* "$pkg_root/assets/"
-          cp dist/checksums.txt "$pkg_root/assets/"
-          test -f "$pkg_root/assets/dws-skills.zip"
-          cat "$pkg_root/package.json"
-
-      - name: Setup Node.js
-        uses: actions/setup-node@v4
-        with:
-          node-version: "20"
-          registry-url: "https://registry.npmjs.org"
-
-      - name: Publish stable to npm
-        if: ${{ github.repository_owner == 'DingTalk-Real-AI' && !contains(inputs.version, '-') }}
-        working-directory: dist/npm/dingtalk-workspace-cli
-        run: npm publish --access public
-        env:
-          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
-
-      - name: Publish prerelease to npm beta
-        if: ${{ github.repository_owner == 'DingTalk-Real-AI' && contains(inputs.version, '-') }}
-        working-directory: dist/npm/dingtalk-workspace-cli
-        run: npm publish --access public --tag beta
-        env:
-          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}

.github/workflows/reviewer-router.yml

@@ -0,0 +1,301 @@
+name: Reviewer routing
+
+on:
+  pull_request_target:
+    branches: [main]
+    types: [opened, synchronize, reopened, ready_for_review]
+
+# pull_request_target deliberately runs only this workflow from the protected
+# base branch. Never check out or execute pull-request code here.
+permissions:
+  contents: write
+  pull-requests: write
+
+concurrency:
+  group: reviewer-router-${{ github.event.pull_request.number }}
+  cancel-in-progress: true
+
+jobs:
+  route:
+    if: github.event.pull_request.draft == false
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    steps:
+      - name: Check out trusted routing policy
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          ref: ${{ github.event.pull_request.base.sha }}
+          persist-credentials: false
+      - name: Route review and enable auto-merge
+        uses: actions/github-script@f28e40c7f34bde8b3046d885e986cb6290c5673b # v7.1.0
+        with:
+          script: |
+            const owner = context.repo.owner;
+            const repo = context.repo.repo;
+            const pullNumber = context.payload.pull_request.number;
+            const eventHeadSha = context.payload.pull_request.head.sha;
+            const {
+              REVIEWER_POOL,
+              requestReviewersWithFallback,
+              resolveReviewRouting,
+              reviewerCandidates,
+            } = require('./.github/reviewer-routing.js');
+            const reviewerPool = REVIEWER_POOL;
+
+            async function getReadyEventPull(phase) {
+              const {data: currentPull} = await github.rest.pulls.get({
+                owner,
+                repo,
+                pull_number: pullNumber,
+              });
+              if (
+                currentPull.head.sha !== eventHeadSha ||
+                currentPull.state !== 'open' ||
+                currentPull.draft ||
+                currentPull.base.ref !== 'main'
+              ) {
+                core.info(
+                  `PR #${pullNumber} state or revision no longer matches this ready-main event during ${phase}; routing stopped.`,
+                );
+                return null;
+              }
+              return currentPull;
+            }
+            const pullRequest = await getReadyEventPull('initial read');
+            if (!pullRequest) {
+              return;
+            }
+            const author = pullRequest.user.login.toLowerCase();
+            const headSha = pullRequest.head.sha;
+            const latestPusher =
+              context.payload.action === 'synchronize'
+                ? context.payload.sender?.login?.toLowerCase()
+                : author;
+
+            async function routeReview() {
+              let changedFiles;
+              try {
+                changedFiles = await github.paginate(github.rest.pulls.listFiles, {
+                  owner,
+                  repo,
+                  pull_number: pullNumber,
+                  per_page: 100,
+                });
+              } catch (error) {
+                core.warning(
+                  `Could not inspect changed files for PR #${pullNumber}; using load-balanced fallback (${error.status || 'unknown status'}).`,
+                );
+                changedFiles = [];
+              }
+
+              const eligible = reviewerPool.filter(
+                reviewer =>
+                  reviewer.toLowerCase() !== author &&
+                  reviewer.toLowerCase() !== latestPusher,
+              );
+              if (eligible.length === 0) {
+                core.warning(`No eligible reviewer remains for PR #${pullNumber}.`);
+                return;
+              }
+
+              const existingRequestedReviewers = new Set(
+                (pullRequest.requested_reviewers || []).map(({login}) => login.toLowerCase()),
+              );
+
+              let reviews;
+              try {
+                reviews = await github.paginate(github.rest.pulls.listReviews, {
+                  owner,
+                  repo,
+                  pull_number: pullNumber,
+                  per_page: 100,
+                });
+              } catch (error) {
+                core.warning(
+                  `Could not inspect existing reviews for PR #${pullNumber}; skipping reviewer routing to avoid a duplicate request (${error.status || 'unknown status'}).`,
+                );
+                return;
+              }
+
+              const latestDecisionByLogin = new Map();
+              for (const review of reviews) {
+                const login = review.user?.login?.toLowerCase();
+                if (
+                  !login ||
+                  !['APPROVED', 'CHANGES_REQUESTED', 'DISMISSED'].includes(
+                    review.state,
+                  )
+                ) {
+                  continue;
+                }
+                const previous = latestDecisionByLogin.get(login);
+                if (!previous || review.id > previous.id) {
+                  latestDecisionByLogin.set(login, review);
+                }
+              }
+              const currentHeadReviewers = new Set(
+                [...latestDecisionByLogin.values()]
+                  .filter(
+                    review =>
+                      review.commit_id === headSha &&
+                      eligible.some(
+                        reviewer => reviewer.toLowerCase() === review.user.login.toLowerCase(),
+                      ) &&
+                      ['APPROVED', 'CHANGES_REQUESTED'].includes(review.state),
+                  )
+                  .map(review => review.user.login.toLowerCase()),
+              );
+
+              const loads = new Map(eligible.map(reviewer => [reviewer, 0]));
+              try {
+                const openPullRequests = await github.paginate(github.rest.pulls.list, {
+                  owner,
+                  repo,
+                  state: 'open',
+                  per_page: 100,
+                });
+                for (const openPullRequest of openPullRequests) {
+                  for (const reviewer of openPullRequest.requested_reviewers || []) {
+                    const candidate = eligible.find(
+                      login => login.toLowerCase() === reviewer.login.toLowerCase(),
+                    );
+                    if (candidate) {
+                      loads.set(candidate, loads.get(candidate) + 1);
+                    }
+                  }
+                }
+              } catch (error) {
+                core.warning(
+                  `Could not read current reviewer load; using deterministic rotation (${error.status || 'unknown status'}).`,
+                );
+              }
+
+              const offset = pullNumber % eligible.length;
+              const rotated = eligible.slice(offset).concat(eligible.slice(0, offset));
+              const tieOrder = new Map(rotated.map((reviewer, index) => [reviewer, index]));
+              const staleChangeRequester = [...latestDecisionByLogin.values()]
+                .filter(review => review.state === 'CHANGES_REQUESTED')
+                .sort((left, right) => right.id - left.id)
+                .map(review =>
+                  eligible.find(
+                    reviewer =>
+                      reviewer.toLowerCase() === review.user.login.toLowerCase(),
+                  ),
+                )
+                .find(Boolean);
+              const ranked = [...eligible].sort(
+                (left, right) =>
+                  Number(right === staleChangeRequester) -
+                    Number(left === staleChangeRequester) ||
+                  loads.get(left) - loads.get(right) ||
+                  tieOrder.get(left) - tieOrder.get(right),
+              );
+
+              const routing = resolveReviewRouting({
+                files: changedFiles,
+                author,
+                latestPusher,
+                fallbackReviewers: ranked,
+              });
+              const candidates = reviewerCandidates({
+                preferredReviewers: routing.reviewers,
+                fallbackReviewers: ranked,
+                eligibleReviewers: eligible,
+              });
+              const desiredReviewers = candidates.slice(0, routing.requiredReviewers);
+              if (routing.reason === 'unknown_paths' && currentHeadReviewers.size > 0) {
+                core.info(
+                  `PR #${pullNumber} has a current-head review for unknown paths; leaving manual ownership unchanged.`,
+                );
+                return;
+              }
+              core.info(
+                `PR #${pullNumber} routing: ${routing.reason}; modules=${routing.modules.map(module => module.id).join(',') || 'unknown'}; reviewers=${desiredReviewers.join(',') || 'load-balanced fallback'}.`,
+              );
+
+              const satisfiedReviewers = new Set([
+                ...currentHeadReviewers,
+                ...[...existingRequestedReviewers].filter((reviewer) =>
+                  candidates.some((candidate) => candidate.toLowerCase() === reviewer),
+                ),
+              ]);
+              const requestResult = await requestReviewersWithFallback({
+                candidates,
+                requiredReviewers: routing.requiredReviewers,
+                satisfiedReviewers: [...satisfiedReviewers],
+                requestReviewer: async (reviewer) => {
+                  const currentPull = await getReadyEventPull('review request');
+                  if (!currentPull) {
+                    return false;
+                  }
+                  await github.rest.pulls.requestReviewers({
+                    owner,
+                    repo,
+                    pull_number: pullNumber,
+                    reviewers: [reviewer],
+                  });
+                  core.info(
+                    `Requested @${reviewer} for PR #${pullNumber} (open request load: ${loads.get(reviewer)}).`,
+                  );
+                  return true;
+                },
+                onFailure: (reviewer, error) => {
+                  core.warning(
+                    `Could not request @${reviewer} for PR #${pullNumber}; trying the next candidate (${error.status || 'unknown status'}).`,
+                  );
+                },
+              });
+              if (requestResult.aborted) {
+                return;
+              }
+
+              if (requestResult.satisfiedReviewers.length < routing.requiredReviewers) {
+                core.warning(
+                  `Only ${requestResult.satisfiedReviewers.length} of ${routing.requiredReviewers} required reviewers could be satisfied for PR #${pullNumber}.`,
+                );
+              }
+            }
+
+            async function enableAutoMerge() {
+              try {
+                const currentPull = await getReadyEventPull('auto-merge enable');
+                if (!currentPull) {
+                  return;
+                }
+                if (currentPull.auto_merge) {
+                  core.info(`Auto-merge is already enabled for PR #${pullNumber}.`);
+                  return;
+                }
+                await github.graphql(
+                  `mutation EnableAutoMerge($pullRequestId: ID!) {
+                    enablePullRequestAutoMerge(
+                      input: {
+                        pullRequestId: $pullRequestId
+                        mergeMethod: MERGE
+                      }
+                    ) {
+                      pullRequest {
+                        autoMergeRequest {
+                          enabledAt
+                        }
+                      }
+                    }
+                  }`,
+                  {pullRequestId: currentPull.node_id},
+                );
+                core.info(`Enabled native auto-merge for PR #${pullNumber}.`);
+              } catch (error) {
+                core.warning(
+                  `Could not enable auto-merge for PR #${pullNumber}; checks and review can continue normally (${error.message}).`,
+                );
+              }
+            }
+
+            try {
+              await routeReview();
+            } catch (error) {
+              core.warning(
+                `Reviewer routing hit an unexpected error for PR #${pullNumber}; review can still proceed manually (${error.message}).`,
+              );
+            }
+            await enableAutoMerge();

.github/workflows/sync-release-to-gitee.yml

@@ -1,50 +0,0 @@
-name: Sync release to Gitee
-
-# Manually mirror a published GitHub release's assets to the matching Gitee
-# release. Use this to repair a release whose Gitee mirror is incomplete (e.g.
-# the Release job timed out mid-upload). It runs ONLY the idempotent Gitee sync
-# step — it does not run GoReleaser and does not touch the GitHub release, so
-# there is no release outage. The sync script skips assets already on Gitee, so
-# this only uploads what is missing.
-
-on:
-  workflow_dispatch:
-    inputs:
-      version:
-        description: "Release tag to mirror to Gitee (e.g. v1.0.42)"
-        required: true
-        type: string
-
-permissions:
-  contents: read
-
-jobs:
-  sync-gitee:
-    runs-on: ubuntu-latest
-    timeout-minutes: 60
-    steps:
-      - name: Check out repository
-        uses: actions/checkout@v4
-
-      - name: Download GitHub release assets
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          set -eu
-          mkdir -p dist
-          gh release download "${{ inputs.version }}" \
-            --repo "${{ github.repository }}" \
-            --dir dist \
-            --pattern 'dws-*' \
-            --pattern 'checksums.txt' \
-            --clobber
-          ls -la dist
-
-      - name: Mirror release to Gitee (China)
-        # Idempotent: uploads only assets not already present on the Gitee release.
-        run: ./scripts/release/sync-to-gitee.sh
-        env:
-          VERSION: ${{ inputs.version }}
-          GITEE_TOKEN: ${{ secrets.GITEE_TOKEN }}
-          GITEE_USER: ${{ secrets.GITEE_USER }}
-          GITEE_REPO: ${{ secrets.GITEE_REPO }}

.github/workflows/withdraw-release.yml

@@ -0,0 +1,148 @@
+name: Withdraw release
+run-name: Withdraw ${{ inputs.version }}
+
+on:
+  workflow_dispatch:
+    inputs:
+      version:
+        description: "Exact published version to withdraw (vX.Y.Z or vX.Y.Z-beta.N)"
+        required: true
+        type: string
+      reason:
+        description: "Public, single-line withdrawal reason (8-300 characters)"
+        required: true
+        type: string
+      confirmation:
+        description: "Type WITHDRAW followed by a space and the exact version"
+        required: true
+        type: string
+
+permissions:
+  contents: read
+
+# Share the publication lock with release.yml. A withdrawal and a publication
+# must never mutate channel pointers concurrently.
+concurrency:
+  group: dws-release-publication
+  cancel-in-progress: false
+
+jobs:
+  withdraw:
+    name: Withdraw release from every distribution channel
+    environment: release-withdrawal
+    runs-on: ubuntu-latest
+    timeout-minutes: 180
+    permissions:
+      actions: read
+      contents: write
+    steps:
+      - name: Verify withdrawal environment protection
+        uses: actions/github-script@v7
+        with:
+          script: |
+            const { owner, repo } = context.repo;
+            const response = await github.request(
+              "GET /repos/{owner}/{repo}/environments/{environment_name}",
+              { owner, repo, environment_name: "release-withdrawal" },
+            );
+            const reviewerRule = response.data.protection_rules.find(
+              (rule) => rule.type === "required_reviewers",
+            );
+            if (
+              !reviewerRule ||
+              reviewerRule.prevent_self_review !== true ||
+              !Array.isArray(reviewerRule.reviewers) ||
+              reviewerRule.reviewers.length === 0
+            ) {
+              core.setFailed("release-withdrawal must require a reviewer and prevent self-review");
+              return;
+            }
+            if (response.data.deployment_branch_policy?.protected_branches !== true) {
+              core.setFailed("release-withdrawal must allow only protected branches");
+            }
+            if (response.data.can_admins_bypass !== false) {
+              core.setFailed("release-withdrawal must not allow administrator bypass");
+            }
+
+      - name: Require the exact current official default-branch commit
+        uses: actions/github-script@v7
+        with:
+          script: |
+            const expectedRepository = "DingTalk-Real-AI/dingtalk-workspace-cli";
+            const defaultBranch = context.payload.repository.default_branch;
+            if (context.eventName !== "workflow_dispatch") {
+              core.setFailed("release withdrawal accepts workflow_dispatch only");
+              return;
+            }
+            if (`${context.repo.owner}/${context.repo.repo}` !== expectedRepository) {
+              core.setFailed(`release withdrawal is restricted to ${expectedRepository}`);
+              return;
+            }
+            if (context.ref !== `refs/heads/${defaultBranch}`) {
+              core.setFailed(`release withdrawal must be dispatched from ${defaultBranch}`);
+              return;
+            }
+            const branch = await github.rest.git.getRef({
+              ...context.repo,
+              ref: `heads/${defaultBranch}`,
+            });
+            if (branch.data.object.sha !== context.sha) {
+              core.setFailed(
+                `default branch advanced to ${branch.data.object.sha}; re-dispatch from the new head`,
+              );
+            }
+
+      - name: Check out trusted withdrawal tooling
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ github.sha }}
+          fetch-depth: 0
+          persist-credentials: false
+
+      - name: Set up Node.js for npm channel withdrawal
+        uses: actions/setup-node@v4
+        with:
+          node-version: "22"
+          registry-url: "https://registry.npmjs.org"
+
+      - name: Withdraw immutable release and roll back channels
+        id: withdrawal
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
+          GITHUB_EVENT_DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
+          WITHDRAW_VERSION: ${{ inputs.version }}
+          WITHDRAW_REASON: ${{ inputs.reason }}
+          WITHDRAW_CONFIRMATION: ${{ inputs.confirmation }}
+          OSS_ACCESS_KEY_ID: ${{ secrets.OSS_ACCESS_KEY_ID }}
+          OSS_ACCESS_KEY_SECRET: ${{ secrets.OSS_ACCESS_KEY_SECRET }}
+          OSS_ENDPOINT: ${{ secrets.OSS_ENDPOINT }}
+          OSS_BUCKET: ${{ secrets.OSS_BUCKET }}
+          OSS_PREFIX: ${{ secrets.OSS_PREFIX }}
+          GITEE_TOKEN: ${{ secrets.GITEE_TOKEN }}
+          GITEE_USER: ${{ secrets.GITEE_USER }}
+          GITEE_REPO: ${{ secrets.GITEE_REPO }}
+          DWS_GITEE_ENABLED: ${{ vars.ENABLE_GITEE_UPLOAD_FALLBACK == 'true' && 'true' || 'false' }}
+          HOMEBREW_PR_TOKEN: ${{ secrets.HOMEBREW_PR_TOKEN }}
+        run: |
+          ./scripts/release/withdraw-release.sh \
+            "$WITHDRAW_VERSION" \
+            "$WITHDRAW_REASON" \
+            "$WITHDRAW_CONFIRMATION"
+
+      - name: Report withdrawal boundary
+        if: ${{ always() }}
+        env:
+          VERSION: ${{ inputs.version }}
+          RESULT: ${{ steps.withdrawal.outcome }}
+        run: |
+          {
+            echo "### Release withdrawal: ${VERSION}"
+            echo
+            echo "- Workflow result: ${RESULT}"
+            echo "- Success means every configured channel was verified and the permanent withdrawn/${VERSION} tombstone remains as the version-reuse barrier."
+            echo "- Failure may occur before or after the tombstone/channel mutations; inspect the failed step and rerun the exact same inputs after fixing the cause."
+            echo "- The problem GitHub Release and original tag are removed after npm and every tag-enabled/configured mirror are rolled back, so GitHub installers stop resolving the bad version while the Homebrew rollback PR is reviewed."
+            echo "- npm is deprecated rather than unpublished; already-installed clients cannot be remotely downgraded."
+            echo "- If a Homebrew rollback PR was opened, this run remains failed until that PR is independently reviewed, merged, and the workflow is rerun."
+          } >> "$GITHUB_STEP_SUMMARY"

.gitignore

@@ -19,6 +19,7 @@ test/cli_compat/testdata/
 /internal/compat/testdata/*
 .gitignore
 .worktrees/
+.qoder/
 
 # Secrets & credentials
 .env
@@ -42,5 +43,29 @@ dws.zip
 # 功能测试运行产物
 results.jsonl
 test/dev_functional/results.jsonl
+/auto-test/
+/eval-runs/
+/.qoder/
 .vercel
 .env*
+# Local Go coverage output
+/coverage.txt
+/coverage-base.txt
+/coverage-policy.txt
+/coverage.html
+dwsbin
+
+# Local shortcut eval / real-backend capture artifacts — may contain real PII
+# (employee names/emails, userIds, conversation & message IDs). Never commit.
+/docs/shortcut-real-read-results.json
+/docs/shortcut-real-write-results.json
+/docs/shortcut-comparison.html
+/docs/shortcut-gsb-eval.*
+/scripts/run_shortcut_real_read_matrix.py
+
+# Local coverage artifacts
+coverage-shortcut.txt
+coverage-*.txt
+
+# stray compiled generator binary (source lives in internal/generator/cmd_param_aliases/)
+/cmd_param_aliases

.goreleaser.yaml

@@ -1,19 +1,14 @@
 # GoReleaser configuration for dws
 # Docs: https://goreleaser.com
 #
-# To release:
-#   git tag -a v0.1.0 -m "Release v0.1.0"
-#   git push origin v0.1.0
+# To release, use scripts/release/release.sh. It seals main, validates the
+# CHANGELOG and packages, then pushes the annotated tag for CI/CD to publish.
 #
 # To test locally (no publish):
 #   goreleaser release --snapshot --clean
 
 version: 2
 
-before:
-  hooks:
-    - go mod tidy
-
 builds:
   - main: ./cmd
     binary: dws
@@ -67,7 +62,9 @@ release:
     # 用当前运行 CI 的仓库 owner: fork CI 发到 fork, 官方 CI 发到官方, 两边都对
     owner: "{{ .Env.GITHUB_REPOSITORY_OWNER }}"
     name: dingtalk-workspace-cli
-  draft: false
+  # Keep the release private until post-processing has replaced the Darwin
+  # archives and verified every finalized asset digest.
+  draft: true
   prerelease: auto
   name_template: "v{{.Version}}"
   mode: replace

AGENTS.md

@@ -0,0 +1,615 @@
+# Repository Agent Guide
+
+This file applies to the entire repository. Keep changes scoped, preserve
+unrelated work, and use `gofmt` for every modified Go file.
+
+## Build and test
+
+- Build: `make build` (wraps `scripts/dev/build.sh` → `go build -o dws ./cmd`; bare `go build ./cmd` fails because output name `cmd` collides with the directory)
+- Full test suite: `DWS_PACKAGE_VERSION=0.0.0-test go test ./...`
+- Param aliases generate: `go generate ./internal/cli` (entry point: `internal/cli/gen.go`; Catalog is not generated)
+- Optional diagnostic MCP dump (not a Schema pin): `make fetch-mcp-metadata` (requires `dws auth login`; writes under `artifacts/`)
+- Check generated drift + assembly determinism: `./scripts/policy/check-generated-drift.sh`
+- Check the Schema contract: `./scripts/policy/check-schema-catalog.sh`
+- Coverage-gate test naming: tests that carry coverage for the macOS platform
+  gate must be named `TestCrossPlatformCoverage*` (or `TestAllShortcuts*`);
+  `scripts/policy/run-platform-coverage-gate.sh` only selects those prefixes,
+  so a covering test with any other name silently leaves its target uncovered.
+- Package-var injection seams (e.g. `pipelineBuildEffectiveRegistry`): swap
+  them in tests only via `testseam.Swap(t, &seam, stub)` from
+  `internal/testseam` — it restores the previous value through `t.Cleanup`
+  structurally. Like the manual pattern it replaces, Swap mutates global state
+  and is **not** safe for `t.Parallel` tests.
+- Cross-package test helpers (e.g. `StoreProductDeclRawForTest`) live in
+  per-package `fortest.go` files, never scattered through production files;
+  the `ForTest` suffix is the boundary and production code must not call them.
+
+Schema Catalog delivery is **声明即 Catalog**: production assembles via
+`RegisterSchemaSourceRoot` → `ResolveSchemaBuild` (factory registered in
+`internal/app`). There is no
+`cmd_schema_catalog` `//go:generate` delivery step. `dws schema -f json` remains
+the wire projection. `cmd_schema_catalog` produces CI/local dumps only;
+`internal/cli/schema_catalog/`, `internal/cli/schema_meta_index.gob`, and
+`internal/cli/schema_meta_index.json` must not be committed. `schema_agent_metadata/` is retired: if that directory
+(or `schema_agent_metadata_audit.json`) is present, policy fails.
+Command identity is no longer a file input: it is collected from
+`ContractFinal.Identity` on the live Cobra leaves
+(`internal/cli/schema_identity_collect.go` → `BuildEffectiveCommandRegistry`).
+The reviewed `schema_command_registry/` was retired together with that
+switchover and must not reappear; identity changes happen by editing the leaf
+declaration. The remaining **reviewed inputs** under `internal/cli` (see Agent
+Schema contract) keep separate authorities — do not merge them with
+`param_concepts.json` or promote any of them into Catalog declaration.
+
+## Command framework declaration
+
+- Framework definition: `docs/rfc-command-framework-convergence.md` **§5.0**
+- Today: `helpers.LeafSpec` / `shortcut.Shortcut` → `corecmd.Spec` (+ optional `Contract`) → `corecmd.New`
+- **Declare = final Schema source**: `Flags` / `Constraints` / `Safety` / `ConstParams` / `Contract` (`corecmd.ContractDecl`; nested fields are `contract.*`)
+- Naming: `ContractDecl` is the authoring leaf declaration. "Schema" means Catalog / `ToolSpec` delivery — do not reintroduce `SchemaDecl`.
+- `Safety` uses `contract.SafetySpec` (`internal/corecmd/contract` only — no `cli.*` type alias). Its `confirmation` drives the runtime gate; `effect` / `risk` / `idempotency` are published unchanged. When `Contract` is set, convert once via `contractfinal.RegisterRuntimeContractFinal` (all callers — `corecmd.New` registers internally); assembly **pass-throughs** Final.
+- Package seam:
+  - types / ProductDecl → `corecmd/contract` (DTO only; **no** Cobra-keyed ContractFinal store)
+  - AnnotateRuntime* writers → `internal/corecmd/runtimeannotate` (framework-owned)
+  - ContractFinal cobra store + Register → `internal/corecmd/contractfinal` (framework-owned)
+  - homology gates → `internal/cli/homology`
+  - Catalog assembly / `ResolveMeta` (`RegisterSchemaSourceRoot` → `ResolveSchemaBuild`); go:embed only for reviewed inputs → `internal/cli` root (package-local aliases for annotate/store APIs live in `runtime_schema_seam.go`; the former `cli/runtimeannotate` / `cli/contractfinal` shim packages are removed — import `corecmd/*` directly)
+  - **Hard rule**: `internal/corecmd` (and its subpackages) must **not** import any `internal/cli` package
+- Authoring tiers (current, not aspirational):
+  - **Tier1** — `corecmd.New` / `helpers.NewLeafCommand` (fully managed declare + execute)
+  - **Tier2** — `DeclareLeafMetadata` (helpers migration; **Shortcut may also use this path — acceptable**)
+  - **Tier3** — bare Cobra (should shrink over time; reviewed exclusions where needed)
+  - Long-term outlook only: broader mcpbind / fewer hand-written `Execute` bodies. **Not** a current hard requirement to delete `Shortcut.Execute` or force mcpbind.
+- Description declare vs delivery: construction requires `ContractDecl.Description` (evidence). Catalog delivery prefers Cobra Long → provenance `cobra_help`; without Long, declared text → `contract_final`. Title: declared first, then Short, then MCP. Do **not** read this as "declare = wire final" or dual authority.
+- **Execute** = hooks (`Validate` / `Call` / `RunE` / `PostMount`) — not a second surface authority
+- Declaration path has **no reviewed parallel fields**; migration-only `runtime_gate` annotate until `Safety` is declared
+- **Do not add** new production `AnnotateRuntimeRisk` / `AnnotateRuntimeGate`
+  (`runtime_gate`) call sites; migrate leaves to declared `Safety` /
+  `ContractDecl` instead. Existing annotate sites may remain until migrated.
+
+## flag / help / schema homology
+
+- Decision (path A — Contract/LeafSpec is CLI-surface authority **and must embed into Schema**): `docs/flag-help-schema-homology.md`
+- Hard rule: every help/Schema fact is **declared** **or** **annotated**; never inference-only (§1.1–§1.3; framework §5.0).
+- Embed path: `corecmd.New` → `dws.schema.*` annotations → Schema catalog assembly
+- MCP metadata must not create CLI flags; optional 1:1 passthrough is a gated subset only.
+- Gate IDs: `HOM-P*`, `HOM-S*`, `HOM-I1`, `HOM-D1` (see that doc §3–§4). `HOM-P1`/`HOM-D1`/`HOM-S1`/`HOM-S2` are on the `check-schema-catalog.sh` policy whitelist; remaining IDs land incrementally.
+
+## Agent Schema contract
+
+The Schema data flow is one way:
+
+```text
+1. app.NewRootCommand()
+   └─ builds the real Cobra command tree and flags
+   └─ leaf Safety / Contract / contract.ParamDecl declare ContractFinal (declare-or-annotate)
+
+2. CollectIdentitySpecs (ContractFinal.Identity on live Cobra leaves)
+   └─ forms EffectiveCommandRegistry
+      └─ binds exactly to real Cobra leaves and aliases
+
+3. Parameter resolution
+   Cobra flags
+   + contract.ParamDecl.Property / native annotations (primary property authority)
+   + schema_parameter_mapping_ledger.go (mapping_exclusions / removals only;
+     active bindings JSON retired after Track 1 Phase 2)
+   └─ produces ParameterSpec and constraints
+
+4. Agent and interface semantics
+   ProductDecl + leaf ContractFinal Selection / Safety / Interface
+   + contract.ParamDecl (interface_type / property)
+   └─ resolves Agent metadata by source precedence
+      Markdown is evidence only; it is not concatenated into final prose
+   └─ schema_hints/ and schema_mcp_metadata.json are fully retired
+
+  5. One typed hub
+   BoundCommandRegistry
+   + ParameterSpec
+   + Agent metadata
+   + Interface metadata
+   └─ resolves every command exactly once into ToolSpec
+   └─ aggregates SchemaRegistry + SchemaIndex
+   └─ ResolveSchemaBuild assembles at runtime; deliverySchemaCatalog wraps it (lazy, sync.Once)
+
+  6. Runtime delivery (no generate-written Catalog authority)
+   SchemaRegistry
+   └─ dws schema list/product/group/leaf/--all (-f json wire)
+   └─ ResolveMeta projects Identity/Safety/Selection from the same registry
+   └─ CI may dump Catalog via cmd_schema_catalog for jq gates / determinism
+```
+
+**Reviewed inputs / 评审输入** (organizational family under `internal/cli`;
+parallel peers, not one merged authority). These are assembly inputs only —
+never Catalog declaration authority, never leaf `Contract` / `ProductDecl`
+substitutes. Keep them side-by-side; do **not** fold one into another:
+
+| Input | Path | Owns |
+|---|---|---|
+| Command identity | collected from `ContractFinal.Identity` on live Cobra leaves (`schema_identity_collect.go`; not a file input) | stable identity, primary CLI path, aliases, navigation |
+| Param concepts | `param_concepts.json` (+ `.schema.json`) | argv synonym / concept dictionary (reduced to `param_aliases_generated.go`) |
+| Exclusions | `schema_command_exclusions.go` | exact reviewed CLI paths excluded from Schema (non-empty reason) |
+| Mapping ledger | `schema_parameter_mapping_ledger.go` | `mapping_exclusions` / removals (CLI flags with no direct RPC property); active bindings JSON retired |
+
+`schema_mcp_metadata.json` is retired and must not reappear. Interface facts
+(`interface_ref`, `interface_type`, …) declare on leaf `Contract` /
+`contract.ParamDecl`. Retiring the pin cleared MCP-sourced `interface_type`
+values from the wire; schema-compat deliberately accepts clearing (missing =
+unknown for consumers) while still rejecting any change to a different
+non-empty value. Re-populating a value requires an explicit `ParamDecl`
+declaration, not a new pin.
+
+**Aliases are three distinct layers** (do not conflate):
+
+| Layer | Owns |
+|---|---|
+| `FlagSpec.Aliases` / Cobra flag aliases | executable flag synonyms on a leaf |
+| `ContractFinal.Identity` `aliases` | reviewed CLI-path aliases for the same command identity |
+| `param_concepts.json` | argv synonym / concept dictionary (central preparse normalization) |
+
+**Visibility vs exclusions:** collected identity `visibility` is dormant (all
+entries default `public`); “runnable but not Agent-visible” belongs in
+`schema_command_exclusions.go`, not new `visibility` values. Native identity
+annotations are consistency assertions only — they must agree with the
+collected identity and never materialize or override it.
+
+Leaf declare (`Contract` / `ParamDecl` / `Safety` / `ProductDecl`) and the live
+Cobra tree remain separate from this table: declare owns semantics; Cobra owns
+executability and flags.
+
+After binding there is no second identity source and no identity precedence
+winner. The binder must reject a missing/non-runnable Cobra path, an alias
+collision, and any native identity annotation that disagrees with the effective
+registry. A missing native identity annotation is allowed because annotations
+are implementation-side assertions, not identity fallbacks.
+
+The assembler resolves every bound command exactly once into one `ToolSpec`.
+CI determinism (`check-schema-assembly.sh`) and policy jq gates consume a
+fresh assembly dump; runtime consumes the same `ResolveSchemaBuild` path via
+`RegisterSchemaSourceRoot`. Neither path may reopen annotations, merge source
+records, or use a previous Catalog JSON as a source.
+
+### Assembly vs consumption
+
+**Assembly** (declare → typed registry; CI + runtime):
+- Runtime entry: `RegisterSchemaSourceRoot` (`internal/app`) →
+  `ResolveSchemaBuild` / `deliverySchemaCatalog` (lazy, sync.Once).
+- CI tool: `cmd_schema_catalog` dumps an assembled Catalog for jq/determinism;
+  it is **not** a `//go:generate` or committed delivery step.
+- `gen.go` only generates `param_aliases_generated.go`.
+- Inputs: **reviewed inputs** (param_concepts / exclusions / mapping ledger —
+  see table above) + ProductDecl/ContractFinal (identity is collected from
+  `ContractFinal.Identity`) + live Cobra tree.
+  `schema_hints/`, `schema_agent_metadata/`, `schema_command_registry/`, and
+  `schema_mcp_metadata.json` must not reappear.
+- Gates: `make generate-schema` (param aliases + assembly determinism),
+  `check-generated-drift.sh`, `check-schema-catalog.sh`.
+
+**Consumption** (runtime, unified API):
+- Entry point: `ResolveMeta(cliPath) → CommandMeta{Identity, Safety, Selection}`
+  in `internal/cli/command_meta.go` — projected from the assembled registry
+  when the app factory is registered.
+- Consumers: `--help` (Safety annotation via `RenderSafetyAnnotation`),
+  agent selection, future skill generation; `dws schema` uses the same
+  assembled Catalog (`-f json` wire unchanged).
+- `SafetyForCLIPath` delegates to `ResolveMeta` (backward compatible).
+
+This split is architecturally isomorphic to Lark's typed metadata registry,
+navigation catalog, and schema renderer. DWS intentionally preserves its
+existing flat JSON wire contract for compatibility; do not treat architectural
+alignment as permission to make an unversioned wire-format change.
+
+The identity collected from `ContractFinal.Identity` (via
+`CollectIdentitySpecs`) is the sole source of stable command identity and
+navigation. The executable Cobra tree remains the source of truth for whether
+a CLI path exists, is runnable, and which flags it accepts. Schema coverage is
+bidirectional:
+
+1. Every final `SchemaRegistry` tool, including its serialized Catalog
+   projection, must resolve to an executable Cobra command.
+2. Every public runnable Cobra leaf must either resolve to Schema or appear as
+   an exact, reviewed exclusion with a non-empty reason in
+   `internal/cli/schema_command_exclusions.go` (central Go groups; not JSON).
+
+Do not use prefix or wildcard exclusions: they can silently hide future
+commands. Remove an exclusion when its command enters Schema; stale, invalid,
+or duplicate exclusions must fail generation and CI.
+
+When adding or changing an Agent-visible command, review all relevant inputs:
+
+- Leaf `ContractFinal.Identity` for canonical identity, primary CLI path,
+  aliases, and stable navigation. Identity is collected from the live Cobra
+  leaves (`CollectIdentitySpecs`); there is no separate identity file. Invalid
+  canonical paths, alias collisions, stale paths, and drift fail collection,
+  binding, and policy.
+- Leaf `Safety` / `Contract` (`corecmd.ContractDecl`) / `contract.ParamDecl`
+  (helpers `LeafSpec` or shortcut `Contract`) for parameter facts, interface
+  disposition, safety, and Agent selection prose. Delivered provenance is
+  `contract_final` from `corecmd.contract` (description may stamp `cobra_help`
+  when Cobra Long wins). Product routing uses `ProductDecl`
+  (`internal/corecmd/contract`; provenance label remains `cli.product_decl`).
+- `internal/cli/schema_hints/` is fully retired. Do not reintroduce HintFiles,
+  audit JSON, or `imported/` baselines; declare on ProductDecl / the owning
+  leaf instead.
+- Native Runtime Schema identity annotations, when present, as consistency
+  assertions against `EffectiveCommandRegistry`. They must agree exactly and
+  must never materialize, infer, or override registry identity.
+- Flag-to-interface property mappings and required/default semantics.
+- Do not expect generate-written Catalog delivery. Run
+  `make generate-schema` only to refresh param aliases and prove assembly
+  determinism. Do not expect or commit `schema_agent_metadata/`.
+
+Run the reverse-completeness tests whenever the Cobra tree changes. A command
+that works through `dws <path>` but cannot be found through the matching
+`dws schema` lookup is a contract failure unless it has a reviewed exact
+exclusion.
+
+`RegisterSchemaHints` / `ToolSchemaHint` overlays are fully removed. Parameter
+and selection facts must be declared on the owning leaf (`contract.ParamDecl` /
+`Contract`) or via `ProductDecl`; do not reintroduce overlay registries.
+
+For Agent-authored selection edits:
+
+1. Confirm the exact command and flag names in the current Cobra tree.
+2. Declare selection prose on the owning leaf (`Contract.Selection` /
+   `DeclareLeafMetadata`) and product routing via `ProductDecl`; declare
+   safety / parameters / interface on the same leaf.
+3. Do not copy generated Catalog fields into source inputs.
+4. Run generation, drift, Schema policy, and the focused CLI tests before
+   proposing the change.
+
+## Agent curation workflow
+
+Use this workflow when refreshing Agent selection prose and confirmation
+alignment. Prefer **agent-authored review** over bulk merge scripts that dump
+Skill Markdown into Catalog fields.
+
+Human-authored inputs:
+
+| Block | Path | Owns |
+|---|---|---|
+| **declaration** | helpers / shortcut `Safety` + `Contract` / `contract.ParamDecl` + `ProductDecl` | `effect` / `risk` / `confirmation` / `idempotency` / `interface_*` / parameter facts / selection prose (`contract_final`) |
+
+`schema_hints/` is fully retired. Do not reintroduce HintFiles or audit JSON.
+
+### Goals
+
+1. **Selection prose** is decision-oriented (Feishu/Lark style): trigger intent,
+   sibling-command routing, and outcome shape — not a restatement of the
+   summary. Delivered Catalog provenance is `contract_final` from leaf
+   `Contract.Selection` / `ProductDecl`.
+2. **Safety** follows Runtime: `confirmation=user_required` when the leaf
+   Contract/Safety (or remaining `runtime_gate` annotate) requires a user gate
+   (for example `confirm_delete`, `typed_yes`, `confirm_dangerous`).
+3. **Parameter facts** are declared on the leaf (`contract.ParamDecl` /
+   `Contract.Parameters` / FlagSpec). Do not reintroduce HintFile or
+   `RegisterSchemaHints` overlays.
+
+### Authoring
+
+For every curated tool:
+
+1. Declare safety/interface/parameters/selection on the owning leaf
+   (`DeclareLeafMetadata` / `Shortcut.Contract` / `contract.ParamDecl`) and product routing
+   via `ProductDecl` when needed.
+2. Run `make generate-schema` (param aliases + assembly determinism). Do not
+   create or commit `schema_catalog/` or Schema meta-index fixtures.
+
+### Pull live MCP descriptions (personal token)
+
+Schema delivery no longer embeds a pinned MCP JSON. Prefer live Schema from a
+logged-in personal session when reviewing interface facts before declaring them
+on the leaf:
+
+```bash
+dws auth status                 # token_valid should be true
+dws schema <mcp-canonical> --jq '{canonical_path,interface_ref,parameters}' -f json
+# or CLI path: dws schema --cli-path "drive copy" --jq '{canonical_path,interface_ref,parameters}' -f json
+```
+
+Resolve MCP identity via declared `interface_ref` when CLI canonical ≠ MCP path
+(example: CLI `drive.copy_document` → live `doc.copy_document`). On pull
+failure, fall back to Skill + Cobra Help, and record evidence
+(for example `live-dws-schema:<path>#FAILED`). Never print or commit tokens.
+`make fetch-mcp-metadata` writes an optional diagnostic dump under `artifacts/`
+only — do not commit it as a Schema pin.
+
+Precedence when sources disagree: **Runtime/Cobra / leaf Contract > live MCP >
+Skill (evidence only)**.
+
+### Parallel product agents
+
+Split work by product groups. Each agent must:
+
+- Read Skill, Cobra/`--help`, Runtime confirmation sites, and live
+  `dws schema <leaf> --compact` for its tools. Mapping/interface/provenance
+  audits may query the full leaf only through a narrow `--jq` / `--fields`
+  projection; do not load an entire full leaf into Agent context.
+- Hand-write selection prose and leaf Contract / ProductDecl declarations;
+  forbid wholesale JSON merges from review dumps.
+- Edit only its product’s leaf declarations (and `ProductDecl` when needed).
+- **Never** `git checkout` unrelated product files to “clean scope”.
+
+### Regenerate and gates
+
+```bash
+make generate-schema
+./scripts/policy/check-runtime-confirmation-truth.sh
+go test ./internal/app -run '^TestSheetFinalSchemaConfirmationMatchesRuntimeGuards$' -count=1
+```
+
+`check-runtime-confirmation-truth.sh` compares live ContractFinal.Safety with the assembled ToolSpec `confirmation=user_required` and probes the runtime gate.
+`schema_hints/` must stay absent.
+
+Example rules (fail generation otherwise):
+
+- At most two examples per tool; no `--yes` in stored examples.
+- Examples must match live Cobra argv (path, flags, required groups).
+- No shell comments in examples.
+
+After generation, spot-check Catalog: selection and safety/interface
+provenance are `contract_final` from ProductDecl / leaf declarations
+(`user_required` must match Runtime confirmation gates).
+
+`make generate-schema` refreshes `param_aliases_generated.go` and runs
+assembly determinism (`check-schema-assembly.sh`). It does not rewrite a
+committed Catalog as delivery authority — runtime reassembles from
+declarations. Byte guards fail if generation mutates parameter-concept
+inputs; policy fails if the retired `schema_command_registry/` reappears.
+
+Selection prose may choose a more or less restrictive recommendation. It cannot
+create a Cobra command or flag, change parameter facts, invent an
+RPC/interface, alter safety metadata, or bypass command completeness. Examples
+must use an executable primary/alias path and flags accepted by the live Cobra
+command; never add `--yes` to stored examples.
+
+Every example is always checked against its real `BoundCommand`: exact path,
+accepted flags, Cobra required flags/positionals, and the effective
+`require_one_of`, `require_together`, and `mutually_exclusive` constraints must
+all pass before execution eligibility is considered. A missing required value,
+constraint failure, runtime error, or MCP resolution error is a contract bug;
+none is a valid reason to skip an example.
+
+Example execution defaults to contract validation only. Runtime execution is
+opt-in: an example enters `dry_run` only when its final `ToolSpec` publishes an
+explicit reviewed dry-run capability. The test never injects `--yes`, and
+`risk`/`confirmation` values do not manufacture preview support. A narrow
+runtime precondition that cannot be derived from the typed contract may use an
+exact zero-based `example_dispositions` entry with `mode=contract_only`,
+`reviewed=true`, one of the schema-enumerated reason codes, and a concrete
+non-empty reason. Such a disposition may only narrow an explicit dry-run
+capability; it cannot turn an ordinary contract-only example into a skip.
+Duplicate, missing, and out-of-range indexes fail validation. Never catch a
+dry-run failure and dynamically downgrade it to `contract_only`.
+
+Normal Go tests run the exhaustive contract gate. Run
+`make test-schema-agent-examples` to additionally execute the eligible subset
+through the real Cobra `--dry-run` path with isolated HOME and blocked proxies.
+The test reports stable `total`, `contract`, `dry_run`, `contract_only`,
+`reviewed_manual`, and per-reason counts; changing those counts requires a
+review of the corresponding typed dry-run capability or manual disposition.
+This target is also part of `make policy`.
+
+Treat every tool `use_when` entry as a reviewed positive selection scenario
+whose expected result is that tool's canonical path, and every `avoid_when`
+entry as a reviewed negative scenario that must not choose that tool. The
+deterministic gate derives a typed evaluation fixture from these same fields;
+it requires exact tool coverage, a real runnable `BoundCommandRegistry`
+primary command, at least one positive and negative assertion per tool, and no
+literal contradictory expectations. It does not claim that string matching
+proves natural-language understanding.
+
+Semantic selection is an explicit opt-in live-model check. Run the smoke set
+(one positive and one negative scenario per product) with
+`DWS_AGENT_SELECTION_LIVE=1 ARK_API_KEY=... ARK_BASE_URL=... ARK_MODEL=... go test ./internal/app -run TestAgentSelectionArkLive -count=1`.
+Add `DWS_AGENT_SELECTION_FULL=1` to evaluate every committed tool scenario, or
+set `DWS_AGENT_SELECTION_CASES` to comma-separated fixture case IDs. Normal CI
+never calls a model; its blockers remain the reproducible fixture, binding,
+example, provenance, and final-delivery facts.
+
+The live evaluator sends only case IDs/scenarios plus one same-product
+candidate table; expected/forbidden assertions stay local and must never be
+included in the model prompt. Built-in Ark HTTPS bases are allowlisted. A
+different HTTPS provider requires its exact base in
+`DWS_AGENT_SELECTION_ALLOWED_BASE_URLS`; plaintext HTTP is accepted only for a
+loopback test server so API credentials are never sent to an arbitrary clear
+text endpoint.
+
+## Safety metadata
+
+Parameter and safety resolution is mostly source-precedence based and
+value-neutral: do not choose a winner because one value looks stricter. A
+higher-priority reviewed metadata/explicit source may intentionally raise or
+lower description, mapping, `effect`, `risk`, `confirmation`, or `idempotency`.
+Preserve all candidates and the selected source in provenance, and fail
+same-precedence conflicts rather than silently merging them.
+
+`required` is the exception. Cobra `MarkFlagRequired` is a hard floor: the
+final Agent projection must keep `required=true` and cannot be lowered by a
+lower-precedence source. A higher-precedence declaration may still raise an
+optional flag to required. `cli_required` continues to mirror the executable
+Cobra marker.
+
+For command-level description: **declare required, delivery Long may win**.
+`ContractDecl.Description` is mandatory at construction (declaration evidence).
+Catalog delivery prefers Cobra Long when present (provenance `cobra_help`,
+resolution `cobra_help_preferred`); without Long, the declared Description is
+delivered as `contract_final`. Title keeps declared ContractDecl /
+ContractFinal first, then Cobra Short, then MCP metadata. This is one authority
+chain with an explicit delivery preference — not two competing sources.
+Generic RPC prose may remain an unselected provenance candidate (and
+parameter-level `interface_description`); it must not overwrite a specialized
+leaf's title or description.
+
+For every delivered `ToolSpec` and `ParameterSpec` field, the provenance
+winner value must exactly equal the delivered value. Checking only source,
+count, presence, or hash is not a sufficient final-delivery invariant.
+
+The same resolved `ToolSpec` must drive every projection. The full leaf payload
+must equal the corresponding tool in `schema --all` and the full Catalog tool.
+Overview/product/group summaries and Catalog summaries must equal
+`ToolSpec.ToSummaryPayload()`. An alias lookup may change only the view fields
+`cli_path` and `is_alias`; it must not re-resolve or mutate the command
+contract.
+
+This build-time rule is distinct from runtime drift handling. If shipped Help
+and leaf Schema disagree, pass only flags accepted by Cobra. For conflicting
+safety information, do not silently take the less restrictive behavior: use
+the safer interpretation or stop and report the contract drift.
+
+Do not infer one safety field from another. In particular, `effect=destructive`
+or `risk=high` does not mechanically rewrite `confirmation`; the final
+precedence winner for each field is authoritative. When
+`confirmation=user_required`, obtain confirmation before adding `--yes`.
+Keep CLI confirmation behavior and Schema metadata consistent, and add a
+semantic regression test through the final embedded loader/query delivery
+path; a generator unit test or JSON count alone is insufficient.
+
+## Unified result Schema and performance
+
+The unified runtime envelope and the per-command Schema result declaration are
+related but distinct contracts:
+
+- Runtime owns the outer machine envelope (`ok`, `outcome`, `data`, `error`,
+  `meta`) and derives it through `internal/output`. Business commands return a
+  `CommandResult`; they must not hand-author the outer JSON shape.
+- A leaf `Contract.Result` / `contract.ResultSpec` describes the reviewed
+  business value inside `data`. It may declare `outcomes`, `data_schema`, and
+  `sensitive_paths`. `Contract.Pagination` is a separate command capability
+  because pagination is emitted under envelope `meta`, not inside `data`.
+- `outcomes` is the set of results a command may produce; it is not the outcome
+  of the current invocation. `data_schema` is a JSON Schema object for business
+  data and must not duplicate the framework envelope.
+- Result declarations are delivered in the full leaf and in the reviewed
+  `--compact` Agent projection. Compact retains the normalized `result` object
+  verbatim but still omits provenance, interface bindings, and other audit-only
+  fields. Product/group summaries remain navigation views and need not repeat
+  every leaf Result. When an Agent needs return-shape facts, query the compact
+  leaf directly; do not load the whole full Catalog.
+- A missing `result` means “no reviewed return-value declaration is published
+  for this leaf.” It does **not** prove that the runtime is legacy, and it must
+  not be filled by inference from examples, MCP samples, or previous command
+  output. Runtime rollout remains an internal per-command fact.
+- The public contract has no `contract_version`, no `--output-contract`, and no
+  Agent-selectable protocol alias. Agents continue to request machine output
+  with `--format json`; migrated commands use the unified result directly and
+  unmigrated commands retain their current legacy output.
+- Existing `dev` / `devapp` pilot coverage is gradual. Active reviewed
+  `devapp` shortcuts are gated on a non-empty Result declaration, while `dev`
+  currently has representative Result coverage. Do not describe that as
+  repository-wide coverage. Any newly activated Agent-visible command should
+  add and test its Result declaration; the remaining pilot gaps should shrink,
+  not expand.
+
+The compact/full leaf `result` object has one stable shape:
+
+```json
+{
+  "result": {
+    "outcomes": ["success", "pending", "partial_failure", "failure"],
+    "data_schema": {
+      "type": "object",
+      "properties": {
+        "items": {
+          "type": "array",
+          "items": {
+            "type": "object",
+            "properties": {
+              "id": {"type": "string", "description": "Stable resource ID"},
+              "name": {"type": "string", "description": "Display name"}
+            }
+          }
+        }
+      }
+    },
+    "sensitive_paths": ["credential.secret"]
+  },
+  "pagination": {
+    "kind": "cursor",
+    "cursor_parameter": "cursor",
+    "meta_path": "meta.pagination",
+    "endpoint_exhausted_path": "meta.pagination.endpoint_exhausted",
+    "next_token_path": "meta.pagination.next_token"
+  }
+}
+```
+
+Field rules:
+
+| Field | Required | Contract |
+|---|---|---|
+| `outcomes` | yes | Non-empty unique subset of `success`, `pending`, `partial_failure`, `failure`; normalization publishes canonical order. |
+| `data_schema` | yes | One recursive JSON Schema **object** describing only the runtime envelope's `data` value. Every named `properties` child must have a non-empty `description`. It must not duplicate `ok`, `outcome`, `error`, or `meta`. |
+| `sensitive_paths` | no | Unique safe dot paths relative to `data`; renderers/redaction consumers must not treat them as shell/JQ expressions. |
+
+Optional members are omitted, never emitted as `null`. A leaf without a
+reviewed Result omits the entire `result` key. Compact must preserve the same
+normalized Result value as the full leaf; it must not summarize, infer, rename,
+or independently rebuild any Result field. Product/group summaries do not
+aggregate child Result objects.
+
+`pagination` is a sibling of `result`, not a child. It declares the canonical
+CLI cursor parameter and the fixed framework paths under `meta.pagination`.
+Product response fields used to derive that metadata remain mapper internals;
+they are not part of `result.data_schema`. Do not execute a second request to
+derive pagination metadata.
+
+Invalid result declarations fail closed during normalization: unknown or
+duplicate outcomes, a non-object/multiple `data_schema`, unsafe or duplicate
+sensitive paths, unsupported pagination kinds, attempts to override framework
+meta paths, and an invalid cursor parameter must be rejected rather than
+silently removed.
+Full-leaf wire round trips must
+preserve the normalized Result exactly. Do not commit generated Schema JSON as
+evidence; tests construct contracts in Go and runtime/CI assemble the Catalog
+from declarations.
+
+### Performance model and rules
+
+- Catalog construction is declaration-driven and cached through the existing
+  lazy `sync.Once` delivery path. Do not reassemble or reopen annotations per
+  command invocation, per leaf lookup, or per renderer.
+- Normalizing one Result declaration is linear in the size of that declaration.
+  Full `schema --all` is linear in tools + parameters + Result schema bytes and
+  is an audit/compatibility export, not the normal Agent discovery path.
+  Overview → compact product/group → compact leaf remains the normal route;
+  only the final leaf carries its Result declaration.
+- Constructing a `CommandResult` defensively clones result data and validates
+  invariants; rendering is buffer-first and then writes once. Both CPU cost and
+  transient memory are O(payload size), with roughly one additional in-memory
+  rendered copy. This buys immutability and prevents partial JSON leakage, but
+  it is not free.
+- Large list/search commands must use bounded pages and publish continuation
+  facts. The current emitter buffers one command result/page before publishing;
+  pagination is the memory bound. Continuous event streams are a separate,
+  command-specific protocol and are not described by `ResultSpec`.
+- A `dual_validate` command must execute the business request exactly once,
+  validate a shadow unified result, and preserve legacy bytes. Never obtain
+  validation by issuing a second network or write request.
+- Filters and alternate formats are render-time work over the same in-memory
+  result. They must not rerun the business operation or rebuild Schema.
+- Performance changes must preserve the one-result, buffer-first, fail-closed,
+  and atomic `--output` guarantees. Do not trade correctness for a microbenchmark
+  improvement. For a material hot-path change, benchmark representative small
+  and page-sized payloads and report allocations/bytes as well as latency.
+
+## Current Schema boundaries
+
+- `schema list` remains a progressive overview. `schema --all` is the stable
+  full-export contract: every final `SchemaIndex` tool must contain its
+  complete leaf parameters, constraints, and safety semantics, including an empty
+  `parameters` object for commands without flags. Keep it suitable for the #602
+  compatibility baseline and fail rather than silently emitting a partial
+  export.
+- `schema --all` is not normal command discovery. Use overview -> compact
+  product/group -> compact leaf for routine Agent work. `--compact` is the
+  reviewed positive-field allowlist for Agent context: new full/audit fields
+  must not appear there until explicitly reviewed. A compact full export is not
+  a complete compatibility baseline.
+- `dws <path> --help` defines whether Cobra exposes a path and which flags the
+  executable accepts. A compact leaf defines Agent selection, CLI parameters,
+  constraints, safety/confirmation semantics, and any reviewed `result`
+  contract. Full leaf fields such as `property`, `interface_ref`, and
+  provenance are audit facts. A conflict is contract drift, not permission to
+  guess.
+- Schema and Help describe commands; neither returns DingTalk business data.
+  After discovery, execute the real read/search/list command to obtain data.

CHANGELOG.md

@@ -6,9 +6,725 @@ The format is inspired by [Keep a Changelog](https://keepachangelog.com/) and th
 
 ## [Unreleased]
 
+## [1.0.58-beta.3] - 2026-08-11
+
+### Added
+
+- **Aitable workflow execution and history** — adds `dws aitable workflow run` for confirmed asynchronous execution of scheduled or record-triggered workflows, plus `dws aitable workflow history` for status-, time-, and page-filtered execution records. The commands map directly to `aitable/run_workflow` and `aitable/get_flow_record_list`, validate trigger-specific arguments locally, and document the `executionId` / `instanceId` correlation.
+- **Streaming-card mentions** — `chat +messages-send-card` now accepts
+  `--at-open-dingtalk-ids` and `--at-all` for group cards, passing mention
+  targets to the initial card-creation request and prepending its returned
+  `atTag` to the automatic streaming update.
+- **Personal OA approval events** — personal event consumers now support task
+  creation, completion, redirection, instance start, termination, and
+  completion events, with typed output and matching usage documentation.
+
+### Fixed
+
+- **Machine-readable export and download receipts** — `dws doc export`,
+  `dws drive download`, and `dws drive download --version` now keep progress
+  logs on stderr under `--format json` and emit one JSON result on stdout after
+  a successful local write. The result includes the saved path and byte size;
+  document exports additionally report the node, requested format, job/task
+  ID, and final status.
+- **IM search and card-write safety** — conversation-scoped search now fails
+  closed when the target cannot be verified, and streaming-card updates require
+  business evidence rather than a transport-only success response.
+- **Document shortcut reliability** — document write, readback verification,
+  pagination, template/version discovery, export, media, and local-file
+  workflows now preserve compatibility while rejecting ambiguous write results.
+- **Event runtime-token handoff** — personal `event consume`, `status`,
+  `stop`, and `+listen-im` honor the root `--token` without falling back to a
+  stale OAuth profile. Detached buses negotiate an owner-only, memory-only IPC
+  credential channel; tokens are never placed in child argv, environment,
+  profiles, logs, or run-state files.
+
+### Changed
+
+- **Minutes `permission apply --policy` type** — `--policy` is now declared as
+  an `int` flag and its required check uses `Flags().Changed`, matching the
+  numeric-parameter convention. `--help` reports `int` instead of `string`;
+  accepted values (2/3/4) and gateway behavior are unchanged.
+- **Minutes skill references** — document `permission apply` in both Minutes
+  skill references: list it in the command trees, describe its policy values and
+  how it differs from `permission add`, and add its intent routing.
+- **Chat paging guidance** — typed chat message commands now document
+  `--page-all`, aggregate result shapes, and cursor behavior in CLI Help and
+  Agent selection examples.
+- **Calendar skill parity** — mono and multi Calendar references are aligned to
+  prevent documentation drift without changing CLI behavior.
+- **Release engineering** — CI now shards helper-package changes through the
+  full race suite, widens a flaky stdio idempotency test budget, governs exact
+  reviewed CLI/Schema type migrations, and lets authorized maintainers trigger
+  internal MCP evaluation with a reviewed `/eval` PR comment.
+
+## [1.0.58-beta.2] - 2026-08-10
+
+### Added
+
+- **`dws sheet create-with-data`（新命令）** — 建表并写入初始数据与样式：`--values`（二维数组写默认表）/ `--sheets`（typed table 多工作表，二者必须给一个）/ `--styles`（`cell_styles` / `row_sizes` / `col_sizes` / `cell_merges`，顶层键对齐飞书 snake_case、列表项内字段兼容 camelCase）。所有结构、字段类型与枚举在创建文档之前校验，非法配置不会留下白建的空文档：`--sheets` 按 `table_put` 的输入契约逐字段校验（`columns` 必填且列名非空不重复、`data` 为二维且行宽与 `columns` 一致、单元格仅限字符串/数字/布尔/null、`dtypes`/`formats` 的键须是列名、`mode`/`header`/`allowOverwrite`/`startCell` 类型与取值、单表 30000 单元格上限），并拒绝未知键、snake_case 变体、`{"sheets":"bad"}` 这类畸形包装与 `sheetId`（服务端会静默丢弃写错的键，导致"只写了表头却报成功"的静默丢数据）；`--values` 校验单元格为标量并受 30000 单元格 / 2000000 字符上限约束；`--styles` 的顶层键与列表项内字段同样拒绝未知键，避免样式只应用一半。写入后回读校验按 `startCell` / `header` / `mode` 推算的首个预期非空单元格，而非固定 A1。该命令是多步编排（建文档 → 探活 → 定位默认工作表 → 写数据 → 回读 → 可选样式），因此如实声明为独立叶子 `sheet.create_with_data` + `interface_mode: composite`（附评审 reason，按契约不带 `interface_ref`）；**`dws sheet create` 保持原样不变**——仍是一次 `create_workspace_sheet` 直连（`interface_mode: mcp`），不新增 flag，避免让 Schema 消费者把编排步骤的参数误当成该 RPC 的入参。
+- **`dws sheet export-csv`（新命令）** — 同步导出单个工作表为 RFC4180 CSV，支持 `--sheet-id` 选表、`--range` 限定范围、`--value-render-option` 选取值模式；`--output` 落盘（为目录时按 `sheet-export.csv` 命名，落盘走 `AtomicWrite` 原子替换，写入失败时已有文件保持原样；父目录不存在按错误处理，不会自动创建），不传则把纯 CSV 打到 stdout（警告只走 stderr）。数据超出单次读取上限时默认报错、既不输出也不写文件，需 `--allow-truncated` 显式接受不完整结果。响应缺 `csv` 字段或类型不对一律报错，不会用 0 字节覆盖已有文件。该分支读的是 `get_range_as_csv`、与 xlsx 的异步导出任务毫无关系，因此独立成叶子 `sheet.export_csv` 并如实声明 `interface_mode: mcp` + `interface_ref: get_range_as_csv`；**`dws sheet export` 保持原样不变**——仍只导 xlsx（`interface_ref: submit_export_job`），flag 面仍是 `--node` / `--output`，csv 专属 flag 不会出现在它上面（此前挂在同一条命令上时，漏写 `--export-format csv` 会让 `--range` 被静默丢弃而导出整篇工作簿）。
+- **`sheet update-dimension --size-type`** — `pixel` / `standard`（恢复默认行高列宽）/ `auto`（按内容自适应行高，仅 ROWS）。
+- **`sheet replace --match-formula`** — 在公式文本中查找替换。
+- **`sheet range set-style` 扩展样式维度** — 新增 `--font-style`（斜体）/ `--font-line`（下划线、删除线）/ `--font-family` / `--border-styles-json`（四边边框；每条边只接受 `style` / `color`，未知键与非字符串 `color` 直接报错，不再静默忽略而画出无颜色的边框，`set-style` / `batch-set-style` / `create-with-data --styles` 三条路径同源校验）。
+- **`sheet range batch-set-style --ranges`** — 一组样式刷多个带工作表前缀的区域，组装为一次原子 `batch_update`。
+
+### Changed
+
+- **Chat message post-send ID handoff** (#897) — CLI Help and bundled Skills
+  now document the `send` → `query-send-status` → `edit`/`recall` workflow,
+  so callers can reuse returned task, message, and conversation IDs instead
+  of searching message history by content.
+- **Sheet mono/multi Skill alignment** — replaces the oversized mono Sheet
+  reference with the progressive routing layout, aligns all 20 Sheet topic
+  references across the mono and multi bundles, and adds a content-policy guard
+  that prevents the paired topic trees from drifting again.
+- **`sheet range set-style` 后端切换为 `set_cell_range`** — 样式统一走 cellStyles 路径（仅设样式、保留原值），这是斜体/下划线删除线/字体族/边框唯一可用的通道。`interface_ref` 由 `update_range` 变为 `set_cell_range`，12 个样式 flag 改为 reviewed mapping exclusion。CLI 用法向后兼容、无 flag 删除；schema-compatibility 经 reviewed 豁免判定为兼容（0 changed fields）。
+- **`sheet range batch-set-style` 改为单次原子提交** — 由本地循环多次 `update_range` 改为一次 `batch_update`，任一项失败默认整批回滚；`--continue-on-error` 由本地控制改为透传服务端。新增批量上限：最多 100 个区域且累计不超过 200000 个单元格。
+- **`sheet range batch-clear` / `batch-set-style` 的 `--ranges` 拒绝空白工作表前缀**（用户可见行为变更）— 此前只按原始串里 `!` 的位置判断，`" !A1:B2"` 修剪后工作表名成了空串，操作却照样带着 `sheetId: ""` 提交：服务端要么让整批 `batch_update` 失败，要么更糟——落到默认工作表而不是用户指定的那张表，且命令报成功。现在工作表名与范围都必须在修剪之后仍非空，否则在发起任何请求之前报错。`batch-set-style --batch` 的纯空白 `sheetId` / `range` 同样拒绝（此前只挡空字符串）；`--batch` 下发仍用原值不替用户修剪，因为 `sheetId` 可以是允许带首尾空格的工作表**名**。两条 `--ranges` 路径现在共用同一个拆分器。
+- **`sheet insert-dimension` / `delete-dimension` / `update-dimension` 的 `--length` 严格校验**（用户可见行为变更）— 解析由 `fmt.Sscanf("%d")` 改为 `strconv.Atoi`。此前只消费前缀数字，`--length 2x` / `3foo` 会被静默当成 `2` / `3` 并对错误的行列数执行操作（删除方向不可回滚）；现在整个值必须是合法正整数，否则报错「`--length` 必须为正整数（>= 1）」且不发起任何请求。**升级影响**：原先依赖这种宽松解析、在传畸形 `--length` 的脚本会开始报错，请把参数修正为纯数字。合法数字值行为不变，上限仍为 5000。`add-dimension` 的 `--length` 是 `Int` 类型 flag，一直由 cobra 严格校验，不受影响。
+- **CLI 接口兼容门禁支持 reviewed flag 类型豁免**（无用户可见变更）— `authoritative-interface-integrity` 与 `check-command-compatibility.sh` 此前一律拒绝历史命令的 flag 类型变更，即使新类型只是把同一套校验从 RunE 前移到解析期，也没有任何评审通道。现在两道门禁各带一张精确豁免表：命令路径 + flag 名 + 旧类型 → 新类型四元组全等才命中、方向敏感（`string`→`int` 与 `int`→`string` 是两个不同的键，只有被评审的方向可用），且仅当该 flag 的其他契约（shorthand / required / hidden / no-opt / scope）纹丝不动时才放行，因此豁免夹带不了别的破坏。首条也是目前唯一一条登记的是 `dws minutes permission apply --policy` 的 `string` → `int`（配合 #912）：旧实现在 RunE 里做 `strconv.ParseInt(v, 10, 64)` 再校验 `[2,4]`，新实现由 pflag 以 `strconv.ParseInt(s, 0, 64)` 解析后仍校验 `[2,4]`，**历史上能成功的调用集是新调用集的子集**（base 0 额外接受 `0x3` 这类写法，只放宽不收紧），非法值依然失败、只是报错文案与时机前移；flag 默认值由 `""` 变 `"0"` 是类型的必然结果，两道门禁都不比较默认值，且该 flag 必须显式给出、默认值不可达。两张表必须逐字一致并有守卫测试锚定漂移——重复是被迫的而非选择：`check-authoritative-interface-baselines.sh` 会把整个 `scripts/policy/interface-baseline` 目录复制进检出历史版本的 worktree 再编译，那份拷贝不能 import 本分支新增的包。
+- **Schema 兼容门禁支持 reviewed 参数类型豁免**（无用户可见变更）— 接上一条。`schema-compatibility` 是同一个 `Interface Integrity` job 里排在两道 CLI 接口门禁之后的第三道检查，此前也一律拒绝已发布参数的 `type` 变更。由于前两道先失败、`set -e` 让它从未在 CI 上暴露，上一条豁免只解决了三分之二。现在 `checkParameterCompatibility` 也带一张精确豁免表：`<product>/<tool id>` + 参数名 + 旧类型 + 新类型四元组全等才命中、方向敏感，且仅当该参数**除 `type` 外的全部已发布字段逐字段相等**时才放行。这里刻意用相等性比较而非「没有产生其他兼容性错误」：放宽 `required` / `cli_required`、清空 `required_when`、扩宽 `enum`、清空 `interface_type`、经 reviewed mapping exclusion 清空 `property`——这些变化单独看都是兼容的、根本不产生错误，若以错误列表代替相等性检查，它们就能搭着一次已评审的类型迁移一起蒙混过关。结构体整体比较还意味着将来给 `parameterSchema` 新增字段时会自动纳入守卫，而不是悄悄放宽每一条既有条目。唯一条目是 `minutes/minutes.apply_minutes_permission` 的 `policy` 由 `"string"` 迁移到 `"integer"`（配合 #912）：该 `type` 由 Cobra flag 类型投影而来（provenance `cobra_flag_type`），描述的是 CLI 如何接受取值；消费方据此拼装的是命令行，而 `--policy 4` 在两种声明下是同一个 argv，加引号的 `--policy "4"` 到 pflag 仍是 4，RunE 也仍校验 `[2,4]`——而且该参数映射的 property `policyId` 一直以数字上报，新声明比旧声明更贴近真实请求。表里的类型值必须是 `schemaType` 实际产出的带引号形态（`"string"` 而非裸 `string`），守卫测试用 `schemaType` 复算并校验类型名属于 JSON Schema 的封闭取值集合——`reviewedInterfaceRefRedirect` 曾因键的书写形态错误两次静默失效，这里不重犯。
+
+### Fixed
+
+- **Event runtime-token handoff** — personal `event consume`, `status`, `stop`, and `+listen-im` now honor the existing root `--token` instead of falling back to a stale local OAuth profile. Detached personal-event buses negotiate the credential only after an additive capability handshake, receive and rotate it through owner-only local IPC, and keep it in memory; the token is never forwarded through child argv, environment variables, profiles, logs, or run-state files. Existing OAuth and multi-profile behavior is unchanged when `--token` is absent. A new client refuses to send a runtime token to an older bus and leaves its existing consumers and subscriptions untouched; the recovery message asks users to inspect `event status --as user`, preview `event stop --as user --all --dry-run`, and explicitly confirm `event stop --as user --all --yes` before retrying.
+
+## [1.0.58-beta.1] - 2026-08-07
+
+### Added
+
+- **Robot image and file messages** (#867) — `dws chat message send-by-bot`
+  now supports image URLs and local-file uploads through explicit message
+  types, while retaining Markdown as the default and preserving its existing
+  title and text requirements.
+- **Conversation shortcut-bar management** (#877) — adds `dws chat toolbar`
+  commands to list, add, hide, sort, and manage custom conversation shortcuts,
+  with validation and confirmation for destructive removal.
+- **Complete AI Table Shortcut surface** (#901) — makes all 92 supported
+  AI Table Shortcuts discoverable through Runtime Schema and adds reliable
+  Base, table, record, attachment, view, dashboard, and workflow operations
+  with explicit confirmation and result-verification semantics for writes.
+
+### Changed
+
+- **Doc import upload fallback** — `dws doc import` no longer fails on file
+  formats outside the conversion whitelist (html, pdf, zip, extensionless,
+  and any future format): it now hands the file to the document-space upload
+  chain (the same primitive as `dws drive upload --workspace`), stores the
+  original file at the requested `--folder`/`--workspace` target, and prints
+  an explicit stderr notice with the supported-format list and the
+  convert-to-md alternative. The fallback shares the import file checks
+  (20MB cap, empty-file guard), keeps `--format json` / `--dry-run` output as
+  a single JSON document, and marks the machine-readable result with
+  `fallback: "upload"` and `converted: false` so agents never mistake the
+  stored file for a converted online document. The fallback fails closed
+  unless the commit response parses as JSON and carries a file identity
+  (exposed as `dentry_id`); empty or unverifiable responses surface as
+  errors instead of fabricated success. Importable formats and
+  `dws sheet import` validation are unchanged.
+- **IM natural-target and history alignment** — Chat shortcuts can resolve natural user/group targets before execution, and message-history workflows expose bounded time ranges, ordering, explicit all-page controls, continuation ledgers, safe local export, and thread-reply pagination without treating empty or incomplete reads as successful results. Bundled mono/multi Skills and intent routing now describe the same executable surface.
+- **Sheet CSV formula writes** — `dws sheet csv-put` and batch `csv-put` now expose the service contract that CSV fields beginning with `=` are written as formulas. Prefix the field with an apostrophe to write literal text beginning with `=`; CSV content continues to pass through unchanged.
+- **Release-equivalent PR compatibility gate** (#889) — pull-request
+  admission now runs command-surface compatibility checks against the current
+  release baseline before code reaches `main`.
+- **Reviewer routing governance** (#903) — updates the Reviewer Router pool
+  used for new ready PRs while retaining the existing current-head review and
+  required-check gates.
+
+### Fixed
+
+- **Fail-closed IM pagination and audit evidence** — `+chat-messages`, `+search-msg`, `+thread-replies`, `+at-me`, `+my-groups`, conversation lists, and favorites preserve partial-read failures, reject missing or stalled continuation state, deduplicate page boundaries, and publish completion evidence. The live-audit regression suite now rejects empty projections and incomplete reads instead of promoting them to passing results.
+- **Sheet formula verification** (#873) — `dws sheet formula-verify` now calls
+  the registered remote tool name `verify_formula`; the previous
+  `formula_verify` name failed at gateway dispatch.
+- **CLI and parameter recovery boundaries** (#864) — command and parameter
+  recovery now fail closed when an Agent-provided path or flag cannot be
+  reconciled with the executable CLI surface, reducing unsafe hallucinated
+  retries.
+
+## [1.0.57-beta.4] - 2026-08-06
+
+### Added
+
+- **Expanded open CLI workflows** (#887) — adds calendar event-instance
+  queries, Drive latest-file selection, Markdown diff, Mail calendar/export/
+  share-to-chat workflows, and Minutes hot-word, permission, and audio-memo
+  operations, with matching Schema and cross-platform coverage.
+
+### Changed
+
+- **Multi-skill framework alignment** (#887) — folds long-tail skills into
+  `dingtalk-misc`, renames the shared package to `dingtalk-shared`, removes
+  stale Preview guidance, and reorganizes shared recipes and routing for more
+  predictable Agent selection.
+- **Bounded Agent Schema delivery** (#887) — keeps compact and wire projections
+  focused on executable contract facts, retires stale MCP metadata candidates,
+  and teaches Agents to prefer `dws schema --compact` for bounded context.
+
+### Deprecated
+
+- **Recovery and discovery-cache compatibility surfaces** (#887) — keeps
+  visible Deprecated `dws recovery` and `dws cache` compatibility stubs while
+  retiring their former recovery engine and dynamic discovery-cache behavior.
+  Recovery plan/execute/finalize now return an explicit “不再支持” notice, and
+  Skills no longer teach either retired workflow.
+
+### Fixed
+
+- **Mail share-to-chat confirmation** (#887) — requires explicit confirmation
+  before the first remote write, while preserving the confirmed sign-retry
+  flow and covering both direct-success and retry responses.
+
+## [1.0.57] - 2026-08-06
+
+This stable release promotes the fully delivered `v1.0.57-beta.4` baseline.
+It includes the v1.0.57 beta-line command-contract, document, chat, OA, Wiki,
+and compatibility improvements, plus the multi-skill framework alignment and
+expanded calendar, Drive, Markdown, Mail, and Minutes workflows validated in
+the final prerelease.
+
+- **Promote v1.0.57-beta.4** — publishes the final validated prerelease
+  baseline as stable `v1.0.57` without adding post-beta product changes.
+
+## [1.0.57-beta.3] - 2026-08-06
+
+### Added
+
+- **Reviewed document shortcuts** (#880) — adds public document shortcuts for
+  safe local downloads, content and history, review, media and style, and
+  document access/sharing workflows, while retaining reviewed compatibility
+  identities and confirmation safeguards for writes.
+- **Mentions in chat replies** (#881) — `chat message reply` now supports
+  `--at-open-dingtalk-ids` and `--at-all`, forwarding reply mention fields and
+  adding any required mention placeholders without changing existing send
+  behavior.
+
+## [1.0.57-beta.2] - 2026-08-05
+
+### Fixed
+
+- **Stable Chat command compatibility** (#876) — restores the hidden migration
+  entries for `chat send`, `chat history`, and their `im` aliases, preserving
+  the v1.0.56 command surface while directing callers to the supported
+  `chat message send/list` commands. Legacy flags now reach the same migration
+  hints instead of failing during flag parsing.
+- **Drive download cancellation-test stability** (#876) — replaces a
+  timing-sensitive worker-cancellation coverage test with a deterministic seam,
+  reducing flaky CI without changing download behavior.
+
+## [1.0.57-beta.1] - 2026-08-05
+
+This beta starts the v1.0.57 line on top of v1.0.56. It packages the unified
+command-contract and runtime Schema architecture, complete Multi IM Chat
+coverage, document whiteboard and OA approval workflows, Wiki activity feeds,
+and compatibility and CI reliability fixes.
+
+### Added
+
+- **Contact personal-status updates** (#872) — adds `contact user update-ownness`
+  (alias `set-ownness`) for updating a user's personal status text. The write
+  operation maps reviewed `userId` and `ownnessText` parameters to the service
+  contract and requires confirmation unless `--yes` is explicitly supplied.
+- **Document whiteboard workflows** (#861) — adds `doc whiteboard insert`,
+  `whiteboard query/update`, and `doc media upload`. These commands support
+  confirmed document-embedded whiteboard creation and updates, structured
+  OpenNodes reads, and preparation of node-bound Vector/SVG resources.
+- **Complete Multi IM Chat coverage** (#860) — hardens deterministic group and
+  stable-ID resolution, sending, querying, downloading, pagination, and JSON
+  export. The remaining reviewed Chat Shortcuts enter Schema coverage, with
+  destructive delete and clear operations aligned to confirmation gates.
+- **OA approval form workflows** (#853) — adds OA form-schema lookup,
+  process forecast, and confirmed approval-instance creation, supporting both
+  simple flags and complete `--request` payloads.
+- **Wiki activity-feed queries** (#862) — adds `wiki feed list` to retrieve
+  workspace document activity, with cursor paging and optional file exclusion.
+
+### Changed
+
+- **Unified command and Schema contract framework** (#830) — Leaf commands and
+  Shortcuts now use the shared typed `corecmd` base for flags, constraints,
+  confirmation, Help, and runtime Schema projection. Schema delivery assembles
+  from leaf Contract declarations at runtime; the retired hint overlays,
+  pinned MCP metadata, and committed Catalog artifacts are no longer delivery
+  authorities.
+- **Faster macOS CI without reducing native coverage** (#857) — narrows the
+  macOS race suite to Keychain, codesign, and Darwin-only tests while adding a
+  reachability contract that prevents native-only tests from being silently
+  excluded.
+
+### Fixed
+
+- **Chat media-download JSON compatibility** (#854) — restores parseable
+  `success`, `downloadUrl`, and `output` fields for
+  `chat message download-media --format json` after a successful download,
+  without progress output corrupting JSON stdout.
+
+### Added
+
+- **Document-embedded whiteboard workflows** — adds `doc whiteboard insert` for confirmed creation and part-ID verification, `whiteboard query/update` for structured OpenNodes reads and confirmed writes, and `doc media upload` for preparing node-bound Vector/SVG resources. The public adapter uses an explicit helper-only whiteboard endpoint, validates update envelopes locally, decodes `resultJson`, and publishes the full command, Schema, Skill, and safety contract migrated from `dws-wukong@e2da8ab947c6`.
+- **Robot image and file messages** — extends `chat message send-by-bot` with public image URL delivery through `--msg-type image --image-url` and local-file upload/send through `--msg-type file --file-path`, while preserving Markdown as the default message type.
+
+### Changed
+
+- **Chat reply mentions** — `dws chat message reply` can @ specified group members with `--at-open-dingtalk-ids` or @ everyone with `--at-all`, forwarding the existing `send_personal_message` mention fields and automatically adding missing current-user `<@id>` / `<@all>` placeholders.
+- **Pinned MCP metadata retired** — deletes `internal/cli/schema_mcp_metadata.json` and removes its embed/loader/fallback role from Schema assembly. Catalog now assembles from Contract/ParamDecl/Interface + Cobra only; `make fetch-mcp-metadata` remains an optional diagnostic dump under `artifacts/` and refuses the retired pin path. Policy bans the pin from reappearing.
+- **MCP service review retired** — deletes `schema_mcp_service_review.json` and removes its policy jq / outputguard / test disposition gate (`notify` → `out_of_surface`, snapshot hash pin). No replacement ledger.
+- **Hints retired; ContractDecl is the leaf Schema source** (#830) — `schema_hints/`, Manual/Schema hint overlays, and `schema_agent_metadata/` delivery are removed. Selection, safety, parameters, and interface facts declare on ProductDecl / leaf `Contract` (`corecmd.ContractDecl` + `contract.ParamDecl` / `Safety`). Authoring renamed `SchemaDecl` → `ContractDecl`; nested fields reuse `contract.*` directly.
+- **Contract package seam** (#830) — types / ProductDecl live under `internal/corecmd/contract` (DTO only). Annotate writers live in `internal/corecmd/runtimeannotate`; Cobra-keyed ContractFinal store + Register live in `internal/corecmd/contractfinal`; homology gates in `internal/cli/homology`. All packages import `corecmd/*` directly; the former `cli/runtimeannotate` / `cli/contractfinal` shim packages are removed, and the `cli` root keeps only package-local aliases (`runtime_schema_seam.go`). Catalog/`ResolveMeta` stay on the `cli` delivery root. `internal/corecmd` must not import any `internal/cli` package.
+- **CommandMeta cache for ResolveMeta** — production `ResolveMeta` / leaf `--help` Safety project from the runtime-assembled `SchemaRegistry` into a `map[cli_path]CommandMeta` installed during `deliverySchemaCatalog` sync.Once. Steady-state lookups are O(1); full Catalog wire maps stay deferred. Registry `Source` stamps `runtime-assembled`.
+
+### Fixed
+
+- **Fail-closed IM pagination and audit evidence** — `+chat-messages`, `+search-msg`, `+thread-replies`, `+at-me`, `+my-groups`, conversation lists, and favorites preserve partial-read failures, reject missing or stalled continuation state, deduplicate page boundaries, and publish completion evidence. The live-audit regression suite now rejects empty projections and incomplete reads instead of promoting them to passing results.
+- **Unified command safety and Shortcut runtime (H0)** — Shortcut leaves now execute through `corecmd.New`, sharing the same typed Safety confirmation gate as Leaf commands. EOF / closed stdin returns `confirmation_required`, and interactive `no` returns the existing non-zero cancellation validation error instead of reporting success for an operation that did not run. Pass `--yes` or `--dry-run` to skip the prompt.
+- **Constraint "provided" for `at_least_one` / `exactly_one` (H0)** — a flag set to an empty string (`--flag ""`) no longer counts as provided; previously bare Cobra `Changed` satisfied the constraint. Pass a non-blank value for a member of the group.
+- **Chat media download JSON compatibility** — `dws chat message download-media --format json` once again returns a clean `{success, downloadUrl, output}` result after the file is saved, preserving the temporary URL and resolved local path without progress text corrupting JSON stdout.
+
+## [1.0.56] - 2026-08-04
+
+This stable release promotes the fully delivered `v1.0.56-beta.4` baseline.
+It includes PR #852's resilient multipart Drive download implementation,
+together with the v1.0.56 beta-line command, Schema, Skill, and runtime
+improvements already validated through the prerelease channel.
+
+### Added
+
+- **Resilient multipart Drive downloads** (#852) — `drive download` and
+  `drive download-version` support parallel chunk transfer, Range probing,
+  fingerprint-validated checkpoint resume, automatic 401/403 credential
+  refresh, and graceful interruption with checkpoint preservation.
+
+## [1.0.56-beta.4] - 2026-08-04
+
+This beta adds PR #852 on top of v1.0.56-beta.3. It makes Drive downloads
+resilient for large files through parallel transfer, validated resumable
+checkpoints, and automatic credential refresh.
+
+### Added
+
+- **Multipart Drive downloads** (#852) — adds `--part-size`, `--parallel`, and
+  `--no-resume` to `drive download` and `drive download-version`. Files above
+  the part-size threshold use a Range probe and parallel chunks, resume from a
+  fingerprint-validated checkpoint, refresh credentials on 401/403, and keep
+  the checkpoint when Ctrl+C interrupts a transfer.
+
+## [1.0.56-beta.3] - 2026-08-03
+
+This beta adds PRs #846 and #851 on top of v1.0.56-beta.2. It adds a
+service-provided Aitable workflow-editing reference command and makes local
+event-bus IPC reliable on shared filesystems by placing Unix sockets in a
+validated private runtime directory.
+
+### Added
+
+- **Aitable workflow editing reference** (#851) — adds `dws aitable workflow edit-example`, a parameter-free read command that returns the service-provided workflow editing documentation and `workflow-dsl/v1` examples through `aitable/edit_workflow_example`.
+
+### Fixed
+
+- **Event bus sockets on shared filesystems** (#846) — Unix event buses now place their local IPC socket in a private per-user runtime directory (`XDG_RUNTIME_DIR` when available, otherwise a `0700` per-UID directory under the system temporary directory) while retaining locks, metadata, logs, and subscription state in the configured Workdir. Listener and dial paths validate directory ownership and permissions before use. This prevents `dws event consume` from failing with `bind: errno 524` when `~/.dws` is hosted on NFS, CSI, FUSE, or another filesystem that does not support Unix Domain Sockets without exposing the socket directly in a shared `/tmp` root. When `XDG_RUNTIME_DIR` is unavailable, the per-UID directory name is deterministic: ownership validation prevents endpoint hijacking, but another local user can pre-create the directory to deny service; multi-user deployments should provide a private `XDG_RUNTIME_DIR`.
+
+## [1.0.56-beta.2] - 2026-07-30
+
+This beta adds PRs #831 and #835 on top of v1.0.56-beta.1. It separates
+Agent Product observability and IM display identity from the stable
+edition-owned PAT and routing identity, and reduces common-path Skill context
+loading without changing the public command or Runtime Schema surface.
+
+### Changed
+
+- **Agent Product identity separation** (#831) — sends `DWS_AGENT_PRODUCT` through the new `x-dws-agent-product` observability Header and uses a valid non-empty value for the IM `clawType` display label whenever `--ai-tag` is enabled. Because `--ai-tag` defaults to `true`, callers that set `DWS_AGENT_PRODUCT` change the displayed label by default. With `--ai-tag=false`, native `chat message send` / `reply` calls preserve their existing wire shape by sending an empty IM `clawType`, while shortcut calls omit the argument. Unset or empty Product values omit the Header and preserve the active edition's IM display default.
+- **Agent Host dimension convention** (#831) — new integrations should send the runtime form (`cloud` or `desktop`) through `DWS_AGENT_HOST` and report the product separately through `DWS_AGENT_PRODUCT`. Legacy combined labels such as `qwenwork_cloud` remain syntactically valid for compatibility.
+- **Reduced common-path Skill context** (#835) — keeps the complete 97-command Chat Shortcut inventory in Runtime Catalog and leaf Schema while routing common intents through compact Skill tables and references. When an exact command path is already known, the mono Skill no longer requires eager loading of a complete product reference. The generated Skill policy now detects drift, forced full-reference loading, and context-budget regressions; the common Chat plus shared activation estimate drops from 7,301 to 4,771 `o200k_base` tokens without changing the 845-tool Schema surface.
+
+### Fixed
+
+- **Stable PAT/routing identity** (#831) — restores the CLI-emitted open-source HTTP `claw-type` and PAT `hostControl.clawType` to the edition-fixed `openClaw` value. `DWS_AGENT_PRODUCT` no longer changes those wire values, and the client continues to derive PAT, authentication, routing, and Discovery behaviour from the existing independent signals.
+- **Portable generated Skill validation** (#835) — resolves the mono Skill name by scanning upward from the generated target, keeping `--check` independent of the repository checkout path and preventing false drift failures when an ancestor directory resembles a Skill name.
+
+## [1.0.56-beta.1] - 2026-07-30
+
+This beta starts the v1.0.56 line on top of v1.0.55 and packages PRs #817,
+#806, and #834, together with release-validation fixes #838 and #839. It closes
+the remaining Agent-visible IM shortcut gaps, introduces reviewed
+command-scoped parameter normalization without guessing business identifiers
+or values, and prevents deterministic personal-event subscription failures
+from becoming unbounded retry storms.
+
+### Added
+
+- **Complete IM shortcut workflows** (#817) — publishes the previously excluded `+chat-messages`, `+messages-send`, `+messages-send-card`, `+search-msg`, and `+thread-replies` shortcuts in Runtime Schema. Unified send, streaming-card delivery, advanced search, thread replies, and opt-in resource downloads now share reviewed parameters, selection guidance, and runtime-aligned safety semantics.
+- **Reviewed parameter concept normalization** (#806) — adds a closed parameter-concept dictionary and generated command-level alias table, covering reviewed IM synonyms while preserving the boundaries between group, conversation, user, open-user, cursor, and paging identifiers.
+
+### Fixed
+
+- **Message delivery and resource handling** (#817) — resolves direct recipients through exact contact search, preserves rich and nested message resources, avoids same-name download overwrites, and prevents read shortcuts from silently returning empty results on non-interactive input.
+- **Parameter parsing safety** (#806) — rejects ambiguous, blocked, or conflicting aliases before dispatch, normalizes explicit boolean values such as `--dry-run false`, and keeps internal pre-parse handler details out of user-visible errors.
+- **Personal-event subscription retry safety** (#834) — adds cross-process attempt claims, deterministic backoff and jitter, `Retry-After` handling, terminal holds, compare-and-swap completion, and fail-closed state handling across all public personal-event subscriptions, preventing deterministic failures from causing unbounded callback retries.
+- **Scoped CI and release validation reliability** (#838, #839) — keeps scoped coverage aligned with intentionally skipped supporting profiles, gives focused race and Multi-profile E2E suites enough time for the current `internal/app` workload, and preserves hidden E2E diagnostics on failure.
+
+## [1.0.55-beta.8] - 2026-07-30
+
+This beta revalidates the `v1.0.55-beta.7` product baseline through a complete
+guarded release delivery. It carries no new product-facing command behavior;
+the new version is required because the published beta.7 artifacts succeeded
+on GitHub, npm, and Homebrew, but its enabled optional Gitee mirror failed and
+left that Release run ineligible for stable promotion.
+
+### Changed
+
+- **Complete promotion evidence** — republishes the validated v1.0.55 command, Runtime Schema, Skill, authentication, and projection changes with the optional Gitee upload fallback disabled, so the release can produce one successful auditable delivery proof before stable promotion.
+
+## [1.0.55] - 2026-07-30
+
+This release promotes the validated `v1.0.55-beta.8` baseline to stable. It
+expands the public Workspace command surface and personal event consumption,
+makes the full built-in shortcut catalog available to Agents, and hardens
+multi-account routing, authentication compatibility, command safety, and
+response projection across the CLI.
+
+### Added
+
+- **Broader Workspace command surface** (#621, #676) — adds roughly 30 reviewed Drive, Doc, Sheet, and Chat leaf commands synchronized from Wukong, including Drive version and permission operations, document styling, Sheet comment/version/formula verification, and in-place text-emotion updates. A reusable declarative `LeafSpec` framework now delivers command identity, safety, selection, and guarded Help metadata consistently.
+- **Complete Agent-visible shortcut delivery** (#802, #815) — publishes all 210 built-in shortcuts as reviewed Runtime Schema leaves across 16 products, including 88 validated Chat shortcuts, with executable paths, parameters, constraints, selection guidance, dry-run capabilities, and runtime-aligned confirmation semantics.
+- **Expanded enterprise and event capabilities** (#790) — adds the HR Brain talent-pool, employee-profile, and structured-search command families; `dws mcp url get` resolves MCP Market endpoints; personal event consumption supports eight additional IM event keys, multi-key consumers, and targeted shutdown.
+- **Agent integration identity** (#804, #816) — adds validated `DWS_AGENT_HOST` and `DWS_AGENT_PRODUCT` labels for observability and product attribution while keeping them separate from authentication and authorization.
+
+### Changed
+
+- **Progressive multi-Skill guidance and account safety** (#621, #821) — reorganizes bundled product guidance for progressive discovery and restores the mandatory rule that Agents must not guess an account when a multi-account organization has no unique current default.
+- **Supported Chat file delivery** — retires the legacy AppKey/AppSecret-backed `chat media upload` command from discovery and routes local files through `chat message send --msg-type file --file-path`, while callers with an existing media ID can continue sending images directly.
+- **Guarded release delivery** (#791) — strengthens immutable GitHub, npm, Homebrew, optional mirror, recovery, and version-allocation checks while keeping beta and stable publication role-gated and auditable.
+
+### Fixed
+
+- **Shortcut and message projection correctness** (#706, #783, #795) — prevents successful read shortcuts from silently projecting non-empty backend responses to empty results, renders rich, forwarded, and encrypted message forms safely, and fixes group-bot, bot-search, mail-thread, media-ID alias, and Todo paging response handling.
+- **Command contract edge cases** (#803) — makes approval revocation and document rollback honor dry-run before confirmation or preflight, fixes Drive and Doc rename semantics, restores Drive-specific metadata, and validates Todo reminder rules.
+- **Authentication and external-contact compatibility** (#756, #757) — migrates legacy global and organization-scoped credentials without cross-account token borrowing, preserves contacts that expose only `openDingTalkId`, and aligns message-resource flags with message-list output fields.
+
+## [1.0.55-beta.7] - 2026-07-29
+
+This beta supersedes the unpublished `v1.0.55-beta.6` candidate and packages
+PRs #621, #676, #757, #815, #816, and #821. It restores the mandatory
+multi-account safety rule caught by the sealed-release E2E gate while retaining
+the reviewed Wukong capability and multi-Skill synchronization, declarative
+command and Schema delivery, hardened Chat shortcuts, external contact
+resolution, and Agent product identity on top of the `v1.0.55-beta.5` baseline.
+
+### Added
+
+- **Wukong capability and multi-Skill synchronization** (#621) — ports roughly 30 reviewed leaf commands into the open-source CLI across Drive, Doc, Sheet, and Chat, including in-place text-emotion updates, Drive version and permission operations, document styling, and Sheet comment/version/formula verification. The bundled multi-Skill framework is reorganized into progressive product references and routing guidance while retaining current open-source command, response, safety, and Runtime Schema contracts.
+- **Declarative leaf commands and unified metadata delivery** (#676) — adds the reusable `LeafSpec` command framework and migrates 27 DevApp commands without changing their paths or flags. Runtime consumers now resolve identity, safety, and selection through one embedded Catalog-backed API, and guarded Help output publishes the command's safety/confirmation annotation.
+- **Agent product identity** (#816) — adds the optional `DWS_AGENT_PRODUCT` override for the existing HTTP `claw-type` header while preserving each edition's default when unset. Product and runtime labels are caller-declared signals, not authentication credentials; services must validate supported values and must not grant access solely from them. The override does not change the separate IM message-display `clawType` parameter controlled by the edition and `--ai-tag`.
+
+### Changed
+
+- **Reviewed Chat shortcut delivery** (#815) — publishes 88 currently available Chat shortcuts after real-business validation, keeps three confirmed lower-service failures unavailable, strengthens semantic availability and dry-run contracts, and adds safe message-resource download plus group-member listing. Conversation filtering, IM routing/reporting, and member mute resolution are aligned with the validated backend identities.
+- **Agent identity label hardening** (#816) — limits `DWS_AGENT_PRODUCT` and `DWS_AGENT_HOST` to 64 ASCII bytes, trims only surrounding ASCII spaces and tabs, and rejects other control or Unicode whitespace. QwenWork integrations should report the two dimensions separately as `DWS_AGENT_PRODUCT=qwenwork` plus `DWS_AGENT_HOST=cloud` or `desktop`; previously used combined Host labels such as `qwenwork_cloud` remain syntactically valid for compatibility.
+
+### Fixed
+
+- **External-contact and message-resource chaining** (#757) — the shared name-to-ID resolver keeps external or cross-organization contacts that expose only `openDingTalkId`, applies reviewed display-name fallbacks, and preserves organization-only filtering for commands that require `userId`. `chat +messages-resource-url` now accepts `--msg-id` and `--open-message-id` as aliases for `--message-id`, matching message-list response fields.
+- **Multi-account Skill safety contract** (#821) — restores the mandatory rule that an Agent must never choose the first, most recently logged-in, or most recently used account when an organization has multiple accounts without one unique `isOrgCurrent=true` default. A PR-level embedded-Skill regression test now catches removal before the full sealed-release E2E gate.
+
+## [1.0.55-beta.6] - 2026-07-29
+
+This beta packages PRs #621, #676, #757, #815, and #816, validating the Wukong
+capability and multi-Skill synchronization, declarative command and Schema
+delivery, hardened Chat shortcuts, external contact resolution, and Agent
+product identity on top of the `v1.0.55-beta.5` baseline.
+
+### Added
+
+- **Wukong capability and multi-Skill synchronization** (#621) — ports roughly 30 reviewed leaf commands into the open-source CLI across Drive, Doc, Sheet, and Chat, including in-place text-emotion updates, Drive version and permission operations, document styling, and Sheet comment/version/formula verification. The bundled multi-Skill framework is reorganized into progressive product references and routing guidance while retaining current open-source command, response, safety, and Runtime Schema contracts.
+- **Declarative leaf commands and unified metadata delivery** (#676) — adds the reusable `LeafSpec` command framework and migrates 27 DevApp commands without changing their paths or flags. Runtime consumers now resolve identity, safety, and selection through one embedded Catalog-backed API, and guarded Help output publishes the command's safety/confirmation annotation.
+- **Agent product identity** (#816) — adds the optional `DWS_AGENT_PRODUCT` override for the existing HTTP `claw-type` header while preserving each edition's default when unset. Product and runtime labels are caller-declared signals, not authentication credentials; services must validate supported values and must not grant access solely from them. The override does not change the separate IM message-display `clawType` parameter controlled by the edition and `--ai-tag`.
+
+### Changed
+
+- **Reviewed Chat shortcut delivery** (#815) — publishes 88 currently available Chat shortcuts after real-business validation, keeps three confirmed lower-service failures unavailable, strengthens semantic availability and dry-run contracts, and adds safe message-resource download plus group-member listing. Conversation filtering, IM routing/reporting, and member mute resolution are aligned with the validated backend identities.
+- **Agent identity label hardening** (#816) — limits `DWS_AGENT_PRODUCT` and `DWS_AGENT_HOST` to 64 ASCII bytes, trims only surrounding ASCII spaces and tabs, and rejects other control or Unicode whitespace. QwenWork integrations should report the two dimensions separately as `DWS_AGENT_PRODUCT=qwenwork` plus `DWS_AGENT_HOST=cloud` or `desktop`; previously used combined Host labels such as `qwenwork_cloud` remain syntactically valid for compatibility.
+
+### Fixed
+
+- **External-contact and message-resource chaining** (#757) — the shared name-to-ID resolver keeps external or cross-organization contacts that expose only `openDingTalkId`, applies reviewed display-name fallbacks, and preserves organization-only filtering for commands that require `userId`. `chat +messages-resource-url` now accepts `--msg-id` and `--open-message-id` as aliases for `--message-id`, matching message-list response fields.
+
+## [1.0.55-beta.5] - 2026-07-28
+
+This beta validates expanded personal event consumption, complete Agent-visible
+Runtime Schema coverage for all 210 built-in shortcuts, Agent host
+observability, and hardened document, Drive, approval, and Todo command
+contracts on top of the `v1.0.55-beta.4` baseline.
+
+### Added
+
+- **Expanded personal event consumption** (#790) — adds eight IM personal event keys, supports subscribing to and consuming multiple event keys in one `dws event consume` invocation, and adds targeted local-consumer shutdown when a subscription is stopped so other consumers can continue on the shared event bus.
+- **Shortcut Runtime Schema delivery** (#802) — publishes all 210 public built-in shortcuts as reviewed Agent-visible leaf tools across 16 product groups, with stable canonical identities, executable `+shortcut` CLI paths, parameter and cross-parameter constraints, selection guidance, interface metadata, and runtime-aligned safety/confirmation semantics. `dws shortcut list` remains the lightweight batch-discovery view, while leaf Schema now carries the complete Agent contract; declared string-slice defaults are also preserved consistently in Cobra and Schema.
+- **Agent host observability** (#804) — accepts an optional, validated `DWS_AGENT_HOST` label and sends it as `x-dws-agent-host` for logs and BI only; invalid values fail before CLI network activity, and the label never participates in authentication or routing.
+
+### Fixed
+
+- **Command contract edge cases** (#803) — approval revocation and document-version rollback now honor `--dry-run` before confirmation or remote preflight; `drive rename` removes only a suffix matching the node's current extension to avoid duplicate extensions while `doc rename` preserves the caller's exact display name; `doc info` keeps its stable MCP contract while `drive info` restores Drive-only metadata such as a non-null `fileSize`; and Todo reminder writes now reject invalid rule JSON while Help, Schema, and Skills distinguish a due time from an independently unreadable reminder rule.
+
+## [1.0.55-beta.4] - 2026-07-27
+
+This beta validates the shortcut projection fixes for group bots, bot search,
+and mail threads, together with hardened release delivery to Gitee and npm on
+top of the `v1.0.55-beta.3` baseline.
+
+### Fixed
+
+- **Shortcut projection fixes** (#795) — `chat +chat-bots` no longer projects a non-empty `list_group_bots` response to an empty list, `+bot-find` recognizes the `search_bots` response shape (`result.bots` entries with `botOpenDingTalkId`), and mail thread listings keep `lastUpdated` when the backend returns `lastModifiedDateTime`.
+
+### Changed
+
+- **Hardened release delivery** — the Gitee mirror workflow can synchronize a specific release's assets on demand, release lookup tolerates Gitee's HTTP 200 null-body response for missing releases, npm dist-tag verification waits through slow registry CDN propagation with incremental backoff, and beta/stable release operations are role-enforced (#791).
+
+## [1.0.55-beta.3] - 2026-07-24
+
+This beta validates the HR Brain command surface, smoother guarded release
+automation, and deterministic Markdown test coverage on top of the
+`v1.0.55-beta.2` baseline.
+
+### Added
+
+- **HR Brain (`dws hrbrain`) command surface** — adds 11 commands across three groups: `talent-pool list/detail/employees` for talent pool browsing, `profile metadata/query/labels/career/performance` for employee profile data, and `search employees/employees-structured/fields` for basic and advanced (rule-based) people search. Ships with bundled mono/multi Skill guidance (`dingtalk-hrbrain`, `cli_version: ">=1.0.54"`); `search employees-structured` validates `--origin-json` as a JSON object and `--fields` as a JSON array before dispatch.
+
+### Changed
+
+- **Smoother guarded releases** — publishes verified stable and beta Homebrew Formula updates directly from the release workflow, retries transient tag-ref visibility failures, lets an exact same-run retry reuse its sealed tag, and allows machine-verified rebuild recovery without a separate approval wait.
+
+### Fixed
+
+- **Deterministic Markdown coverage** — replaces timing-dependent temporary-file deletion tests with synchronized file-stat failures so release admission no longer flakes on scheduler timing.
+
+### Changed
+
+- **Faster guarded releases** — trusts an independently revalidated, exact `CHANGELOG.md`-only successor of an already admitted `main` commit, runs cloud planning alongside governance, and executes sealed-release automation, compatibility, and multi-profile validation in parallel with artifact compilation. Normal cloud publication no longer requires an unshareable local packaging preflight.
+- **Scoped document reads and group mentions** — `doc read --content-format jsonml` can return `outline`, `range`, `section`, or custom-tag fragments with depth and block-boundary controls; document comment create, reply, and update can mention groups through `--mentioned-open-conversation-id`.
+- **Drive overwrite uploads** — `drive upload --node <fileId>` can replace an existing Drive or document-space file, is mutually exclusive with `--folder`, supports dry-run, and requires confirmation before writing.
+- **Chat nickname clearing and cross-organization todos** — omitting `--nick` from `chat group update-nick` now clears the current user's group nickname, while `todo task list --query-all` queries todos across organizations.
+
+### Fixed
+
+- **Legacy authentication compatibility** (#756) — migrates pre-v1.0.53 global and organization-scoped login state into the identity-aware token store, including all legacy organizations, while keeping unresolved accounts isolated from exact `corpId:userId` credentials so external or no-directory identities can complete login without borrowing another user's token.
+
+## [1.0.55-beta.1] - 2026-07-23
+
+This beta validates MCP Market URL resolution, the supported Wukong local-file
+send path after retiring the legacy credential-based media upload command from
+discovery, and reliable message-read rendering for rich content, forwarded
+records, encrypted messages, and media-download ID aliases.
+
+### Added
+
+- **MCP URL resolution** — adds `dws mcp url get <mcpId>` for resolving a DingTalk MCP Market ID to the current user and organization scoped Streamable HTTP URL, while keeping the helper-only `mcp-meta` endpoint out of the public product command surface.
+
+### Changed
+
+- **Chat local-file sending** — hides the open-source-only `chat media upload` compatibility command from Help, Schema, and bundled Skills, and removes its legacy AppKey/AppSecret OAPI path. Historical argv still receives an actionable migration error. Send local images and files through `chat message send --msg-type file --file-path`; callers that already hold a mediaId may continue to use `--msg-type image --media-id`.
+
+### Fixed
+
+- **Shortcut projection silent-empty returns** (#783) — a batch of read shortcuts returned an empty list with exit 0 and no error envelope even when the underlying MCP tool returned data, so agents misread "no data". The projection resolvers now probe the real container keys (`processCodeList`, `values`, `wikiSpaces`, `itemList`, `groupList`, `recentItems`, `emailAccounts`, `deptUserList`, `labelUserList`, `roles`, `report_list`, and the grouped `get_org_labels` `labels[]`), unwrap items nested under a VO wrapper (`shiftVO` / `entityVO` / `userInfo`), and `todo +created-todos` uses the shared pager (`pageSize=20`) because the backend silently returns an empty page for `pageSize>20`. Affects contact/oa/wiki/drive/minutes/calendar/attendance/chat/report/smart shortcuts, each with a guard test asserting the real response shape projects non-empty. `scripts/shortcut_real_result.py` also gains an upper-vs-lower layer comparison so an exit-0 empty projection over a non-empty backend is scored as `projection-data-loss` in the real read-audit path rather than `real-ok`.
+- **Message-read shortcut projection** (#706) — the message-list shortcuts (`chat +chat-messages` / `+messages-list` / `+messages-list-direct` / `+at-me` / `+search-msg` / `+thread-replies`) now render card and out-of-office rich-content JSON as readable text (without ever rewriting ordinary text that merely embeds a JSON fragment), expand a forwarded chat record's nested `forwardMessages` instead of collapsing to a "[卡片]" summary, and mark undecryptable encrypted card messages as `[加密消息]`; the speaker is read from the bare `sender` key, nested `{name:…}` sender objects yield their display name, and the literal string `"null"` is treated as absent. Shared projection helpers now live in `internal/shortcut/chatmsg`. `chat message download-media` also gains `--msg-id` / `--open-message-id` aliases for its `--message-id` flag so agents copying the `openMessageId`/`msgId` output field no longer hit "unknown flag".
+
+## [1.0.54] - 2026-07-21
+
+This release promotes the validated `v1.0.54-beta.2` baseline to stable. It restores the default transport envelope for personal event output with opt-in flattening, plus Schema CLI path and plugin overlay compatibility fixes.
+
+### Changed
+
+- **Personal event output compatibility** (#743) — `event consume` once again preserves the transport envelope by default for `ndjson`/`json`/`pretty`, while retaining the existing `compact` processor. New Agent workflows opt into the event-specific top-level DTO with `--flatten`, which is mutually exclusive with `-f raw` and `--debug-raw-events`; `event schema --flatten` describes that DTO, while the default schema describes `type/event_type/data/headers` and points to `.data | fromjson`.
+
+### Fixed
+
+- **Schema CLI path compatibility** (#738) — user-facing Schema lookups once again accept space-, dot-, and slash-separated CLI paths without weakening strict canonical identity resolution.
+- **Plugin CLI overlays** (#701) — installed plugins register their manifest-authored command trees again for HTTP and stdio servers, and a plugin may now replace a hidden compatibility fallback (for example `conference`) instead of being skipped as a distribution conflict.
+
+## [1.0.54-beta.2] - 2026-07-21
+
+This beta revalidates the same `v1.0.54-beta.1` source through the cloud release path with a sealed `OSS-Mirror: deferred` policy, because the manually tagged `v1.0.54-beta.1` push run failed on the unavailable OSS mirror channel after GitHub and npm delivery.
+
+### Changed
+
+- **Release delivery only** — no source changes since `v1.0.54-beta.1`; see that section for the user-visible changes under validation (#743, #738, #701).
+
+## [1.0.54-beta.1] - 2026-07-21
+
+This beta validates the restored default transport envelope for personal event output with opt-in flattening, plus Schema CLI path and plugin overlay compatibility fixes, on top of the validated `v1.0.53-beta.7` baseline.
+
+### Changed
+
+- **Personal event output compatibility** (#743) — `event consume` once again preserves the transport envelope by default for `ndjson`/`json`/`pretty`, while retaining the existing `compact` processor. New Agent workflows opt into the event-specific top-level DTO with `--flatten`, which is mutually exclusive with `-f raw` and `--debug-raw-events`; `event schema --flatten` describes that DTO, while the default schema describes `type/event_type/data/headers` and points to `.data | fromjson`.
+
+### Fixed
+
+- **Schema CLI path compatibility** (#738) — user-facing Schema lookups once again accept space-, dot-, and slash-separated CLI paths without weakening strict canonical identity resolution.
+- **Plugin CLI overlays** (#701) — installed plugins register their manifest-authored command trees again for HTTP and stdio servers, and a plugin may now replace a hidden compatibility fallback (for example `conference`) instead of being skipped as a distribution conflict.
+
+## [1.0.53] - 2026-07-21
+
+This release promotes the validated `v1.0.53-beta.7` baseline to stable. It adds enterprise onboarding, declarative shortcuts, Sheet/Aitable writes, multi-account profiles, and broader personal IM events, while hardening authentication and the guarded release path.
+
+### Added
+
+- **Enterprise and office command coverage** — adds enterprise creation, employee invitation, and account provisioning commands; 366 declarative service shortcuts; Sheet import commands; and Aitable workflow create/update support with reviewed Schema contracts.
+- **Multiple accounts in one DingTalk organization** — profiles can distinguish accounts by organization and user, select them explicitly, and log out one account or an entire organization without overwriting another account's credentials.
+- **Expanded personal IM event subscriptions** (#651) — adds read-receipt, recall, and reaction events for one-to-one and group chats, plus specified-sender subscriptions by staff ID or OpenDingTalk ID.
+- **Official multi-platform Homebrew channel** — ships separate stable and keg-only beta Formulae for macOS and Linux across amd64 and arm64, with isolated update PRs.
+
+### Changed
+
+- **Personal event output contract** (#651) — `event consume` now emits event-specific top-level structured fields; scripts that consumed the former transport envelope must use the flat fields or select `-f raw`, while `--debug-raw-events` retains the diagnostic envelope.
+- **Guarded release lifecycle** — beta/stable publication now uses explicit promotion, immutable delivery proofs, protected recovery, and tag-bound optional OSS policy; an unprovisioned OSS mirror is sealed as `deferred` so GitHub, npm, and Homebrew are not blocked.
+- **Relaxed stable promotion contract** (#729) — a stable release still requires a delivered, non-withdrawn beta baseline in its commit history, but no longer requires a byte-identical tree with that beta; reviewed commits merged to `main` after the beta can now ship in the stable release. Local releases now accept any sealed commit contained in `main` history and push only the release tag, so `main` is never frozen during the beta-to-stable window.
+
+### Fixed
+
+- **Authentication and credential reliability** — organization-policy denials stop before mutation or polling, long-running clients reload and refresh access tokens consistently, concurrent credential writes are atomic, and Windows portable-auth commands fail before reading or writing unsupported credential bundles.
+- **Command validation and compatibility** — invalid Sheet/task targets fail locally, IM shortcuts preserve AI-tag and alias compatibility, and Aitable import uploads require and forward a positive file size.
+- **Release publication reliability** — GitHub draft publication is bound to one verified release ID and exact assets, preflight uses isolated installer worktrees, guarded local tags remain compatible, cloud planning fingerprints the actual allocated release refs, and npm channel verification waits for bounded registry propagation without moving tags.
+- **Package-manager version verification** (#735) — npm-vendored, Homebrew-installed, and packaged release binaries are now verified by searching their raw bytes for the injected version marker, so a correctly versioned stable binary is no longer rejected when the short version marker coalesces with adjacent printable linker metadata; incorrect or missing markers still fail closed.
+
+## [1.0.53-beta.7] - 2026-07-21
+
+This beta validates bounded npm channel verification after registry publication.
+
+### Fixed
+
+- **npm dist-tag eventual consistency** — Release delivery now tolerates a briefly stale `latest` or `beta` read after publishing by retrying only when npm reports a valid older version. Registry errors, invalid or incomparable tags, and channels that never converge still fail closed without moving any tag during verification.
+
+## [1.0.53-beta.6] - 2026-07-21
+
+This beta validates guarded local release compatibility and tag-bound OSS deferral so an unprovisioned mirror cannot block the primary release channels.
+
+### Changed
+
+- **Tag-bound optional OSS release mirror** — Official cloud Release runs no longer block GitHub, npm, and Homebrew delivery when an OSS bucket has not been provisioned. Cloud tags immutably record `OSS-Mirror: enabled|deferred`; publication, repair, and withdrawal consume that sealed policy instead of the current repository variable. Enabled releases remain fail-closed, while deferred releases skip the nonexistent channel and cannot be backfilled without a future audited repair proof.
+
+### Fixed
+
+- **Guarded local release compatibility** — The tag-push Release workflow now accepts the `Channel`-only annotated tags created by the guarded local release entry while continuing to reject any partial cloud-only seal metadata.
+- **Cloud release tag allocation fingerprint** — Release planning now fingerprints the actual `v*` and `withdrawn/v*` refs fetched from GitHub, matching the seal job's API view instead of hashing an empty non-wildcard ref prefix and rejecting every publish before tag creation.
+
+## [1.0.53-beta.5] - 2026-07-21
+
+This beta validates long-running access-token recovery and the faster, recoverable guarded release path introduced after v1.0.53-beta.4.
+
+### Changed
+
+- **Fast guarded beta and stable releases** — successful local release checks now leave a six-hour proof bound to the exact version, commit, repository identity, remote `main`, and stable baseline, so the subsequent guarded `--publish` invocation revalidates authority without repeating tests and packaging. A default-branch governance smoke uses the same dedicated immutable-release credential as the tag workflow before any tag is allocated.
+- **Protected existing-tag recovery** — `dws-release recover <version>` can resume a failed, unpublished annotated tag through the normal contract, build, Developer ID signing, immutable GitHub Release, Homebrew, npm, and OSS jobs. Recovery requires the exact tag object, peeled commit, failed tag-push run, typed version confirmation, and the protected `release-recovery` environment; successful runs are accepted as future beta/stable delivery evidence.
+
+### Fixed
+
+- **Long-running event authentication recovery** — personal and portal event streams resolve the current access token for every ticket request, refresh a server-rejected token with compare-and-refresh semantics, and reconnect with backoff when refresh is temporarily blocked by network failures, rate limits, or 5xx responses.
+- **Consistent access-token caching and errors** — runtime, recovery, Skill, PAT polling, and personal/portal event clients now resolve user access tokens through one expiry- and publication-aware manager, so long-running processes reload rotated credentials while keychain, refresh, parse, permission, and cancellation failures remain observable instead of being collapsed into “not authenticated.”
+- **Tag-push GitHub Release publication** — Draft publication now locks one GitHub Release database ID, verifies its exact tag, channel, notes, recovery marker, asset set, and uploaded bytes, then publishes and rechecks that same ID as immutable. Recovery runs use the trusted default-branch release helpers instead of the sealed tag's historical scripts, fixing the Draft-only `GET /releases/tags/{tag}` 404 without allowing the release identity to drift during recovery.
+- **Release preflight reliability** — source-mode installer tests now use isolated temporary checkouts and HOME directories instead of overwriting and deleting the real repository `dws` binary, release preflight explicitly rebuilds before policy checks, and the full-suite runner gives the growing script package a non-flaky five-minute per-suite budget.
+
+## [1.0.53-beta.4] - 2026-07-17
+
+This beta validates the expanded personal IM event subscriptions and the flattened `event consume` structured output introduced after v1.0.53-beta.3.
+
+### Added
+
+- **Expanded personal IM event subscriptions** (#651) — adds one-to-one and group events for message read receipts, recalls, and reactions; publishes the specified-sender receive event; and lets one-to-one/sender subscriptions target either a staff `--user` or an `--open-dingtalk-id`. Event Schema now exposes these alternatives through machine-readable parameter constraints.
+
+### Changed
+
+- **Personal event structured output is now flat** (#651) — `event consume` projects NDJSON/JSON/pretty/compact output into event-specific top-level DTOs, so consumers read fields such as `content`, `sender`, and `conversation_id` directly instead of parsing `.data | fromjson`. This is a breaking change for scripts using the former transport envelope; the original server payload remains available through `-f raw`, while `--debug-raw-events` preserves the full diagnostic envelope.
+
+## [1.0.53-beta.3] - 2026-07-17
+
+This beta validates multi-account profile support and the post-v1.0.53-beta.2 compatibility fixes for Windows portable authentication, IM shortcuts, and Aitable import uploads.
+
+### Added
+
+- **Multiple accounts in one DingTalk organization** — profiles are keyed by `corpId:userId`, `--profile` accepts organization IDs/names plus user IDs/names, and organization-only selection uses its explicitly remembered current account or asks for an exact account when ambiguous.
+
+### Changed
+
+- **Profile-scoped logout and consistent token storage** — `dws auth logout --profile` can remove one account or every account in an organization, while identity token slots remain the source of truth and legacy organization/global mirrors stay compatible without overwriting newer account credentials.
+
+### Fixed
+
+- **Windows portable-auth contract** — `dws auth export` and `dws auth import` now fail early without reading credentials, bundles, or writing files instead of claiming portable-bundle support for DPAPI-protected HKCU Registry credentials.
+- **IM shortcut message tags and compatibility aliases** (#646) — IM send shortcuts now add the same AI-sent marker as `chat message send` by default, support `--ai-tag=false` to opt out, and preserve compatible search, conversation-ID, and page-size aliases.
+- **Aitable import upload file-size validation** (#654) — `dws aitable import upload` and `dws aitable +import-upload` now require a positive `--file-size` and always send it to the upload-preparation API, preventing invalid requests without the actual file size.
+
+## [1.0.53-beta.2] - 2026-07-16
+
+This beta validates the accumulated post-v1.0.52 command surface, release automation, and runtime hardening changes, including enterprise contact onboarding, declarative shortcuts, Sheet/Aitable writes, multi-platform Homebrew formulas, and credential and target-validation fixes.
+
+### Added
+
+- **Contact enterprise onboarding commands** — adds `contact org create`, `contact user invite`, and `contact account create` for creating a DingTalk enterprise, inviting an employee by mobile, and provisioning an enterprise login account, with reviewed Schema contracts and mono/multi Skill routing.
+- **Declarative shortcut commands** (#592) — adds 366 `dws <service> +<command>` shortcuts across 16 services, including one-to-one MCP wrappers and multi-step smart workflows. Shortcuts publish stable Agent-visible contracts with named flags, validation and confirmation metadata, dry-run protection for writes, catalog/help routing, and optional local YAML extensions and usage recording.
+- **Sheet imports and Aitable workflow writes** (#624) — adds `dws sheet import` / `sheet import create` for converting local xlsx/xls files into new online sheets, `sheet import get` for polling import tasks, and `dws aitable workflow create/update` for applying validated `workflow-dsl/v1` definitions, with matching reviewed Agent Schema and bundled Skill guidance.
+- **Official multi-platform Homebrew channel** — stable `Formula/dingtalk-workspace-cli.rb` and keg-only `Formula/dingtalk-workspace-cli-beta.rb` live in this repository and select signed macOS Intel/Apple Silicon or Linux amd64/arm64 artifacts at install time. Stable and beta releases open isolated Formula update PRs after final artifact signing, so beta never replaces the stable Formula. Agent Skills stay under `pkgshare` without mutating the user's home directory, and both tracks are covered by the six-channel post-release verifier.
+
+### Changed
+
+- **Guarded prerelease and stable automation** — adds the guided `dws-release` entry for one-command CHANGELOG preparation, validation-only and annotated-tag publication flows; promotes only an explicitly validated beta; verifies command-tree compatibility and all six packaged binaries; and serializes immutable GitHub Release, npm channel, OSS, Homebrew, and optional Gitee delivery with fail-closed recovery checks.
+- **Reviewed historical release recovery proofs** — release preflight can recognize an explicitly pinned successful recovery delivery for a historical stable tag while still rejecting arbitrary workflow dispatches, mismatched commits, and incomplete release, signing, or publication jobs.
+
+### Fixed
+
+- **PAT organization-policy denials stop immediately** — `PAT_ORG_POLICY_DENIED` now remains terminal even if a backend also returns `flowId`, authorization URLs, or client credentials; the CLI does not mutate process credentials, open a browser, poll, or retry until an organization administrator changes the policy.
+- **Sheet and task invalid-target failures** — `sheet range read/get` now rejects a null cell-info response instead of printing `null` and exiting successfully, while task completion and attachment listing verify that a task exists before calling lenient backend endpoints. Attachment listing is also published through Runtime Schema for schema-first Agent discovery.
+- **Concurrent credential writes and reentrant CLI execution** — secure-token writers now use isolated, exclusive temporary files before atomic replacement so concurrent processes cannot remove each other's in-flight data, and repeated in-process CLI runs close the previous file logger before replacing it instead of retaining the prior log-file handle.
+
+## [1.0.52] - 2026-07-14
+
+This release seals the `v1.0.52` line with personal event subscriptions, a deterministic 22-product Agent command catalog, local user-operation auditing, expanded Open product commands, safer macOS credentials and release signing, and more reliable Connect and IM delivery.
+
+### Added
+
+- **Personal event subscriptions** (#589) — adds `dws event list/schema/consume/status/stop` for user @ mentions, selected one-to-one chats, and selected group chats. `consume` can create or reuse a personal subscription, multiple local consumers share one bus while keeping outputs isolated by event type and subscription, and the mono/multi event Skills ship with the binary.
+- **Open product command capabilities** (#608) — adds Sheet table, pivot-table, and gridline commands; Chat message favorites; Drive statistics and shortcuts; and Doc comment update/delete, with matching mono/multi Skill documentation and command-contract coverage.
+- **Local user-operation audit log** (#555) — operations executed through `dws` now produce redacted daily JSONL records with actor, command and endpoint, result or error category, duration, CLI/platform metadata, and a SHA-256 previous-hash chain for tamper evidence. Writers coordinate through a cross-process file lock and rotate logs safely; `dws audit tail` inspects recent records, `dws audit export` emits date-filtered JSONL or CSV, and `dws audit verify` reports the first broken link in a file's hash chain.
+- **Stable Agent command catalog** (#598) — `dws schema` now ships a deterministic 22-product / 564-tool catalog generated from the executable Cobra tree, with progressive product/group/leaf queries, complete parameter contracts, reviewed command identity and aliases, safety/confirmation metadata, field provenance, and final-delivery completeness/drift gates. The catalog is embedded at build time and does not require runtime MCP `tools/list` discovery.
+- **Reviewed Schema for local commands** (#598, #609) — `event consume/list/schema/status/stop` and `audit export/tail/verify` enter the reviewed `CommandRegistry`, bind to the real Cobra tree at generation time, and ship through the same typed `ToolSpec` and embedded Catalog path as public MCP-backed commands. Leaf, group, product, and `--all` queries are projections of that single delivered model.
+- **Safe macOS Keychain → file-DEK migration** (#597) — `dws auth migrate-keychain --to file-dek` preflights every legacy/profile auth entry before rewriting, ignores unrelated application secrets, supports side-effect-free `--dry-run`, requires explicit `--yes`, and lets sandboxed and normal processes share an existing login without exposing tokens.
+
+### Changed
+
+- **`event consume` AI-subprocess contract** (#609) — emits a fixed ready line and a final controlled-exit summary, supports parent-pipe stdin EOF as graceful shutdown, forwards `--profile` to the detached bus, surfaces bus startup errors, and cleans up subscriptions according to ownership so orchestrators can drive event streams without sleeps or leaked server-side subscriptions.
+- **Wukong IM read-result parity** (#618) — `chat message list` preserves quoted merged-forward and image context; message-search entitlement failures retain the server-provided friendly hint and action URL; and `ding message list` exposes each DING's content alongside its ID and status.
+- **Developer ID signing for official macOS archives** (#605) — official releases now require both Darwin archives to be signed with the configured Apple Developer ID certificate, timestamp, and hardened runtime. The release job validates credentials and signatures and fails closed instead of silently publishing ad-hoc-signed official binaries.
+
+### Fixed
+
+- **Smart-category mappings and runtime network diagnostics** (#591) — `chat category create-smart` now maps category names, group-name keywords, and member OpenDingTalk IDs to the live MCP contract, rejects blank or empty supplied values locally, and reports runtime `tools/call` connection failures as actionable API/network errors instead of internal discovery failures.
+- **Connect daemon restart lifecycle** (#599) — pins the Stream SDK reconnect-race fix, snapshots the running executable before detaching, uses a real 30-second keepalive, and manages each worker as its own Unix process group so launcher cleanup or worker panics no longer cause restart loops or orphan local-agent processes.
+- **Complex Connect messages and attachments** (#606, #612) — rich-text messages retain all embedded pictures in order, queued turns keep every pending attachment, and unknown or future callback shapes reach each Agent backend with their message type and raw JSON instead of being discarded. Attachment recovery is locator-based, nested `chatRecord` pictures/audio/video/files can be recovered from message APIs after Stream ACK, and OpenCode uses a full-duration storyboard for large videos to avoid base64 OOMs while preserving the original download for the turn.
+- **macOS auth survives Keychain mode changes** (#597) — credential reads try existing compatible DEKs without creating key material, updates preserve the DEK that decrypted existing ciphertext, unreadable slots fail closed before token exchange, profile slots use the canonical auth backend, and `auth status` reports ciphertext/key mismatches instead of treating them as ordinary logout. Dedicated macOS race and Windows DPAPI coverage protect the cross-platform paths.
+
+## [1.0.51] - 2026-07-10
+
+This release promotes the sealed `v1.0.51-beta.1` contents to stable. It syncs the hardcoded Wukong command surface, prevents `dev connect` conversations from blocking on messages received mid-turn, and makes local credential failures diagnosable without mutating key material.
+
+### Added
+
+- **Agoal product commands** (#585) — adds `dws agoal` strategy, contract, scorecard, user-objective, report, and objective-template command groups, together with static routing and the bundled mono/multi Agoal skills.
+- **Wukong chat command parity** (#585) — adds `chat group notice create|edit|get|list`, `group share-invite`, `text translate`, `category create-smart`, and `message list-emotion-replies`.
+- **Wukong document import commands** (#585) — adds `doc import` for starting imports and `doc import get` for querying import tasks.
+- **Wukong mail command parity** (#585) — adds mailbox profile, message batch-get, sent-message recall and recall-detail, auto-reply update, plus allow-list and block-list management.
+- **Wukong sheet grouping commands** (#585) — adds `sheet group-dimension` and `sheet ungroup-dimension` for whole-row or whole-column ranges.
+- **Keychain health diagnostics** (#578) — `dws doctor` now includes a keychain check, while `dws auth status` distinguishes ordinary logged-out state from `keychain_unavailable` and `dek_missing` failures and returns remediation hints in table and JSON output.
+
 ### Changed
 
 - **`dws pat chmod` defaults to permanent grants** (#584) — running `dws pat chmod <scope>` without `--grant-type` now requests a `permanent` grant instead of `session`, aligning the direct CLI path with the recommend-authorization helper. Session grants remain available by passing `--grant-type session --session-id <id>`.
+- **The `dev connect --channel gemini` path now uses the Gemini `generateContent` API** (#587) — configure it with `GEMINI_API_KEY` or `GOOGLE_API_KEY`, optionally override the compatible endpoint with `GEMINI_API_BASE_URL` or `GOOGLE_GEMINI_API_BASE_URL`, and select a model with `--agent-model` or `GEMINI_MODEL`; a local `gemini` executable is no longer required.
+
+### Fixed
+
+- **Non-blocking `dev connect` turn scheduling** (#587) — stream and `@`-poll callbacks no longer wait for the active turn to finish. Turns stay serialized per conversation, messages received mid-turn are coalesced into one pending follow-up, and different conversations can continue in parallel.
+- **Connect agent recovery and headless execution** (#587) — stale addressable sessions retry once with a fresh session, unsupported Qoder control requests receive an immediate response instead of hanging, OpenCode and bypass-mode channels receive non-interactive permission settings, and backend/API failures are no longer posted as successful assistant replies.
+- **Side-effect-free credential reads** (#578) — keychain reads inspect encrypted credential data before looking up the DEK and never generate a replacement key on a read path. Missing DEKs and unavailable macOS Keychains are surfaced as explicit diagnostic failures instead of silently mutating credential state.
 
 ## [1.0.50] - 2026-07-08
 
@@ -62,7 +778,6 @@ This release promotes the sealed **remove-discovery delivery** from the beta lin
 
 - **Command-surface regression tests** — root-command tests now cover real `contact label`/`role` dry-runs, hidden top-level contact compatibility entries, `chat file upload` downline behavior, and `calendar event list --dry-run`.
 - **Release hygiene tests** — skill markdown policy still blocks unsupported conference routes, plugin loader tests assert optional validation failures stay quiet at WARN level, and doc version cursor extraction has nested-envelope coverage.
-
 ## [1.0.47] - 2026-07-05
 
 This release adds **connector supervision & health monitoring** (`dev connect list/status/restart/stop`) and fixes **bot-to-bot @-mention** delivery end-to-end.

CONTRIBUTING.md

@@ -27,7 +27,9 @@ notes that are intentionally kept out of the repository root.
 
 ## Local Checks
 
-Run the verification commands that match the surface you changed before you hand work back.
+Run the verification commands that match the surface you changed before you
+hand work back. The goal is useful, change-specific evidence, not a second
+local execution of every CI job.
 
 Common repository checks already used here include:
 
@@ -36,27 +38,71 @@ Common repository checks already used here include:
 ./scripts/policy/check-open-source-assets.sh
 go test ./...
 make test
+make test-plan
 make lint
-bash test/scripts/run_all_tests.sh --jobs 8
 ./scripts/policy/check-generated-drift.sh
 ./scripts/policy/check-command-surface.sh --strict
 ./scripts/release/verify-package-managers.sh
 git diff --check
 ```
 
+Select the PR risk tier before choosing checks:
+
+| Tier | Typical scope | Developer evidence | CI expansion |
+|---|---|---|---|
+| Documentation-only | Prose and documentation assets with no executable, generated, workflow, packaging, or interface change | Links/content/rendering plus repository asset checks | Lightweight documentation validation; all nine named contexts still report |
+| Standard | Ordinary implementation work with a stable package graph | Focused unit/integration tests and observable behavior for the changed path | Race tests for changed packages and their reverse dependencies, scope-matched HEAD/base coverage, and representative Darwin/Windows compilation |
+| High-risk | Workflow/policy, package graph, generated Schema/registry, platform, auth/keychain, installer, packaging, release, transport, recovery, or an unprovable infrastructure change | Relevant full or domain suite plus focused behavior evidence | Complete race suite, native platform tests, and all affected domain gates; protected `main` uses this tier |
+
+Classification fails closed: an incomplete diff, package add/remove/rename, or
+uncertain dependency graph selects the high-risk suite. Native changed-code
+coverage is additionally selected for platform-sensitive code.
+
 ## Pull Request Checklist
 
 1. Keep implementation and tests in sync.
-2. Run `./scripts/dev/ci-local.sh`.
-3. Run `./scripts/policy/check-command-surface.sh --strict` when command paths/flags change.
-4. Run `./scripts/policy/check-generated-drift.sh` when generated artifacts may change.
-5. Run `./scripts/release/verify-package-managers.sh` when packaging or installer surfaces change (run `make package` first).
-6. Update docs and `CHANGELOG.md` for behavior/interface changes.
-7. Include verification evidence in your PR description.
+2. Select the documentation-only, standard, or high-risk tier and run the
+   smallest checks that prove the change. Use `./scripts/dev/ci-local.sh` when
+   a complete local pass is warranted; it is not required for every ordinary
+   PR.
+3. Include both the commands/results and user-visible or contract-level
+   behavior evidence in the PR description.
+4. Run `./scripts/policy/check-command-surface.sh --strict` when command
+   paths/flags change. CI resolves the exact merge-base, latest reachable
+   non-withdrawn stable GA tag, and committed candidate SHA, then enters the single compatibility
+   decision seam through
+   `make authoritative-interface-integrity BASE_REF=<merge-base> STABLE_REF=<latest-GA-tag> CANDIDATE_REF=<candidate-sha>`.
+   The Make target delegates to the authoritative wrapper; CI does not invoke a
+   second comparator or the legacy fixture checker. See
+   [CLI flag compatibility migration governance](docs/cli-interface-flag-migrations.md)
+   for the reviewed two-stage `pending` → `consumed` lifecycle.
+   Agent-visible flag migrations must also run
+   `make schema-compatibility BASE_REF=<merge-base> STABLE_REF=<latest-GA-tag> CANDIDATE_REF=<candidate-sha>`;
+   it consumes the same base-owned ledger rather than a second exception list.
+5. Run `./scripts/policy/check-generated-drift.sh` when generated artifacts may
+   change.
+6. Run `./scripts/release/verify-package-managers.sh` when packaging or
+   installer surfaces change (run `make package` first).
+7. Update docs and `CHANGELOG.md` for behavior/interface changes.
 
 ## Submission Flow
 
 1. Make the smallest atomic change that satisfies the task.
 2. Keep doc edits factual and limited to implemented behavior.
 3. Run the relevant verification commands.
-4. Report the validation results with the handoff.
+4. Report the validation results and risk tier with the handoff.
+5. Open a ready PR against `main`. Base-owned automation assigns one eligible
+   peer reviewer, balancing the current open-review load and excluding the
+   author. A new head push re-enters the same routing flow when the latest
+   revision still needs review.
+6. After the latest push has one peer approval and the exact nine required
+   contexts are current and green, auto-merge completes the PR. If `main`
+   advances first, strict status checks revalidate the branch; no separate
+   routine merge request is needed.
+
+Contributors without repository write access stop at the PR flow. Explicitly
+authorized collaborators with `write`, `maintain`, or `admin` access can use
+[Actions → Release](https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/actions/workflows/release.yml)
+to publish beta releases without manual approval. The same internal roles may
+start a stable release, but a different repository administrator must approve
+the `release-stable` Environment deployment before publication continues.

Formula/dingtalk-workspace-cli-beta.rb

@@ -0,0 +1,63 @@
+class DingtalkWorkspaceCliBeta < Formula
+  desc "Automate DingTalk workspace tasks from the terminal (beta channel)"
+  homepage "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli"
+  version "1.0.58-beta.3"
+  license "Apache-2.0"
+  keg_only "it is the beta channel and conflicts with dingtalk-workspace-cli"
+
+  on_macos do
+    if Hardware::CPU.arm?
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.58-beta.3/dws-darwin-arm64.tar.gz"
+      sha256 "29b4fb9e081f36a699933c0919fe7530544f62de3e27c785d27626a575a2efc2"
+    else
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.58-beta.3/dws-darwin-amd64.tar.gz"
+      sha256 "a6b9c4ef212c533e02b414bd9c3be0b8d1874d4af983a7896072825bdfec1033"
+    end
+  end
+
+  on_linux do
+    if Hardware::CPU.arm?
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.58-beta.3/dws-linux-arm64.tar.gz"
+      sha256 "ae3a9ebe151702fd05dee0c56d778a20789d98c390cf8c0a0b2ec695b57b5ec3"
+    else
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.58-beta.3/dws-linux-amd64.tar.gz"
+      sha256 "544b480701e9ec9ec5366467ad4c855fca06dea887e3d577ff50e4b3eeb13ac2"
+    end
+  end
+
+  resource "skills" do
+    url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.58-beta.3/dws-skills.zip"
+    sha256 "322f1840442ff183ad4b6d4f2a2b38825ff9f96a3fcde06ba57b8f80647468ae"
+  end
+
+  def install
+    root = Dir["dws-*"].find { |entry| File.directory?(entry) } || "."
+    binary = File.join(root, "dws")
+    raise "binary not found: #{binary}" unless File.exist?(binary)
+
+    bin.install binary => "dws"
+
+    %w[LICENSE NOTICE README.md CHANGELOG.md].each do |name|
+      source = File.join(root, name)
+      pkgshare.install source if File.exist?(source)
+    end
+
+    skill_dest = pkgshare/"skills/dws"
+    skill_dest.mkpath
+    resource("skills").stage do
+      cp_r(Dir["*"], skill_dest)
+    end
+  end
+
+  def caveats
+    <<~EOS
+      Agent Skills are bundled in #{pkgshare}/skills/dws.
+      Run `dws skill setup` to install them into your Agent directories.
+      This beta is keg-only. Add #{opt_bin} to PATH to use its `dws` binary.
+    EOS
+  end
+
+  test do
+    assert_match version.to_s, shell_output("#{bin}/dws version")
+  end
+end

Formula/dingtalk-workspace-cli.rb

@@ -0,0 +1,63 @@
+class DingtalkWorkspaceCli < Formula
+  desc "Automate DingTalk workspace tasks from the terminal"
+  homepage "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli"
+  version "1.0.57"
+  license "Apache-2.0"
+
+
+  on_macos do
+    if Hardware::CPU.arm?
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.57/dws-darwin-arm64.tar.gz"
+      sha256 "c01c28dc13948a70fca905207073dc8dbd22f7ba7fc90e68b3316eb9a9c98e88"
+    else
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.57/dws-darwin-amd64.tar.gz"
+      sha256 "d7baa218beefc851c6a933b456055195f8272984ce008d7e0122bdfc5dad94ea"
+    end
+  end
+
+  on_linux do
+    if Hardware::CPU.arm?
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.57/dws-linux-arm64.tar.gz"
+      sha256 "0bbe9c233a3ff585077bae1ac5000937c32d967846d14cc44c46f98d49b95ae2"
+    else
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.57/dws-linux-amd64.tar.gz"
+      sha256 "f113ce3654f21d1f9ecc7c196f815aeafbca54d377a347b244a15116c5cba698"
+    end
+  end
+
+  resource "skills" do
+    url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.57/dws-skills.zip"
+    sha256 "0c9667209cf30761427a8f9348149cbbf1e397aa3c25587e99f205bc7525e101"
+  end
+
+  def install
+    root = Dir["dws-*"].find { |entry| File.directory?(entry) } || "."
+    binary = File.join(root, "dws")
+    raise "binary not found: #{binary}" unless File.exist?(binary)
+
+    bin.install binary => "dws"
+
+    %w[LICENSE NOTICE README.md CHANGELOG.md].each do |name|
+      source = File.join(root, name)
+      pkgshare.install source if File.exist?(source)
+    end
+
+    skill_dest = pkgshare/"skills/dws"
+    skill_dest.mkpath
+    resource("skills").stage do
+      cp_r(Dir["*"], skill_dest)
+    end
+  end
+
+  def caveats
+    <<~EOS
+      Agent Skills are bundled in #{pkgshare}/skills/dws.
+      Run `dws skill setup` to install them into your Agent directories.
+
+    EOS
+  end
+
+  test do
+    assert_match version.to_s, shell_output("#{bin}/dws version")
+  end
+end

Makefile

@@ -1,6 +1,16 @@
 GO ?= go
+DWS_PACKAGE_VERSION ?= 0.0.0-test
+REMOTE ?=
+PUBLISH ?= 0
+YES ?= 0
+DWS_POLICY_TMPDIR ?= $(CURDIR)/.worktrees/policy-tmp
+POLICY_GOTMPDIR ?= $(DWS_POLICY_TMPDIR)/go
+SCHEMA_CATALOG_OUTPUT ?= artifacts/schema_catalog
+SCHEMA_META_INDEX_OUTPUT ?= artifacts/schema_meta_index.gob
+POLICY_ENV = DWS_POLICY_TMPDIR="$(DWS_POLICY_TMPDIR)" GOTMPDIR="$(POLICY_GOTMPDIR)"
+GO_SOURCE_LIST = git ls-files -z --cached --others --exclude-standard -- '*.go'
 
-.PHONY: all help build rebuild test lint fmt policy edition-test package release publish-homebrew-formula setup-hooks
+.PHONY: all help build rebuild test test-plan test-auth-legacy-compat lint format-check fmt policy edition-test interface-integrity authoritative-interface-integrity coverage-gate coverage-gate-platform update-interface-baseline reset-interface-baseline schema-compatibility skill-command-integrity skill-context-budget multi-im-skill-chain-integrity cli-smoke mock-mcp-smoke test-schema-agent-examples generate-schema fetch-mcp-metadata generate-schema-catalog package release release-pre release-stable changelog-pre changelog-stable publish-homebrew-formula setup-hooks
 
 all: setup-hooks fmt lint build test rebuild
 
@@ -8,11 +18,32 @@ help:
 	@printf "Available targets:\n"
 	@printf "  make build         - Build the dws CLI binary\n"
 	@printf "  make test          - Run the Go test suite\n"
-	@printf "  make lint          - Run formatting checks and golangci-lint when available\n"
-	@printf "  make fmt           - Format Go source files\n"
-	@printf "  make policy        - Run open-source asset and command-surface checks\n"
-	@printf "  make package       - Build all release artifacts locally (goreleaser snapshot)\n"
-	@printf "  make release       - Build and publish a release via goreleaser\n"
+	@printf "  make test-plan     - Verify every default Go package belongs to one CI test shard\n"
+	@printf "  make test-auth-legacy-compat - Run stable legacy authentication compatibility regressions\n"
+	@printf "  make lint          - Run formatting checks, go vet, and staticcheck\n"
+	@printf "  make format-check  - Check all repository Go source files with gofmt\n"
+	@printf "  make fmt           - Format all repository Go source files\n"
+	@printf "  make policy        - Check the built dws plus open-source and Schema policies\n"
+	@printf "  make interface-integrity [BASE_REF=<ref>] [STABLE_REF=<tag>] [CANDIDATE_REF=<ref>] - Check authoritative CLI history\n"
+	@printf "  make authoritative-interface-integrity BASE_REF=<ref> [STABLE_REF=<tag>] [CANDIDATE_REF=<ref>] - Check Git-owned CLI history\n"
+	@printf "  make coverage-gate BASE_REF=<ref> - Enforce overall non-regression and 100%% changed-code coverage\n"
+	@printf "  make coverage-gate-platform BASE_REF=<ref> PROFILE=<file> - Enforce 100%% native changed-code coverage\n"
+	@printf "  make update-interface-baseline - Update the non-authoritative CLI smoke fixture\n"
+	@printf "  make reset-interface-baseline - DANGEROUS: replace the non-authoritative CLI smoke fixture\n"
+	@printf "  make schema-compatibility BASE_REF=<ref> [STABLE_REF=<tag>] [CANDIDATE_REF=<ref>] - Check the authoritative Schema history\n"
+	@printf "  make skill-command-integrity - Check dws commands referenced by skills exist\n"
+	@printf "  make skill-context-budget - Check generated Skill drift and common-path context budgets\n"
+	@printf "  make multi-im-skill-chain-integrity - Check reviewed IM intents keep one default Skill route\n"
+	@printf "  make cli-smoke     - Verify help for every public top-level command\n"
+	@printf "  make mock-mcp-smoke - Verify HTTP and stdio MCP request/response transport\n"
+	@printf "  make test-schema-agent-examples - Contract-check all Agent examples and dry-run the eligible subset\n"
+	@printf "  make generate-schema - Refresh param_aliases + verify Schema assembly determinism\n"
+	@printf "  make generate-schema-catalog - Optional assembled Catalog dump under artifacts/ (not a delivery step)\n"
+	@printf "  make package       - Build all release artifacts locally\n"
+	@printf "  make changelog-pre VERSION=vX.Y.Z-beta.N - Prepare prerelease notes\n"
+	@printf "  make changelog-stable VERSION=vX.Y.Z FROM_BETA=vX.Y.Z-beta.N - Prepare stable notes\n"
+	@printf "  make release-pre VERSION=vX.Y.Z-beta.N - Validate prerelease; publish official releases from Actions\n"
+	@printf "  make release-stable VERSION=vX.Y.Z FROM_BETA=vX.Y.Z-beta.N - Validate stable; publish official releases from Actions\n"
 	@printf "  make publish-homebrew-formula - Push dist/homebrew/dingtalk-workspace-cli.rb to a tap repo\n"
 
 build:
@@ -22,24 +53,178 @@ rebuild:
 	@./scripts/dev/build.sh
 
 test:
-	@./test/scripts/run_all_tests.sh
+	@DWS_PACKAGE_VERSION="$(DWS_PACKAGE_VERSION)" $(GO) test -count=1 -timeout=10m ./...
+
+test-plan:
+	@./scripts/ci/test-packages.sh verify
+
+test-auth-legacy-compat:
+	@mkdir -p "$(POLICY_GOTMPDIR)"
+	@GO="$(GO)" $(POLICY_ENV) ./scripts/policy/check-auth-legacy-compat.sh
 
 lint:
 	@./scripts/dev/lint.sh
 
-fmt:
-	@find cmd internal test -name '*.go' -print0 2>/dev/null | xargs -0r gofmt -w
+format-check:
+	@set -eu; \
+	go_files="$$(mktemp "$${TMPDIR:-/tmp}/dws-go-files.XXXXXX")"; \
+	trap 'rm -f "$$go_files"' EXIT HUP INT TERM; \
+	$(GO_SOURCE_LIST) > "$$go_files"; \
+	unformatted="$$(xargs -0 sh -c 'if [ "$$#" -gt 0 ]; then exec gofmt -l -- "$$@"; fi' sh < "$$go_files")"; \
+	if [ -n "$$unformatted" ]; then \
+		printf '%s\n' "$$unformatted"; \
+		printf '%s\n' "Go files are not formatted. Run 'make fmt'." >&2; \
+		exit 1; \
+	fi
 
-policy:
-	@./scripts/policy/check-open-source-assets.sh
-	@./scripts/policy/check-command-surface.sh --strict
+fmt:
+	@set -eu; \
+	go_files="$$(mktemp "$${TMPDIR:-/tmp}/dws-go-files.XXXXXX")"; \
+	trap 'rm -f "$$go_files"' EXIT HUP INT TERM; \
+	$(GO_SOURCE_LIST) > "$$go_files"; \
+	xargs -0 sh -c 'if [ "$$#" -gt 0 ]; then exec gofmt -w -- "$$@"; fi' sh < "$$go_files"
+
+policy: test-auth-legacy-compat
+	@mkdir -p "$(POLICY_GOTMPDIR)"
+	@$(POLICY_ENV) ./scripts/policy/check-open-source-assets.sh
+	@$(POLICY_ENV) ./scripts/policy/check-skill-context-budget.sh
+	@$(POLICY_ENV) ./scripts/policy/check-multi-im-skill-chain.sh
+	@python3 scripts/run_chat_shortcut_live_audit_test.py
+	@$(POLICY_ENV) ./scripts/policy/check-command-surface.sh --strict
+	@$(POLICY_ENV) ./scripts/policy/check-generated-drift.sh
+	@$(POLICY_ENV) ./scripts/policy/check-param-concepts.sh
+	@$(POLICY_ENV) ./scripts/policy/check-param-alias-cooccurrence.sh
+	@$(POLICY_ENV) $(GO) test -count=1 ./internal/app -run '^(TestParamAlias(FixtureThroughEmbeddedDeliveryPath|ReadCommandFinalPayload|WriteCommandFinalPayload|CanonicalConflictFailsBeforeRunE|BlockedFlagReachesReviewedFinalError)|TestFlagConflictErrorFormattingIsDeterministic)$$'
+	@$(POLICY_ENV) ./scripts/policy/check-schema-catalog.sh
+	@$(POLICY_ENV) ./scripts/policy/check-schema-binary.sh
+	@$(POLICY_ENV) $(MAKE) test-schema-agent-examples
 
 edition-test:
 	$(GO) test -v -count=1 ./pkg/editiontest/...
 
+interface-integrity:
+	@base_ref="$(BASE_REF)"; \
+	candidate_ref="$(CANDIDATE_REF)"; \
+	if [ -z "$$base_ref" ]; then base_ref="origin/main"; fi; \
+	if [ -z "$$candidate_ref" ]; then candidate_ref="HEAD"; fi; \
+	./scripts/policy/check-authoritative-interface-baselines.sh \
+		--base-ref "$$base_ref" \
+		--stable-ref "$(STABLE_REF)" \
+		--candidate-ref "$$candidate_ref"
+
+authoritative-interface-integrity:
+	@candidate_ref="$(CANDIDATE_REF)"; \
+	if [ -z "$$candidate_ref" ]; then candidate_ref="HEAD"; fi; \
+	./scripts/policy/check-authoritative-interface-baselines.sh \
+		--base-ref "$(BASE_REF)" \
+		--stable-ref "$(STABLE_REF)" \
+		--candidate-ref "$$candidate_ref"
+
+coverage-gate:
+	@./scripts/policy/check-coverage-gate.sh --base-ref "$(BASE_REF)" --scope-buildable
+
+coverage-gate-platform:
+	@./scripts/policy/run-platform-coverage-gate.sh --base-ref "$(BASE_REF)" --profile "$(PROFILE)"
+
+update-interface-baseline:
+	@./scripts/policy/check-interface-baseline.sh --update
+
+reset-interface-baseline:
+	@./scripts/policy/check-interface-baseline.sh --reset
+
+schema-compatibility:
+	@candidate_ref="$(CANDIDATE_REF)"; \
+	if [ -z "$$candidate_ref" ]; then candidate_ref="HEAD"; fi; \
+	./scripts/policy/check-authoritative-schema-compatibility.sh \
+		--base-ref "$(BASE_REF)" \
+		--stable-ref "$(STABLE_REF)" \
+		--candidate-ref "$$candidate_ref"
+
+skill-command-integrity:
+	@./scripts/policy/check-skill-commands.sh
+
+skill-context-budget:
+	@./scripts/policy/check-skill-context-budget.sh
+
+multi-im-skill-chain-integrity:
+	@./scripts/policy/check-multi-im-skill-chain.sh
+
+skill-mono-multi-content:
+	@./scripts/policy/check-mono-multi-skill-content.sh
+
+cli-smoke:
+	@./scripts/policy/check-cli-smoke.sh
+
+mock-mcp-smoke:
+	$(GO) test -v -count=1 -run '^(TestHTTPClientEndToEnd|TestStdioClientEndToEnd)$$' ./internal/transport
+
+test-schema-agent-examples:
+	DWS_AGENT_EXAMPLES_DRY_RUN=1 $(GO) test -v -count=1 ./internal/app -run '^TestAgentExamplesDryRun$$'
+
+# generate-schema refreshes param_aliases_generated.go and verifies that
+# ResolveSchemaBuild assembly is deterministic. Catalog is runtime-assembled
+# (声明即 Catalog); cmd_schema_catalog is not a committed delivery step.
+# schema_agent_metadata/ and schema_hints/ must stay absent.
+generate-schema:
+	@set -e; \
+	concepts_guard=$$(mktemp); \
+	concepts_schema_guard=$$(mktemp); \
+	command_fallbacks_guard=$$(mktemp); \
+	command_fallbacks_schema_guard=$$(mktemp); \
+	trap 'rm -f "$$concepts_guard" "$$concepts_schema_guard" "$$command_fallbacks_guard" "$$command_fallbacks_schema_guard"' EXIT HUP INT TERM; \
+	cp internal/cli/param_concepts.json "$$concepts_guard"; \
+	cp internal/cli/param_concepts.schema.json "$$concepts_schema_guard"; \
+	cp internal/cli/command_path_fallbacks.json "$$command_fallbacks_guard"; \
+	cp internal/cli/command_path_fallbacks.schema.json "$$command_fallbacks_schema_guard"; \
+	$(GO) generate ./internal/cli; \
+	rm -rf internal/cli/schema_agent_metadata internal/cli/schema_agent_metadata_audit.json; \
+	rm -f internal/cli/schema_meta_index.json; \
+	if [ -e internal/cli/schema_command_registry ]; then \
+		printf '%s\n' 'retired schema_command_registry/ must not reappear after generation' >&2; \
+		exit 1; \
+	fi; \
+	cmp -s internal/cli/param_concepts.json "$$concepts_guard" || { \
+		printf '%s\n' 'generation modified reviewed input internal/cli/param_concepts.json' >&2; \
+		exit 1; \
+	}; \
+	cmp -s internal/cli/param_concepts.schema.json "$$concepts_schema_guard" || { \
+		printf '%s\n' 'generation modified reviewed input internal/cli/param_concepts.schema.json' >&2; \
+		exit 1; \
+	}; \
+	cmp -s internal/cli/command_path_fallbacks.json "$$command_fallbacks_guard" || { \
+		printf '%s\n' 'generation modified reviewed input internal/cli/command_path_fallbacks.json' >&2; \
+		exit 1; \
+	}; \
+	cmp -s internal/cli/command_path_fallbacks.schema.json "$$command_fallbacks_schema_guard" || { \
+		printf '%s\n' 'generation modified reviewed input internal/cli/command_path_fallbacks.schema.json' >&2; \
+		exit 1; \
+	}; \
+	if [ -e internal/cli/schema_hints ]; then \
+		printf '%s\n' 'retired schema_hints/ must not reappear after generation' >&2; \
+		exit 1; \
+	fi; \
+	if [ -e internal/cli/schema_meta_index.json ]; then \
+		printf '%s\n' 'retired schema_meta_index.json must not remain after generation' >&2; \
+		exit 1; \
+	fi; \
+	./scripts/policy/check-schema-assembly.sh
+
+# Optional local/CI dump of an assembled Catalog under artifacts/ by default.
+# Override SCHEMA_CATALOG_OUTPUT and SCHEMA_META_INDEX_OUTPUT as needed. This
+# is not a go:generate or production delivery step.
+generate-schema-catalog:
+	$(GO) run -a ./internal/generator/cmd_schema_catalog \
+		-root . \
+		-output "$(SCHEMA_CATALOG_OUTPUT)" \
+		-meta-index "$(SCHEMA_META_INDEX_OUTPUT)"
+
+fetch-mcp-metadata:
+	@printf '  %sFetching diagnostic MCP dump (not a Schema pin)%s\n' "$(COLOR_RUN)" "$(COLOR_RESET)"
+	@./scripts/dev/fetch_mcp_metadata.sh
+
 package:
-	@./scripts/dev/build-all.sh
-	@./scripts/release/post-goreleaser.sh
+	@version="$(if $(VERSION),$(VERSION),v0.0.0-SNAPSHOT)"; VERSION="$${version#v}" ./scripts/dev/build-all.sh
+	@version="$(if $(VERSION),$(VERSION),v0.0.0-SNAPSHOT)"; DWS_PACKAGE_VERSION="$$version" ./scripts/release/post-goreleaser.sh
 
 publish-homebrew-formula:
 	@./scripts/release/publish-homebrew-formula.sh
@@ -47,6 +232,32 @@ publish-homebrew-formula:
 setup-hooks:
 	@git config core.hooksPath scripts/hooks 2>/dev/null || true
 
+changelog-pre:
+	@test -n "$(VERSION)" || (printf 'VERSION is required, e.g. v1.2.3-beta.1\n' >&2; exit 2)
+	@./scripts/release/prepare-changelog.sh prerelease "$(VERSION)"
+
+changelog-stable:
+	@test -n "$(VERSION)" || (printf 'VERSION is required, e.g. v1.2.3\n' >&2; exit 2)
+	@test -n "$(FROM_BETA)" || (printf 'FROM_BETA is required, e.g. v1.2.3-beta.2\n' >&2; exit 2)
+	@./scripts/release/prepare-changelog.sh stable "$(VERSION)" --from-beta "$(FROM_BETA)"
+
+release-pre:
+	@test -n "$(VERSION)" || (printf 'VERSION is required, e.g. v1.2.3-beta.1\n' >&2; exit 2)
+	@test -n "$(REMOTE)" || (printf 'REMOTE is required, e.g. origin\n' >&2; exit 2)
+	@args=""; \
+	  if [ "$(PUBLISH)" = "1" ]; then args="$$args --publish"; fi; \
+	  if [ "$(YES)" = "1" ]; then args="$$args --yes"; fi; \
+	  ./scripts/release/release.sh prerelease "$(VERSION)" --remote "$(REMOTE)" $$args
+
+release-stable:
+	@test -n "$(VERSION)" || (printf 'VERSION is required, e.g. v1.2.3\n' >&2; exit 2)
+	@test -n "$(FROM_BETA)" || (printf 'FROM_BETA is required, e.g. v1.2.3-beta.2\n' >&2; exit 2)
+	@test -n "$(REMOTE)" || (printf 'REMOTE is required, e.g. origin\n' >&2; exit 2)
+	@args=""; \
+	  if [ "$(PUBLISH)" = "1" ]; then args="$$args --publish"; fi; \
+	  if [ "$(YES)" = "1" ]; then args="$$args --yes"; fi; \
+	  ./scripts/release/release.sh stable "$(VERSION)" --from-beta "$(FROM_BETA)" --remote "$(REMOTE)" $$args
+
 release:
-	goreleaser release --clean
-	@./scripts/release/post-goreleaser.sh
+	@printf 'Use make release-pre or make release-stable; direct goreleaser publishing is disabled.\n' >&2
+	@exit 2

README.md

@@ -71,9 +71,7 @@ The installer ships skills in one of two layouts. CLI commands (`dws aitable ...
 | Mode | What gets installed | Best for |
 |------|----------------------|----------|
 | **mono** (stable, default) | One `dws` skill covering all products | Cross-product workflows; single entry point |
-| **multi** 🧪 **EXPERIMENTAL** | 22 per-product skills (`dingtalk-aitable`, `dingtalk-calendar`, `dingtalk-chat`, ...) | Single-product tasks; smaller context per call |
-
-> 🧪 **`multi` is currently EXPERIMENTAL / preview.** 22 product-scoped skills all pass the dispatch verifier, but interface, naming and cross-skill references may change in future releases. For production / shared environments, prefer `mono`. File issues if you hit problems.
+| **multi** | Per-product skills (`dingtalk-aitable`, `dingtalk-calendar`, `dingtalk-chat`, ...) | Single-product tasks; smaller context per call |
 
 How to pick:
 
@@ -93,6 +91,30 @@ How to pick:
 npm install -g dingtalk-workspace-cli
 ```
 
+Install the latest beta:
+
+```bash
+npm install -g dingtalk-workspace-cli@beta
+```
+
+**Homebrew** (macOS / Linux):
+
+```bash
+brew tap DingTalk-Real-AI/dingtalk-workspace-cli https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli.git
+brew install dingtalk-workspace-cli
+```
+
+> The Formula lives in this repository, so the first `tap` command must include the explicit repository URL. Afterwards, use `brew upgrade dingtalk-workspace-cli` normally.
+
+Install the keg-only Homebrew beta without replacing the stable Formula:
+
+```bash
+brew install dingtalk-workspace-cli-beta
+$(brew --prefix dingtalk-workspace-cli-beta)/bin/dws version
+```
+
+To make the beta `dws` the default for the current shell, prepend `$(brew --prefix dingtalk-workspace-cli-beta)/bin` to PATH.
+
 **Pre-built binary**: download from [GitHub Releases](https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases).
 
 > **macOS users**: If you see "cannot be opened because Apple cannot check it for malicious software", run:
@@ -168,6 +190,18 @@ dws upgrade -y                 # skip confirmation prompt
 
 By default, `dws upgrade` follows the stable release track. Use `--beta` only when you explicitly want the newest GitHub pre-release build.
 
+### Six-channel post-release verification
+
+Maintainers and release validators can run the release-quality smoke checks for curl, PowerShell, npm stable, npm beta, Homebrew, and `dws upgrade`:
+
+```bash
+git clone https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli.git /tmp/dws-verify
+cd /tmp/dws-verify/verify
+bash verify-all-channels.sh
+```
+
+The verifier uses isolated directories and does not replace the `dws` on the current PATH. It reports `PASS`, `FAIL`, and `SKIP`; a platform skip is not a pass and must be covered on the matching host. See [`verify/README.md`](verify/README.md) for the platform matrix.
+
 <details>
 <summary><strong>How it works</strong></summary>
 
@@ -247,16 +281,32 @@ Credentials are securely persisted after first login (Keychain). Subsequent runs
 <details>
 <summary><strong>Multiple organizations (profiles)</strong></summary>
 
-`dws` can stay logged in to several DingTalk organizations at once. Each organization is one **profile**; the current profile decides which org a command runs against (credentials are stored per organization).
+`dws` can stay logged in to several DingTalk accounts at once, including multiple accounts in the same organization. A profile is uniquely identified by `corpId:userId`; the current profile decides which identity a command runs as.
 
 ```bash
-dws auth login                              # log in to another org → adds a profile (first login becomes the primary)
-dws profile list                            # list logged-in orgs (primary / current marker, status)
-dws profile switch <name|corpId>            # switch the default org (use - to toggle back to the previous one)
-dws --profile <name|corpId> contact user search --query "..."   # run one command against a specific org, without changing the default
+dws auth login                              # add or refresh one account
+dws profile list                            # list every logged-in account
+dws profile switch <corpId:userId>          # persistently switch; use - to toggle back
+dws profile switch "<corpName>:<userName>"  # friendly input; names must be unique
+dws --profile <corpId> contact user search --query "..."        # use that org's explicitly recorded current account
+dws --profile <corpId:userId> contact user search --query "..." # use one exact account without changing the default
 ```
 
-Cross-org reads are orchestrated by the agent rather than a built-in `--all-orgs`: list the profiles, run the query per org with `--profile`, then merge. Writes default to the current org only — confirm the target org before writing across orgs.
+Selectors support `corpId:userId`, `corpId:userName`, `corpName:userId`, and `corpName:userName`. Friendly names are input aliases only; use the stable `profile` value returned by `profile list` for automation. Duplicate organization or account names fail with explicit `corpId:userId` candidates. If an organization has multiple accounts but no recorded current account, `--profile <corpId>` fails instead of choosing the first or most recently used account.
+
+`currentProfile`, `previousProfile`, and per-organization defaults are stored as exact identities. `primaryProfile` remains in JSON only for compatibility and is not used for selection. `profile list` reads status and expiry from each real identity Token without refreshing it. `auth logout --profile <corpId>` removes all local accounts in that organization; an exact selector or local profile name removes one account.
+
+Cross-org reads are orchestrated by the agent rather than a built-in `--all-orgs`: list profiles, group by `corpId`, and use the unique `isOrgCurrent=true` account for each organization. If a multi-account organization has no default, ask the user to choose an account first. Writes default to the current account — confirm both organization and account before cross-org writes.
+
+On macOS, an unreadable registered token slot blocks a new OAuth login rather than risking a mixed Keychain/file-DEK state. If normal terminal commands can still read the login while a sandbox using `DWS_DISABLE_KEYCHAIN=1` cannot, migrate the legacy and profile auth entries without exposing tokens:
+
+```bash
+env -u DWS_DISABLE_KEYCHAIN dws auth migrate-keychain --to file-dek --dry-run --format json
+env -u DWS_DISABLE_KEYCHAIN dws auth migrate-keychain --to file-dek --yes --format json
+DWS_DISABLE_KEYCHAIN=1 dws auth status --format json
+```
+
+The migration validates every selected auth ciphertext before writing, ignores unrelated application secrets, and can be rerun after an interrupted commit. If validation identifies genuinely damaged ciphertext, remove only the affected account with `dws auth logout --profile <corpId:userId>`, or all accounts in one organization with `--profile <corpId>`, then log in again. Use `dws auth reset` only when you intend to discard every local profile.
 
 </details>
 
@@ -277,6 +327,9 @@ dws auth status   # confirm "Refresh Token: valid"
 ```
 
 The bundle includes the encrypted keychain under `~/.local/share/dws-cli` (with `auth-token.enc` and `dek`) plus required `~/.dws` config files.
+Windows export and import are intentionally rejected before credentials or
+bundles are read: Windows stores credentials as DPAPI-protected HKCU Registry
+values, and the current file-DEK bundle has no safe DPAPI-to-portable conversion.
 
 </details>
 
@@ -313,25 +366,35 @@ dws contact user get-self --jq '.result[0].orgEmployeeModel | {name: .orgUserNam
 
 ### Command Help and Schema
 
-Product commands are compiled into the binary in static endpoint mode. Use `--help` and the bundled Agent Skills as the source of truth; `dws schema` is retained for helper-only schemas such as `dev.*`.
+Use Cobra help and Schema for different parts of the command contract:
+
+- `dws <path> --help` is the source of truth for whether a command exists and which flags the binary accepts.
+- `dws schema "<path>" --compact` is the normative Agent view for command selection, CLI parameters and constraints, risk, and confirmation; use a full leaf with a narrow `--jq` projection for mapping or provenance audits.
+- If Help and Schema disagree, treat it as contract drift: pass only flags accepted by Cobra and use the more conservative safety semantics.
+- Schema describes commands; it does not read or search DingTalk business data. Execute the real product command after discovery.
 
 ```bash
-# Inspect the current compiled command surface
+# Confirm that the command exists and inspect accepted flags
 dws aitable record query --help
 
-# Helper-only schema introspection
-dws schema "dev app create"
+# Discover within a product, then inspect the selected leaf contract
+dws schema aitable --compact
+dws schema "aitable record query" --compact
 
-# Construct the call
+# Execute the real business query
 dws aitable record query --base-id BASE_ID --table-id TABLE_ID --limit 10
 ```
 
+`dws schema --all` exports the complete contract for tooling, CI, audits, and compatibility baselines. Agents should query progressively with `--compact`; its positive field allowlist prevents new full/audit fields from silently expanding Agent context.
+
 ### Agent Skills
 
-The repo ships a complete Agent Skill system under `skills/`, now organized into two layouts:
+The repo ships a complete Agent Skill system under `skills/`, organized into two layouts:
 
 - `skills/mono/` — single-skill layout (one `SKILL.md` + `references/products/`), recommended default.
-- `skills/multi/` — per-product skills (`dingtalk-aitable/`, `dingtalk-calendar/`, `dingtalk-chat/`, ... 22 products in total), each with its own `SKILL.md`. 🧪 **EXPERIMENTAL / preview — see banner in each multi `SKILL.md` for caveats.**
+- `skills/multi/` — per-product skills (`dingtalk-aitable/`, `dingtalk-calendar/`, `dingtalk-chat/`, ...), each with its own `SKILL.md`.
+
+Leaf safety/parameters/selection prose for Schema generation come from ProductDecl / ContractFinal declarations in Go. The former `internal/cli/schema_hints/` HintFile tree is fully retired and must not reappear.
 
 After installing, AI tools like Claude Code / Cursor can operate DingTalk directly through natural language:
 
@@ -378,7 +441,6 @@ Env vars: `DWS_SKILL_MODE=mono|multi` (also honored by `install.sh` / `install.p
 | Intent guide | `skills/mono/references/intent-guide.md` | Disambiguation for confusing scenarios (e.g. report vs todo) |
 | Global reference | `skills/mono/references/global-reference.md` | Auth, output formats, global flags |
 | Error codes | `skills/mono/references/error-codes.md` | Error codes + debugging workflows |
-| Recovery guide | `skills/mono/references/recovery-guide.md` | `RECOVERY_EVENT_ID` handling |
 | Ready-made scripts | `skills/mono/scripts/*.py` | 13 batch operation scripts (see below) |
 
 <details>
@@ -406,6 +468,88 @@ Env vars: `DWS_SKILL_MODE=mono|multi` (also honored by `install.sh` / `install.p
 
 ## Features
 
+<details>
+<summary><strong>Personal Event Subscription</strong> — real-time DingTalk messages for event-driven agents</summary>
+
+`dws event consume` subscribes as the currently logged-in user over a managed Stream WebSocket and emits each event as one NDJSON line on stdout. The public catalog covers scoped and all one-to-one/group messages, specified senders, read/recall/reaction events, group lifecycle events, and six OA approval task/instance events.
+
+The default `ndjson`, `json`, and `pretty` output preserves the transport envelope (`type`, `event_type`, string `data`, and `headers`) for existing scripts; `compact` retains its existing processor. Add `--flatten` to emit the stable top-level business fields used by Agent workflows. `--format` controls JSON serialization; `--flatten` controls the data structure and cannot be combined with `-f raw` or `--debug-raw-events`.
+
+> **Prerequisite**: run `dws auth login`. Personal identity is resolved from the OAuth token and cannot be supplied through command-line identity flags.
+
+For an event-focused installation, use the official convenience installer:
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/main/scripts/install-event.sh | sh
+
+# Or install the standalone multi skill from an existing dws installation
+dws skill setup --mode multi -s event
+```
+
+```bash
+# Inspect the public personal event catalog and schema
+dws event list
+dws event schema user_im_message_receive_o2o --flatten
+dws event list --category oa
+dws event schema user_oa_approval_task_created --flatten
+
+# Listen for messages that mention the current user
+dws event +listen-im --kind at-me -f ndjson
+
+# Listen for messages from a specified sender
+dws event +listen-im --kind sender --user <userId> -f ndjson
+
+# Listen by openDingtalkId (external contact, bot, or cross-organization identity)
+dws event +listen-im --kind sender --open-dingtalk-id <openDingtalkId> -f ndjson
+
+# Listen for messages in a specified group
+dws event +listen-im --kind group --chat-id <openConversationId> -f ndjson
+
+# Listen for all one-to-one or all group messages
+dws event +listen-im --kind all-direct -f ndjson
+dws event +listen-im --kind all-group -f ndjson
+
+# Listen for a specified group's title changes, member changes, or disband event
+dws event consume user_im_group_updated --group <openConversationId> --flatten -f ndjson
+dws event consume user_im_group_member_added --group <openConversationId> --flatten -f ndjson
+dws event consume user_im_group_member_exited --group <openConversationId> --flatten -f ndjson
+dws event consume user_im_group_disbanded --group <openConversationId> --flatten -f ndjson
+
+# Listen for messages, reads, and recalls from the same sender in one process
+dws event +listen-im --kind sender --user <userId> \
+  --events message,read,recall -f ndjson
+
+# Listen for all six public OA approval events in one process
+dws event consume \
+  user_oa_approval_task_created \
+  user_oa_approval_task_finished \
+  user_oa_approval_task_redirected \
+  user_oa_approval_instance_started \
+  user_oa_approval_instance_terminated \
+  user_oa_approval_instance_finished \
+  --flatten -f ndjson
+
+# Inspect local consumers and cancel a subscription
+dws event status
+dws event stop <subscribe_id>
+```
+
+For one-to-one and specified-sender events, use exactly one target identity: `--user` for an internal `userId`, or `--open-dingtalk-id` for an `openDingtalkId`. The CLI does not infer or convert between these identity types.
+
+| Feature | Details |
+|---------|---------|
+| Managed lifecycle | `consume` creates or reuses the personal subscription; `stop` cancels it and cleans local state |
+| Shared connection | Consumers for the same user share one local bus and cloud connection |
+| Multi-event process | One consume process can listen for compatible events for the same target while retaining one subscription per event |
+| Subscription isolation | Normal consumers match both event type and `subscribe_id` |
+| Agent-friendly output | Stream events are written to stdout as NDJSON; status and diagnostics use stderr |
+| Observability | `status` shows remote subscriptions, the personal bus, and local consumers |
+| Cross-platform | Unix Socket on macOS/Linux, Windows Named Pipe on Windows |
+
+See `skills/multi/dingtalk-event/SKILL.md` for the Agent workflow and supported event parameters.
+
+</details>
+
 <details>
 <summary><strong>Raw API Access</strong> — call any DingTalk OpenAPI directly</summary>
 
@@ -487,7 +631,7 @@ dws aitable record query --base-id BASE_ID --tabel-id TABLE_ID       # --tabel-i
 ```bash
 # Built-in jq expressions
 dws aitable record query --base-id BASE_ID --table-id TABLE_ID --jq '.invocation.params'
-dws schema "dev app create" --jq '.tool.required'
+dws schema "dev app create" --jq '.parameters'
 
 # Return only specific fields
 dws aitable record query --base-id BASE_ID --table-id TABLE_ID --fields invocation,response
@@ -496,12 +640,13 @@ dws aitable record query --base-id BASE_ID --table-id TABLE_ID --fields invocati
 </details>
 
 <details>
-<summary><strong>Schema Introspection</strong> — helper-only schemas in static endpoint mode</summary>
+<summary><strong>Schema Introspection</strong> — Agent command discovery and execution contracts</summary>
 
 ```bash
-dws schema                                              # static endpoint mode note
-dws schema "dev app create"                             # view helper-only schema
-dws schema "dev app create" --jq '.tool.required'        # view required fields
+dws schema aitable --compact                            # discover product commands
+dws schema "aitable record query" --compact             # view the selected Agent leaf contract
+dws schema "aitable record query" --jq '[.parameters | to_entries[] | select(.value.required)]' # view required fields
+dws schema --all                                        # full export for CI/audit/baselines
 ```
 
 </details>
@@ -553,7 +698,7 @@ See [`docs/robot-quickstart.md`](./docs/robot-quickstart.md) for the full 4-step
 
 | Service | Command | Capabilities |
 |---------|---------|--------------|
-| Contact | `contact` | Look up users by name / mobile / job-number, departments, labels & roles, roster profiles & dismissals |
+| Contact | `contact` | Look up users, departments, labels, roster profiles and dismissals; create enterprises and enterprise accounts; invite employees |
 | Chat / IM | `chat` (`im`) | Send / reply / search messages, group & member management, bot & webhook messaging, reactions, recall |
 | Calendar | `calendar` | Events CRUD, attendees, meeting rooms, free/busy & time suggestions |
 | Todo | `todo` | Create / list / update / complete tasks and comments |

README_zh.md

@@ -71,9 +71,7 @@ irm https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/ma
 | 模式 | 安装内容 | 适合场景 |
 |------|----------|----------|
 | **mono**（稳定，默认） | 一个 `dws` skill，覆盖全部产品 | 跨产品组合操作；单一入口召唤 |
-| **multi** 🧪 **试验版 / Preview** | 22 个独立产品 skill（`dingtalk-aitable` / `dingtalk-calendar` / `dingtalk-chat` ...） | 单产品任务；每次召唤上下文更小 |
-
-> 🧪 **multi 模式当前为 EXPERIMENTAL（试验版 / Preview）**。22 个独立 skill 全部通过 dispatch verifier，但接口、命名、跨 skill 引用后续可能调整。生产 / 共享环境建议优先用 `mono`。问题请提 issue 反馈。
+| **multi** | 按产品拆分的独立 skill（`dingtalk-aitable` / `dingtalk-calendar` / `dingtalk-chat` ...） | 单产品任务；每次召唤上下文更小 |
 
 怎么选：
 
@@ -93,6 +91,30 @@ irm https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/ma
 npm install -g dingtalk-workspace-cli
 ```
 
+安装最新 beta：
+
+```bash
+npm install -g dingtalk-workspace-cli@beta
+```
+
+**Homebrew**（macOS / Linux）：
+
+```bash
+brew tap DingTalk-Real-AI/dingtalk-workspace-cli https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli.git
+brew install dingtalk-workspace-cli
+```
+
+> Formula 与代码位于同一个仓库，因此首次 `tap` 需要显式指定仓库 URL。后续可直接使用 `brew upgrade dingtalk-workspace-cli`。
+
+安装 Homebrew beta（keg-only，不覆盖稳定版）：
+
+```bash
+brew install dingtalk-workspace-cli-beta
+$(brew --prefix dingtalk-workspace-cli-beta)/bin/dws version
+```
+
+如需让 beta 的 `dws` 成为当前 shell 默认版本，将 `$(brew --prefix dingtalk-workspace-cli-beta)/bin` 放到 PATH 最前面。
+
 **预编译二进制文件**：从 [GitHub Releases](https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases) 下载。
 
 > **macOS 用户注意**：如果提示“无法打开，因为 Apple 无法检查其是否包含恶意软件”，请执行：
@@ -165,6 +187,18 @@ dws upgrade -y                 # 跳过确认直接升级
 
 默认情况下，`dws upgrade` 只跟随正式 release 轨道。只有显式传入 `--beta` 时，才会选择 GitHub pre-release 里的 beta 构建。
 
+### 六渠道发布后验证
+
+维护者和验证同学可按发版质量保障 SOP，对 curl、PowerShell、npm stable、npm beta、Homebrew、`dws upgrade` 执行安装与冒烟验证：
+
+```bash
+git clone https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli.git /tmp/dws-verify
+cd /tmp/dws-verify/verify
+bash verify-all-channels.sh
+```
+
+脚本使用隔离目录，不会替换当前 PATH 中的 `dws`；输出 `PASS`、`FAIL`、`SKIP` 汇总。跨平台渠道必须由对应平台补测，`SKIP` 不计为通过。验证范围和平台矩阵见 [`verify/README.md`](verify/README.md)。
+
 <details>
 <summary><strong>工作原理</strong></summary>
 
@@ -244,16 +278,32 @@ dws auth login --client-id <your-app-key> --client-secret <your-app-secret>
 <details>
 <summary><strong>多组织（profile）</strong></summary>
 
-`dws` 可以同时登录多个钉钉组织。一个组织就是一个 **profile**，当前 profile 决定本次命令操作哪个组织（凭证按组织分别存储）。
+`dws` 可以同时登录多个钉钉账号，同一组织也能保留多个账号。一个 profile 由 `corpId + userId` 唯一确定。
 
 ```bash
-dws auth login                              # 再登录一个组织 → 新增一个 profile（首次登录的为主组织）
-dws profile list                            # 列出已登录组织（主 / 当前标记、状态）
-dws profile switch <名称|corpId>            # 切换默认组织（用 - 切回上一个）
-dws --profile <名称|corpId> contact user search --query "..."   # 单次对指定组织执行，不改默认组织
+dws auth login                              # 新增或刷新一个账号
+dws profile list                            # 列出全部账号，profile 字段是稳定的 corpId:userId
+dws profile switch <corpId:userId>          # 持久切换账号；用 - 切回上一个
+dws profile switch "组织名:用户名"          # 名称输入要求唯一
+dws --profile <corpId> contact user search --query "..."        # 使用该组织明确记录的当前账号
+dws --profile <corpId:userId> contact user search --query "..." # 单次精确指定账号，不改默认账号
 ```
 
-跨组织读取由 agent 编排，而非内置 `--all-orgs`：先 `dws profile list` 拿到组织，再对每个组织带 `--profile` 各查一遍，然后合并。写操作默认只在当前组织进行——跨组织写之前先确认目标组织。
+支持 `corpId:userId`、`corpId:userName`、`corpName:userId`、`corpName:userName`。名称只用于输入，自动化应使用 `profile list` 返回的稳定 `profile`。组织名或用户名重名时会列出候选并报错；同组织多账号但没有明确当前账号时，只传组织也会报错，不会选择第一项或最近使用账号。
+
+`currentProfile`、`previousProfile` 和组织默认账号都保存精确身份。`primaryProfile` 只为 JSON 兼容保留，不再参与选择。`profile list` 直接读取各身份 Token 计算状态和到期时间，不触发刷新。`auth logout --profile <corpId>` 退出该组织全部账号；精确选择器或本地 profile 名只退出一个账号。
+
+跨组织读取由 agent 编排，而非内置 `--all-orgs`：先 `dws profile list`，每个组织使用唯一的 `isOrgCurrent=true` 账号；若多账号组织没有默认账号，先让用户指定账号。写操作默认只在当前账号执行——跨组织写之前先确认目标组织和账号。
+
+macOS 下，如果已登记的 token slot 无法解密，为避免把系统 Keychain 和 file-DEK 写成混合状态，新的 OAuth 登录会直接拒绝。如果普通终端仍能读取登录态、只有设置 `DWS_DISABLE_KEYCHAIN=1` 的沙箱读不到，可在不暴露 token 的情况下迁移 legacy 与各 profile 的认证条目：
+
+```bash
+env -u DWS_DISABLE_KEYCHAIN dws auth migrate-keychain --to file-dek --dry-run --format json
+env -u DWS_DISABLE_KEYCHAIN dws auth migrate-keychain --to file-dek --yes --format json
+DWS_DISABLE_KEYCHAIN=1 dws auth status --format json
+```
+
+迁移会先验证全部认证密文再写入、忽略无关的应用密钥；提交中断后可安全重跑。如果预检确认是密文本身损坏，优先使用 `dws auth logout --profile <corpId:userId>` 只清理受影响账号；只有确认要丢弃全部本地 profile 时才用 `dws auth reset`。
 
 </details>
 
@@ -310,25 +360,35 @@ dws contact user get-self --jq '.result[0].orgEmployeeModel | {name: .orgUserNam
 
 ### 命令帮助与 Schema
 
-产品命令在静态端点模式下已经编译进二进制。Agent 以 `--help` 和内置 Skill 为事实源；`dws schema` 仅保留给 `dev.*` 等 helper-only schema 查询。
+命令帮助和 Schema 分别负责命令契约的不同部分：
+
+- `dws <path> --help` 是命令是否存在、当前二进制接受哪些 flags 的事实源。
+- `dws schema "<path>" --compact` 是 Agent 选命令、CLI 参数与约束、风险和确认语义的规范视图；映射或 provenance 审计使用 full leaf 配合 `--jq` 精确投影。
+- Help 与 Schema 冲突时视为契约漂移：执行只传 Cobra 接受的参数，安全语义取更保守值。
+- Schema 只描述命令，不读取或搜索钉钉业务数据；发现命令后仍需执行真实产品命令。
 
 ```bash
-# 查看当前编译出的命令面
+# 确认命令存在并查看当前接受的 flags
 dws aitable record query --help
 
-# helper-only schema 自省
-dws schema "dev app create"
+# 先在产品内发现命令，再查看选中 leaf 的契约
+dws schema aitable --compact
+dws schema "aitable record query" --compact
 
-# 构造正确的调用
+# 执行真实业务查询
 dws aitable record query --base-id BASE_ID --table-id TABLE_ID --limit 10
 ```
 
+`dws schema --all` 会完整导出命令契约，供工具、CI、审计和兼容性基线使用。Agent 应使用 `--compact` 渐进查询；该视图采用正向字段白名单，full 新增的审计字段不会自动进入 Agent 上下文。
+
 ### Agent Skills
 
-仓库内置完整的 Agent Skill 体系（`skills/` 目录），目前重组为两套布局：
+仓库内置完整的 Agent Skill 体系（`skills/` 目录），分为两套布局：
 
 - `skills/mono/` — 单 skill 布局（一个 `SKILL.md` + `references/products/`），默认推荐。
-- `skills/multi/` — 每个产品一个独立 skill（`dingtalk-aitable/` / `dingtalk-calendar/` / `dingtalk-chat/` ... 共 22 个），每个 skill 自带 `SKILL.md`。🧪 **试验版 / Preview — 各 multi `SKILL.md` 头部有详细注意事项。**
+- `skills/multi/` — 每个产品一个独立 skill（`dingtalk-aitable/` / `dingtalk-calendar/` / `dingtalk-chat/` ...），每个 skill 自带 `SKILL.md`。
+
+Schema 生成的叶子 safety/参数/选型文案由 Go 中的 ProductDecl / ContractFinal 声明驱动。原 `internal/cli/schema_hints/` HintFile 目录已完全退役，不得重新引入。
 
 安装之后，Claude Code / Cursor 等 AI 工具就能通过自然语言直接操作钉钉：
 
@@ -375,7 +435,6 @@ DWS_SKILL_SOURCE=/path/to/skills dws skill setup --mode multi
 | 意图指南 | `skills/mono/references/intent-guide.md` | 易混淆场景消歧（如 report vs todo） |
 | 全局参考 | `skills/mono/references/global-reference.md` | 认证、输出格式、全局 flag |
 | 错误码 | `skills/mono/references/error-codes.md` | 错误码 + 调试流程 |
-| Recovery 指南 | `skills/mono/references/recovery-guide.md` | `RECOVERY_EVENT_ID` 处理 |
 | 现成脚本 | `skills/mono/scripts/*.py` | 13 个批量操作脚本（见下方） |
 
 <details>
@@ -403,6 +462,88 @@ DWS_SKILL_SOURCE=/path/to/skills dws skill setup --mode multi
 
 ## 功能特性
 
+<details>
+<summary><strong>个人事件订阅</strong> — 实时接收钉钉消息，驱动事件触发的 Agent</summary>
+
+`dws event consume` 使用当前 OAuth 登录用户建立托管的 Stream WebSocket 长连接，并把每条事件以 NDJSON 一行输出到 stdout。当前公开目录覆盖指定范围和全量单聊/群消息、指定发送人、已读/撤回/表情回应、群生命周期，以及六个 OA 审批任务/实例事件。
+
+默认 `ndjson`、`json`、`pretty` 输出保留兼容 transport envelope（`type`、`event_type`、字符串 `data`、`headers`），`compact` 继续沿用原 processor。Agent 或新脚本显式加 `--flatten` 后，输出稳定的顶层业务字段。`--format` 控制 JSON 序列化，`--flatten` 控制数据结构，且不能与 `-f raw` 或 `--debug-raw-events` 同时使用。
+
+> **前置条件**：先运行 `dws auth login`。个人身份从 OAuth token 解析，不允许通过命令行伪造。
+
+只需要 event 能力时，可以使用官方便捷安装脚本：
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/main/scripts/install-event.sh | sh
+
+# 或在已有 dws 环境中安装独立的 multi skill
+dws skill setup --mode multi -s event
+```
+
+```bash
+# 查看公开个人事件目录和 schema
+dws event list
+dws event schema user_im_message_receive_o2o --flatten
+dws event list --category oa
+dws event schema user_oa_approval_task_created --flatten
+
+# 监听当前用户被 @ 的消息
+dws event +listen-im --kind at-me -f ndjson
+
+# 监听指定发送人的消息
+dws event +listen-im --kind sender --user <userId> -f ndjson
+
+# 使用 openDingtalkId 监听外部联系人、机器人或跨组织身份
+dws event +listen-im --kind sender --open-dingtalk-id <openDingtalkId> -f ndjson
+
+# 监听指定群的消息
+dws event +listen-im --kind group --chat-id <openConversationId> -f ndjson
+
+# 监听所有单聊或所有群消息
+dws event +listen-im --kind all-direct -f ndjson
+dws event +listen-im --kind all-group -f ndjson
+
+# 监听指定群标题变更、成员进退群或群解散
+dws event consume user_im_group_updated --group <openConversationId> --flatten -f ndjson
+dws event consume user_im_group_member_added --group <openConversationId> --flatten -f ndjson
+dws event consume user_im_group_member_exited --group <openConversationId> --flatten -f ndjson
+dws event consume user_im_group_disbanded --group <openConversationId> --flatten -f ndjson
+
+# 一个进程监听同一发送人的消息、已读和撤回
+dws event +listen-im --kind sender --user <userId> \
+  --events message,read,recall -f ndjson
+
+# 一个进程监听全部六个公开 OA 审批事件
+dws event consume \
+  user_oa_approval_task_created \
+  user_oa_approval_task_finished \
+  user_oa_approval_task_redirected \
+  user_oa_approval_instance_started \
+  user_oa_approval_instance_terminated \
+  user_oa_approval_instance_finished \
+  --flatten -f ndjson
+
+# 查看本地 consume，并取消指定订阅
+dws event status
+dws event stop <subscribe_id>
+```
+
+单聊和指定发送人事件必须且只能选择一种目标身份：企业内部 `userId` 使用 `--user`，`openDingtalkId` 使用 `--open-dingtalk-id`。CLI 不会自动猜测或转换身份类型。
+
+| 特性 | 说明 |
+|------|------|
+| 自动编排 | `consume` 创建或复用个人订阅，`stop` 取消订阅并清理本地状态 |
+| 共享连接 | 同一用户的多个 consumer 共享本地 bus 和云端长连接 |
+| 多事件进程 | 同一目标的兼容事件可由一个 consume 进程监听，每个事件仍有独立订阅 |
+| 订阅隔离 | 正常 consumer 同时按事件类型和 `subscribe_id` 匹配 |
+| Agent 友好输出 | Stream 事件写入 stdout，连接状态和诊断信息写入 stderr |
+| 状态可观测 | `status` 同时显示服务端订阅、personal bus 和本地 consumers |
+| 跨平台 | macOS/Linux 使用 Unix Socket，Windows 使用 Named Pipe |
+
+Agent 工作流和事件参数详见 `skills/multi/dingtalk-event/SKILL.md`。
+
+</details>
+
 <details>
 <summary><strong>Raw API 调用</strong> — 直接调用钉钉 OpenAPI</summary>
 
@@ -484,7 +625,7 @@ dws aitable record query --base-id BASE_ID --tabel-id TABLE_ID       # --tabel-i
 ```bash
 # 内置 jq 表达式
 dws aitable record query --base-id BASE_ID --table-id TABLE_ID --jq '.invocation.params'
-dws schema "dev app create" --jq '.tool.required'
+dws schema "dev app create" --jq '.parameters'
 
 # 只返回指定字段
 dws aitable record query --base-id BASE_ID --table-id TABLE_ID --fields invocation,response
@@ -493,12 +634,13 @@ dws aitable record query --base-id BASE_ID --table-id TABLE_ID --fields invocati
 </details>
 
 <details>
-<summary><strong>Schema 自省</strong> — 静态端点模式下的 helper-only schema</summary>
+<summary><strong>Schema 自省</strong> — Agent 命令发现与执行契约</summary>
 
 ```bash
-dws schema                                              # 静态端点模式提示
-dws schema "dev app create"                             # 查看 helper-only schema
-dws schema "dev app create" --jq '.tool.required'        # 查看必填字段
+dws schema aitable --compact                            # 发现产品命令
+dws schema "aitable record query" --compact             # 查看 Agent leaf 契约
+dws schema "aitable record query" --jq '[.parameters | to_entries[] | select(.value.required)]' # 定向查看必填字段
+dws schema --all                                        # CI/审计/基线的全量导出
 ```
 
 </details>
@@ -545,7 +687,7 @@ dws dev connect --channel auto --robot-client-id <id> --robot-client-secret <sec
 
 | 服务 | 命令 | 能力 |
 |------|------|------|
-| 通讯录 | `contact` | 按姓名 / 手机号 / 工号查人，部门、角色标签、花名册与离职 |
+| 通讯录 | `contact` | 按姓名 / 手机号 / 工号查人，部门、角色标签、花名册与离职；创建企业、企业账号及邀请员工 |
 | 群聊 | `chat`（`im`）| 发送 / 回复 / 搜索消息，群与成员管理，机器人与 Webhook 发消息，表情反应，撤回 |
 | 日历 | `calendar` | 日程 CRUD、参与者、会议室、闲忙与时间建议 |
 | 待办 | `todo` | 创建 / 列表 / 修改 / 完成待办及评论 |

auto-test/cli_to_mcp/README.md

@@ -1,16 +0,0 @@
-# cli_to_mcp smoke tests
-
-This directory contains lightweight command-to-tool contract tests for hardcoded
-DWS commands synced from `dws-wukong`.
-
-The tests do not call live DingTalk APIs. They exercise command help, validation,
-and `--dry-run` output so command paths and MCP argument mappings stay stable.
-
-Run with an already built binary:
-
-```bash
-DWS_BIN=/path/to/dws pytest auto-test/cli_to_mcp/testcases
-```
-
-If `DWS_BIN` is not set, the runner falls back to `go run ./cmd` from the repo
-root.

auto-test/cli_to_mcp/testcases/agoal/test_01_agoal_commands.py

@@ -1,193 +0,0 @@
-from test_utils import combined_output, dry_run_args
-
-
-def assert_ok(result):
-    output = combined_output(result)
-    assert result.returncode == 0, output
-    return output
-
-
-def test_agoal_strategy_and_contract_cli_to_mcp(dws):
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "strategy",
-            "list",
-            "--scope-type",
-            "PERSONAL",
-            "--scope-id",
-            "user123",
-            "--request-id",
-            "req-1",
-            "--dry-run",
-        )
-    )
-    assert "list_strategy_decodings" in output
-    assert dry_run_args(output) == {
-        "scopeType": "PERSONAL",
-        "openId": "user123",
-        "requestId": "req-1",
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "strategy",
-            "update",
-            "--profile-id",
-            "profile123",
-            "--content",
-            '[{"id":"e1","title":{"title":"new"}}]',
-            "--dry-run",
-        )
-    )
-    assert "update_strategy_decoding" in output
-    assert dry_run_args(output) == {
-        "profileId": "profile123",
-        "content": [{"id": "e1", "title": {"title": "new"}}],
-    }
-
-    output = assert_ok(dws.run_raw("agoal", "contract", "fields", "--dry-run"))
-    assert "list_op_contract_fields" in output
-    assert dry_run_args(output) == {}
-
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "contract",
-            "update",
-            "--contract-id",
-            "contract123",
-            "--dimensions",
-            '[{"id":"dim1","title":"metric"}]',
-            "--audit-config",
-            '{"needAudit":true}',
-            "--objective-template",
-            '{"id":"tpl1"}',
-            "--dry-run",
-        )
-    )
-    assert "update_op_contract" in output
-    assert dry_run_args(output) == {
-        "contractId": "contract123",
-        "dimensions": [{"id": "dim1", "title": "metric"}],
-        "auditConfig": '{"needAudit":true}',
-        "objectiveTemplate": '{"id":"tpl1"}',
-    }
-
-
-def test_agoal_scorecard_user_report_template_cli_to_mcp(dws):
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "scorecard",
-            "detail",
-            "--selected-time",
-            "2026-01-01T00:00:00+08:00",
-            "--dept-id",
-            "dept123",
-            "--dry-run",
-        )
-    )
-    assert "get_score_card_detail" in output
-    args = dry_run_args(output)
-    assert args["deptId"] == "dept123"
-    assert args["selectedTime"] == 1767196800000
-
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "scorecard",
-            "update",
-            "--dept-id",
-            "dept123",
-            "--selected-time",
-            "2026-01-01",
-            "--id",
-            "sc123",
-            "--tracking-period-type",
-            "MONTHLY",
-            "--content",
-            '[{"id":"dim1","items":[]}]',
-            "--dry-run",
-        )
-    )
-    assert "update_score_card" in output
-    args = dry_run_args(output)
-    assert args["selectedTime"] == 1767196800000
-    assert args["content"] == [{"id": "dim1", "items": []}]
-
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "user",
-            "objectives",
-            "--user-id",
-            "user123",
-            "--rule-id",
-            "rule123",
-            "--period-ids",
-            "p1,p2",
-            "--dry-run",
-        )
-    )
-    assert "list_user_objectives" in output
-    assert dry_run_args(output) == {
-        "dingUserId": "user123",
-        "objectiveRuleId": "rule123",
-        "periodIds": ["p1", "p2"],
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "report",
-            "submit-detail",
-            "--template-id",
-            "tpl123",
-            "--submit-state",
-            "LATE",
-            "--query-date",
-            "2026-06-18T00:00:00+08:00",
-            "--page",
-            "1",
-            "--page-size",
-            "20",
-            "--keyword",
-            "alice",
-            "--dry-run",
-        )
-    )
-    assert "get_submit_detail" in output
-    assert dry_run_args(output) == {
-        "templateId": "tpl123",
-        "submitState": "LATE",
-        "queryDate": "2026-06-18",
-        "page": 1,
-        "pageSize": 20,
-        "keyword": "alice",
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "obj-template",
-            "create-or-update",
-            "--title",
-            "tpl",
-            "--dimensions",
-            '[{"title":"dim"}]',
-            "--objective-weight",
-            "--dimension-weight",
-            "--compute-by-weight",
-            "--dry-run",
-        )
-    )
-    assert "create_or_update_obj_template" in output
-    assert dry_run_args(output) == {
-        "title": "tpl",
-        "dimensions": '[{"title":"dim"}]',
-        "objectiveWeight": True,
-        "dimensionWeight": True,
-        "computeByWeight": True,
-    }

auto-test/cli_to_mcp/testcases/chat/test_29_group_notice_and_new_commands.py

@@ -1,163 +0,0 @@
-from test_utils import combined_output, dry_run_args
-
-
-def assert_ok(result):
-    output = combined_output(result)
-    assert result.returncode == 0, output
-    return output
-
-
-def test_group_notice_cli_to_mcp(dws):
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "group",
-            "notice",
-            "create",
-            "--group",
-            "cid123",
-            "--content",
-            "maintenance tonight",
-            "--sticky",
-            "--send-ding",
-            "--dry-run",
-        )
-    )
-    assert "create_group_notice" in output
-    assert dry_run_args(output) == {
-        "openConversationId": "cid123",
-        "content": "maintenance tonight",
-        "sticky": True,
-        "sendDing": True,
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "group",
-            "notice",
-            "edit",
-            "--group",
-            "cid123",
-            "--notice-id",
-            "notice123",
-            "--content",
-            "updated",
-            "--dry-run",
-        )
-    )
-    assert "edit_group_notice" in output
-    assert dry_run_args(output) == {
-        "openConversationId": "cid123",
-        "dataId": "notice123",
-        "content": "updated",
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "group",
-            "notice",
-            "get",
-            "--group",
-            "cid123",
-            "--notice-id",
-            "notice123",
-            "--dry-run",
-        )
-    )
-    assert "get_group_notice" in output
-    assert dry_run_args(output) == {
-        "openConversationId": "cid123",
-        "dataId": "notice123",
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "group",
-            "notice",
-            "list",
-            "--group",
-            "cid123",
-            "--limit",
-            "20",
-            "--cursor",
-            "next",
-            "--scheduled",
-            "--dry-run",
-        )
-    )
-    assert "list_group_notices" in output
-    assert dry_run_args(output) == {
-        "openConversationId": "cid123",
-        "limit": 20,
-        "cursor": "next",
-        "scheduled": True,
-    }
-
-
-def test_chat_misc_new_commands_cli_to_mcp(dws):
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "group",
-            "share-invite",
-            "--source",
-            "sourceCid",
-            "--target",
-            "targetCid",
-            "--expires-seconds",
-            "3600",
-            "--uuid",
-            "uuid-1",
-            "--dry-run",
-        )
-    )
-    assert "share_group_invite_url" in output
-    assert dry_run_args(output) == {
-        "sourceOpenConversationId": "sourceCid",
-        "targetOpenConversationId": "targetCid",
-        "expiresSeconds": 3600,
-        "uuid": "uuid-1",
-    }
-
-    output = assert_ok(
-        dws.run_raw("chat", "text", "translate", "--query", "hello", "--to", "zh_CN", "--dry-run")
-    )
-    assert "translate" in output
-    assert dry_run_args(output) == {"query": "hello", "to": "zh_CN"}
-
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "category",
-            "create-smart",
-            "--name",
-            "priority",
-            "--keywords",
-            "alpha,beta",
-            "--members",
-            "uid1,uid2",
-            "--dry-run",
-        )
-    )
-    assert "create_smart_conv_category" in output
-    assert dry_run_args(output) == {
-        "title": "priority",
-        "keywords": ["alpha", "beta"],
-        "memberOpenDingTalkIds": ["uid1", "uid2"],
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "message",
-            "list-emotion-replies",
-            "--msg-ids",
-            "msg1,msg2",
-            "--dry-run",
-        )
-    )
-    assert "list_message_emotion_replies" in output
-    assert dry_run_args(output) == {"openMessageIds": ["msg1", "msg2"]}

auto-test/cli_to_mcp/testcases/conftest.py

@@ -1,8 +0,0 @@
-import pytest
-
-from test_utils import DWSRunner
-
-
-@pytest.fixture(scope="session")
-def dws():
-    return DWSRunner()

auto-test/cli_to_mcp/testcases/doc/test_12_import.py

@@ -1,54 +0,0 @@
-from test_utils import combined_output
-
-
-def assert_ok(result):
-    output = combined_output(result)
-    assert result.returncode == 0, output
-    return output
-
-
-def test_doc_import_help_and_validation(dws, tmp_path):
-    output = assert_ok(dws.run_raw("doc", "import", "--help"))
-    assert "dws doc import" in output
-    assert "--file string" in output
-    assert "--workspace string" in output
-    assert "--name string" in output
-
-    result = dws.run_raw("doc", "import", "--file", str(tmp_path / "missing.md"), "--dry-run")
-    output = combined_output(result)
-    assert result.returncode != 0
-    assert "cannot read file" in output
-
-    bad = tmp_path / "bad.exe"
-    bad.write_text("bad", encoding="utf-8")
-    result = dws.run_raw("doc", "import", "--file", str(bad), "--dry-run")
-    output = combined_output(result)
-    assert result.returncode != 0
-    assert "unsupported file format" in output
-
-
-def test_doc_import_dry_run(dws, tmp_path):
-    source = tmp_path / "sample.md"
-    source.write_text("# Sample\n\nhello\n", encoding="utf-8")
-
-    output = assert_ok(
-        dws.run_raw(
-            "doc",
-            "import",
-            "--file",
-            str(source),
-            "--name",
-            "Imported Sample",
-            "--workspace",
-            "workspace123",
-            "--dry-run",
-        )
-    )
-    assert "Imported Sample" in output
-    assert "sample.md" in output
-    assert "md" in output
-
-
-def test_doc_import_get_dry_run(dws):
-    output = assert_ok(dws.run_raw("doc", "import", "get", "--task-id", "task123", "--dry-run"))
-    assert "task123" in output

auto-test/cli_to_mcp/testcases/mail/test_11_mail_new_commands.py

@@ -1,143 +0,0 @@
-import os
-
-from test_utils import combined_output, dry_run_args
-
-
-def mail_email() -> str:
-    return os.environ.get("DINGTALK_MAIL_EMAIL", "user@example.com")
-
-
-def assert_ok(result):
-    output = combined_output(result)
-    assert result.returncode == 0, output
-    return output
-
-
-def assert_fails(result, expected: str):
-    output = combined_output(result)
-    assert result.returncode != 0, output
-    assert expected in output
-
-
-def test_mailbox_profile_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "mailbox", "profile", "--help"))
-    assert "dws mail mailbox profile" in output
-    assert "--email string" in output
-
-    assert_fails(dws.run_raw("mail", "mailbox", "profile"), "email")
-
-    output = assert_ok(
-        dws.run_raw("mail", "mailbox", "profile", "--email", mail_email(), "--dry-run")
-    )
-    assert "get_mailbox_profile" in output
-    assert dry_run_args(output) == {"email": mail_email()}
-
-
-def test_message_batch_get_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "message", "batch-get", "--help"))
-    assert "dws mail message batch-get" in output
-    assert "--email string" in output
-    assert "--ids string" in output
-
-    assert_fails(
-        dws.run_raw("mail", "message", "batch-get", "--email", mail_email()),
-        "ids",
-    )
-
-    too_many_ids = ",".join(f"msg_{i:02d}" for i in range(21))
-    assert_fails(
-        dws.run_raw(
-            "mail",
-            "message",
-            "batch-get",
-            "--email",
-            mail_email(),
-            "--ids",
-            too_many_ids,
-            "--dry-run",
-        ),
-        "20",
-    )
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "message",
-            "batch-get",
-            "--email",
-            mail_email(),
-            "--ids",
-            "msg_001,msg_002",
-            "--dry-run",
-        )
-    )
-    assert "get_email_by_message_id" in output
-    assert "msg_001" in output
-    assert "msg_002" in output
-
-
-def test_sent_message_recall_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "sent-message", "recall", "--help"))
-    assert "dws mail sent-message recall" in output
-    assert "--subject string" in output
-    assert "--yes" in output
-
-    assert_fails(
-        dws.run_raw(
-            "mail",
-            "sent-message",
-            "recall",
-            "--email",
-            mail_email(),
-            "--id",
-            "msg_001",
-            "--subject",
-            "subject",
-        ),
-        "--yes",
-    )
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "sent-message",
-            "recall",
-            "--email",
-            mail_email(),
-            "--id",
-            "msg_001",
-            "--subject",
-            "subject",
-            "--yes",
-            "--dry-run",
-        )
-    )
-    assert "recall_sent_message" in output
-    assert dry_run_args(output) == {
-        "email": mail_email(),
-        "id": "msg_001",
-        "subject": "subject",
-    }
-
-
-def test_sent_message_recall_detail_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "sent-message", "recall-detail", "--help"))
-    assert "dws mail sent-message recall-detail" in output
-    assert "--email string" in output
-    assert "--id string" in output
-    assert "FINISHED" in output
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "sent-message",
-            "recall-detail",
-            "--email",
-            mail_email(),
-            "--id",
-            "task_001",
-            "--dry-run",
-        )
-    )
-    assert "get_recall_detail" in output
-    assert dry_run_args(output) == {"email": mail_email(), "id": "task_001"}

auto-test/cli_to_mcp/testcases/mail/test_12_mail_allowlist_blocklist_autoreply.py

@@ -1,133 +0,0 @@
-import os
-
-from test_utils import combined_output, dry_run_args
-
-
-def mail_email() -> str:
-    return os.environ.get("DINGTALK_MAIL_EMAIL", "user@example.com")
-
-
-def assert_ok(result):
-    output = combined_output(result)
-    assert result.returncode == 0, output
-    return output
-
-
-def test_auto_reply_update_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "auto-reply", "update", "--help"))
-    for flag in ("--email string", "--enabled string", "--start string", "--end string", "--scope string", "--content string"):
-        assert flag in output
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "auto-reply",
-            "update",
-            "--email",
-            mail_email(),
-            "--enabled",
-            "true",
-            "--start",
-            "2026/07/01 09:00:00 +0800",
-            "--end",
-            "2026/07/07 18:00:00 +0800",
-            "--scope",
-            "all",
-            "--content",
-            "out of office",
-            "--dry-run",
-        )
-    )
-    assert "update_auto_reply" in output
-    assert dry_run_args(output) == {
-        "email": mail_email(),
-        "enabled": True,
-        "startTime": "2026/07/01 09:00:00 +0800",
-        "endTime": "2026/07/07 18:00:00 +0800",
-        "scope": "all",
-        "content": "out of office",
-    }
-
-
-def test_allow_list_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "allow-list", "list", "--email", mail_email(), "--dry-run"))
-    assert "list_mailbox_allowlist" in output
-    assert dry_run_args(output) == {"email": mail_email()}
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "allow-list",
-            "add",
-            "--email",
-            mail_email(),
-            "--entries",
-            "partner@example.com,@example.org",
-            "--dry-run",
-        )
-    )
-    assert "add_mailbox_allowlist" in output
-    assert dry_run_args(output) == {
-        "email": mail_email(),
-        "entries": ["partner@example.com", "@example.org"],
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "allow-list",
-            "remove",
-            "--email",
-            mail_email(),
-            "--entries",
-            "partner@example.com",
-            "--dry-run",
-        )
-    )
-    assert "remove_mailbox_allowlist" in output
-    assert dry_run_args(output) == {
-        "email": mail_email(),
-        "entries": ["partner@example.com"],
-    }
-
-
-def test_block_list_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "block-list", "list", "--email", mail_email(), "--dry-run"))
-    assert "list_mailbox_blocklist" in output
-    assert dry_run_args(output) == {"email": mail_email()}
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "block-list",
-            "add",
-            "--email",
-            mail_email(),
-            "--entries",
-            "spam@example.com,@junk.example",
-            "--dry-run",
-        )
-    )
-    assert "add_mailbox_blocklist" in output
-    assert dry_run_args(output) == {
-        "email": mail_email(),
-        "entries": ["spam@example.com", "@junk.example"],
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "block-list",
-            "remove",
-            "--email",
-            mail_email(),
-            "--entries",
-            "spam@example.com",
-            "--dry-run",
-        )
-    )
-    assert "remove_mailbox_blocklist" in output
-    assert dry_run_args(output) == {
-        "email": mail_email(),
-        "entries": ["spam@example.com"],
-    }

auto-test/cli_to_mcp/testcases/sheet/test_36_group_dimension.py

@@ -1,80 +0,0 @@
-from test_utils import combined_output, dry_run_args
-
-
-def assert_ok(result):
-    output = combined_output(result)
-    assert result.returncode == 0, output
-    return output
-
-
-def test_group_dimension_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("sheet", "group-dimension", "--help"))
-    assert "dws sheet group-dimension" in output
-    assert "--group-state string" in output
-
-    output = assert_ok(
-        dws.run_raw(
-            "sheet",
-            "group-dimension",
-            "--node",
-            "node123",
-            "--sheet-id",
-            "Sheet1",
-            "--range",
-            "3:7",
-            "--group-state",
-            "fold",
-            "--dry-run",
-        )
-    )
-    assert "group_dimension" in output
-    assert dry_run_args(output) == {
-        "nodeId": "node123",
-        "sheetId": "Sheet1",
-        "range": "3:7",
-        "groupState": "fold",
-    }
-
-
-def test_ungroup_dimension_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("sheet", "ungroup-dimension", "--help"))
-    assert "dws sheet ungroup-dimension" in output
-
-    output = assert_ok(
-        dws.run_raw(
-            "sheet",
-            "ungroup-dimension",
-            "--node",
-            "node123",
-            "--sheet-id",
-            "Sheet1",
-            "--range",
-            "C:F",
-            "--dry-run",
-        )
-    )
-    assert "ungroup_dimension" in output
-    assert dry_run_args(output) == {
-        "nodeId": "node123",
-        "sheetId": "Sheet1",
-        "range": "C:F",
-    }
-
-
-def test_group_dimension_rejects_invalid_state(dws):
-    result = dws.run_raw(
-        "sheet",
-        "group-dimension",
-        "--node",
-        "node123",
-        "--sheet-id",
-        "Sheet1",
-        "--range",
-        "3:7",
-        "--group-state",
-        "invalid",
-        "--dry-run",
-    )
-    output = combined_output(result)
-    assert result.returncode != 0
-    assert "group-state" in output

auto-test/cli_to_mcp/testcases/test_utils/__init__.py

@@ -1,58 +0,0 @@
-import json
-import os
-import re
-import shlex
-import subprocess
-from pathlib import Path
-
-
-def repo_root(start_file: str) -> Path:
-    current = Path(start_file).resolve()
-    for parent in [current, *current.parents]:
-        if (parent / "go.mod").exists():
-            return parent
-    raise RuntimeError(f"cannot locate repo root from {start_file}")
-
-
-def resolve_dws_cmd(start_file: str) -> list[str]:
-    root = repo_root(start_file)
-    if env_bin := os.environ.get("DWS_BIN"):
-        return shlex.split(env_bin)
-
-    for rel in ("dws", "build/dws", "bin/dws", "dingtalk-workspace-cli"):
-        candidate = root / rel
-        if candidate.exists() and os.access(candidate, os.X_OK):
-            return [str(candidate)]
-
-    return ["go", "run", "./cmd"]
-
-
-def combined_output(result: subprocess.CompletedProcess) -> str:
-    return (result.stdout or "") + (result.stderr or "")
-
-
-def dry_run_args(output: str) -> dict:
-    match = re.search(r"Arguments:\s*(\{.*\})", output, re.S)
-    assert match, f"dry-run output does not contain Arguments JSON: {output}"
-    return json.loads(match.group(1))
-
-
-class DWSRunner:
-    def __init__(self):
-        self.root = repo_root(__file__)
-        self.cmd = resolve_dws_cmd(__file__)
-
-    def run_raw(self, *args: str, timeout: int = 45) -> subprocess.CompletedProcess:
-        return subprocess.run(
-            [*self.cmd, *args],
-            cwd=self.root,
-            text=True,
-            capture_output=True,
-            timeout=timeout,
-        )
-
-    def run(self, *args: str, timeout: int = 45):
-        result = self.run_raw(*args, timeout=timeout)
-        output = combined_output(result)
-        assert result.returncode == 0, output
-        return json.loads(result.stdout)

build/homebrew-release.rb.tmpl

@@ -0,0 +1,63 @@
+class __CLASS_NAME__ < Formula
+  desc "__DESCRIPTION__"
+  homepage "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli"
+  version "__VERSION__"
+  license "Apache-2.0"
+__KEG_ONLY_LINE__
+
+  on_macos do
+    if Hardware::CPU.arm?
+      url "__DARWIN_ARM64_URL__"
+      sha256 "__DARWIN_ARM64_SHA256__"
+    else
+      url "__DARWIN_AMD64_URL__"
+      sha256 "__DARWIN_AMD64_SHA256__"
+    end
+  end
+
+  on_linux do
+    if Hardware::CPU.arm?
+      url "__LINUX_ARM64_URL__"
+      sha256 "__LINUX_ARM64_SHA256__"
+    else
+      url "__LINUX_AMD64_URL__"
+      sha256 "__LINUX_AMD64_SHA256__"
+    end
+  end
+
+  resource "skills" do
+    url "__SKILLS_URL__"
+    sha256 "__SKILLS_SHA256__"
+  end
+
+  def install
+    root = Dir["dws-*"].find { |entry| File.directory?(entry) } || "."
+    binary = File.join(root, "dws")
+    raise "binary not found: #{binary}" unless File.exist?(binary)
+
+    bin.install binary => "dws"
+
+    %w[LICENSE NOTICE README.md CHANGELOG.md].each do |name|
+      source = File.join(root, name)
+      pkgshare.install source if File.exist?(source)
+    end
+
+    skill_dest = pkgshare/"skills/dws"
+    skill_dest.mkpath
+    resource("skills").stage do
+      cp_r(Dir["*"], skill_dest)
+    end
+  end
+
+  def caveats
+    <<~EOS
+      Agent Skills are bundled in #{pkgshare}/skills/dws.
+      Run `dws skill setup` to install them into your Agent directories.
+__CHANNEL_CAVEAT__
+    EOS
+  end
+
+  test do
+    assert_match version.to_s, shell_output("#{bin}/dws version")
+  end
+end

build/homebrew.rb.tmpl

@@ -1,5 +1,5 @@
 class __CLASS_NAME__ < Formula
-  desc "DingTalk Workspace CLI"
+  desc "Install locally built DingTalk workspace CLI artifacts for verification"
   homepage "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli"
   url "__ARCHIVE_URL__"
   sha256 "__ARCHIVE_SHA256__"
@@ -12,8 +12,6 @@ __KEG_ONLY_LINE__
   end
 
   def install
-    require "fileutils"
-
     root = Dir["dws-*"].find { |entry| File.directory?(entry) } || "."
     binary = File.join(root, "dws")
     raise "binary not found: #{binary}" unless File.exist?(binary)
@@ -28,44 +26,15 @@ __KEG_ONLY_LINE__
     skill_dest = pkgshare/"skills/dws"
     skill_dest.mkpath
     resource("skills").stage do
-      FileUtils.cp_r(Dir["*"], skill_dest)
+      cp_r(Dir["*"], skill_dest)
     end
   end
 
-  def post_install
-    require "fileutils"
-
-    skill_root = pkgshare/"skills/dws"
-    entries = Dir["#{skill_root}/*"]
-    return if entries.empty?
-
-    targets = [
-      Pathname.new(File.join(Dir.home, ".agents/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".claude/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".cursor/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".qoder/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".qoderwork/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".gemini/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".codex/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".github/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".windsurf/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".augment/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".cline/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".amp/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".kiro/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".trae/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".openclaw/skills/dws")),
-      Pathname.new(File.join(Dir.home, ".hermes/skills/dws")),
-    ]
-
-    targets.each_with_index do |dest, index|
-      parent_gate = dest.parent.parent
-      next if index > 0 && !parent_gate.directory?
-
-      FileUtils.rm_rf(dest)
-      FileUtils.mkdir_p(dest)
-      FileUtils.cp_r(entries, dest)
-    end
+  def caveats
+    <<~EOS
+      Agent Skills are bundled in #{pkgshare}/skills/dws.
+      Run `dws skill setup` to install them into your Agent directories.
+    EOS
   end
 
   test do

cmd/fetch_mcp_metadata/main.go

@@ -0,0 +1,413 @@
+// Command fetch_mcp_metadata pulls tools/list from ALL live MCP server endpoints
+// and writes a local diagnostic dump. It is NOT a Schema delivery refresh:
+// schema_mcp_metadata.json is retired; production Catalog assembles from
+// Contract/ParamDecl/Interface + Cobra only.
+//
+// Usage:
+//
+//	dws auth login                     # ensure valid auth
+//	make fetch-mcp-metadata             # writes artifacts/mcp_metadata_diagnostic.json
+//
+// The tool loads auth from the DWS keychain, iterates static server endpoints
+// (internal/syncdata.StaticServers), calls tools/list on each, merges results,
+// and writes the requested -output path (refuses the retired pin path).
+package main
+
+import (
+	"context"
+	"encoding/json"
+	"flag"
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"sort"
+	"strings"
+	"time"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/app"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/auth"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/cli"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/syncdata"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/transport"
+)
+
+// toolLister is the tools/list capability consumed by run; production code
+// uses transport.Client, tests inject fakes.
+type toolLister interface {
+	ListTools(ctx context.Context, endpoint string) (transport.ToolsListResult, error)
+}
+
+// Injection points so run() is fully testable without network/keychain/exit.
+var (
+	osExit               = os.Exit
+	getenv               = os.Getenv
+	loadTokenData        = auth.LoadTokenDataKeychain
+	staticServers        = syncdata.StaticServers
+	registrySource       = collectedIdentityInterfaceRefs
+	collectIdentitySpecs = cli.CollectIdentitySpecs
+	listToolsTimeout     = 30 * time.Second
+	gitHeadPath          = ".git/HEAD"
+	newToolLister        = func(token string) toolLister {
+		return transport.NewClient(&http.Client{Timeout: 60 * time.Second}).WithAuth(token, nil)
+	}
+)
+
+func main() {
+	osExit(run(os.Args[1:], os.Stderr))
+}
+
+func run(args []string, stderr io.Writer) int {
+	flags := flag.NewFlagSet("fetch_mcp_metadata", flag.ContinueOnError)
+	flags.SetOutput(stderr)
+	output := flags.String("output", "artifacts/mcp_metadata_diagnostic.json", "diagnostic dump path (not a Schema pin)")
+	if err := flags.Parse(args); err != nil {
+		return 2
+	}
+	if retiredPinnedMCPMetadataPath(*output) {
+		fmt.Fprintln(stderr, "fetch_mcp_metadata: refusing to write retired Schema pin internal/cli/schema_mcp_metadata.json")
+		return 2
+	}
+
+	token := resolveToken(stderr)
+	if token == "" {
+		fmt.Fprintln(stderr, "fetch_mcp_metadata: no auth token. Run 'dws auth login' first.")
+		return 1
+	}
+
+	client := newToolLister(token)
+
+	// Iterate ALL static server endpoints (26 servers covering all products).
+	servers := staticServers()
+	fmt.Fprintf(stderr, "fetch_mcp_metadata: querying %d server endpoints\n", len(servers))
+
+	// Collect command identity to build tool_name → interface_ref mapping.
+	registryMap := loadRegistryInterfaceRefs(stderr)
+	fmt.Fprintf(stderr, "fetch_mcp_metadata: registry mapping: %d entries\n", len(registryMap))
+
+	// Load a previous diagnostic dump (if any) to preserve hand-curated
+	// cross-server interface_ref mappings that automated matching can't derive.
+	prevData, prevErr := os.ReadFile(*output)
+	prevTools := map[string]map[string]any{}
+	if prevErr == nil {
+		var prev struct {
+			Tools map[string]map[string]any `json:"tools"`
+		}
+		if json.Unmarshal(prevData, &prev) == nil {
+			prevTools = prev.Tools
+		}
+	}
+
+	// Start from previous data (preserves cross-server refs), then overwrite
+	// with fresh MCP data where available.
+	allTools := make(map[string]map[string]any)
+	for k, v := range prevTools {
+		allTools[k] = v
+	}
+
+	// Reviewed cross-server interface_refs live only in the previous snapshot
+	// (the registry stores canonical paths, not MCP identities). Build a
+	// live-key → canonicals index so those tools get refreshed instead of
+	// being skipped and frozen at the previous snapshot forever.
+	crossRefs := buildCrossServerRefs(prevTools, registryMap)
+	if len(crossRefs) > 0 {
+		fmt.Fprintf(stderr, "fetch_mcp_metadata: cross-server ref index: %d live keys\n", len(crossRefs))
+	}
+	// Canonicals with a reviewed cross-server identity must only be fed by
+	// that identity; a same-named tool on another server is a coincidence,
+	// not a data source.
+	crossOwned := map[string]bool{}
+	for _, canonicals := range crossRefs {
+		for _, canonical := range canonicals {
+			crossOwned[canonical] = true
+		}
+	}
+	totalRaw := 0
+	failedServices := []string{}
+
+	for _, srv := range servers {
+		endpoint := strings.TrimSpace(srv.Endpoint)
+		if endpoint == "" {
+			continue
+		}
+		ctx, cancel := context.WithTimeout(context.Background(), listToolsTimeout)
+		result, err := client.ListTools(ctx, endpoint)
+		cancel()
+		if err != nil {
+			fmt.Fprintf(stderr, "  [skip] %s: %v\n", srv.ID, err)
+			failedServices = append(failedServices, srv.ID)
+			continue
+		}
+		fmt.Fprintf(stderr, "  [ok]   %s: %d tools\n", srv.ID, len(result.Tools))
+		totalRaw += len(result.Tools)
+		for _, tool := range result.Tools {
+			name := strings.TrimSpace(tool.Name)
+			if name == "" {
+				continue
+			}
+			// Direct match: CLI canonical equals server-prefixed tool name
+			// (e.g., "doc.copy_document"). Cross-owned canonicals are skipped
+			// here — their reviewed identity feeds them below.
+			canonicalKey := srv.ID + "." + name
+			if ref, hasRef := registryMap[canonicalKey]; hasRef && !crossOwned[canonicalKey] {
+				mergeLiveMCPTool(allTools, canonicalKey, tool, ref)
+			}
+			// Cross-server match: registry canonicals whose reviewed
+			// interface_ref points at this live tool (one live tool may feed
+			// several canonicals, e.g. advperm_enable/disable → set_advanced_permission).
+			for _, canonical := range crossRefs[canonicalKey] {
+				mergeLiveMCPTool(allTools, canonical, tool, registryMap[canonical])
+			}
+		}
+	}
+
+	matched := 0
+	for _, t := range allTools {
+		if _, ok := t["interface_ref"]; ok {
+			matched++
+		}
+	}
+	fmt.Fprintf(stderr, "fetch_mcp_metadata: MCP matched=%d, with interface_ref=%d\n", len(allTools), matched)
+
+	// Fill gaps: for registry canonicals not covered by MCP tools/list OR
+	// previous data, add stub entries (interface_ref only).
+	stubs := 0
+	for canonicalKey, ref := range registryMap {
+		if _, exists := allTools[canonicalKey]; exists {
+			continue
+		}
+		allTools[canonicalKey] = map[string]any{
+			"interface_ref": ref,
+		}
+		stubs++
+	}
+	if stubs > 0 {
+		fmt.Fprintf(stderr, "fetch_mcp_metadata: added %d registry stubs (no MCP data, interface_ref only)\n", stubs)
+	}
+
+	// Compute coverage fields required by check-schema-catalog.sh. Failed
+	// services must be reported honestly so policy can spot snapshot gaps.
+	if len(failedServices) > 0 {
+		fmt.Fprintf(stderr, "fetch_mcp_metadata: %d/%d services unreachable: %s\n",
+			len(failedServices), len(servers), strings.Join(failedServices, ", "))
+	}
+
+	metadata := map[string]any{
+		"version":  1,
+		"source":   "mcp-tools-list+cli-registry",
+		"coverage": buildCoverage(len(servers), failedServices, totalRaw, len(allTools), stubs),
+		"tools":    allTools,
+	}
+
+	// source_revision: git commit hash (proves provenance).
+	if rev, err := os.ReadFile(gitHeadPath); err == nil {
+		metadata["source_revision"] = strings.TrimSpace(string(rev))
+	}
+
+	if err := writeMetadata(*output, metadata); err != nil {
+		fmt.Fprintf(stderr, "fetch_mcp_metadata: %v\n", err)
+		return 1
+	}
+	fmt.Fprintf(stderr, "fetch_mcp_metadata: wrote %d tools to %s\n", len(allTools), *output)
+	return 0
+}
+
+// resolveToken returns the access token from DWS_ACCESS_TOKEN or, as a
+// fallback, the DWS keychain.
+func resolveToken(stderr io.Writer) string {
+	token := strings.TrimSpace(getenv("DWS_ACCESS_TOKEN"))
+	if token != "" {
+		return token
+	}
+	td, err := loadTokenData()
+	if err != nil || td == nil || td.AccessToken == "" {
+		return ""
+	}
+	fmt.Fprintf(stderr, "fetch_mcp_metadata: loaded token from keychain (%d chars)\n", len(td.AccessToken))
+	return td.AccessToken
+}
+
+// writeMetadata marshals the snapshot and writes it to the output path.
+func writeMetadata(path string, metadata map[string]any) error {
+	data, err := json.MarshalIndent(metadata, "", "  ")
+	if err != nil {
+		return fmt.Errorf("marshal failed: %w", err)
+	}
+	data = append(data, '\n')
+	if err := os.WriteFile(path, data, 0644); err != nil {
+		return fmt.Errorf("write %s failed: %w", path, err)
+	}
+	return nil
+}
+
+// buildCoverage reports snapshot coverage honestly: snapshot_services only
+// counts services whose tools/list succeeded, missing_services names the
+// failures, and matched_tools excludes registry stubs (entries carrying no
+// live MCP metadata) so a stub-heavy snapshot cannot claim full matching.
+func buildCoverage(sourceServices int, failedServices []string, sourceTools, surfaceTools, stubs int) map[string]any {
+	missing := failedServices
+	if missing == nil {
+		missing = []string{}
+	}
+	return map[string]any{
+		"surface_scope":     "source_revision",
+		"source_services":   sourceServices,
+		"snapshot_services": sourceServices - len(missing),
+		"missing_services":  missing,
+		"source_tools":      sourceTools,
+		"surface_tools":     surfaceTools,
+		"matched_tools":     surfaceTools - stubs,
+		"aliased_tools":     0,
+		"unmatched_tools":   stubs,
+	}
+}
+
+// mergeLiveMCPTool replaces stale live-derived fields while retaining an
+// existing reviewed interface_ref. Some CLI canonicals intentionally route to
+// a differently named product/RPC, so the previous cross-server mapping must
+// survive even though title, description, and parameters are refreshed.
+func mergeLiveMCPTool(allTools map[string]map[string]any, canonicalKey string, tool transport.ToolDescriptor, fallbackRef map[string]string) {
+	interfaceRef := any(fallbackRef)
+	if previous := allTools[canonicalKey]; previous != nil {
+		if reviewedRef, ok := previous["interface_ref"]; ok && reviewedRef != nil {
+			interfaceRef = reviewedRef
+		}
+	}
+
+	entry := map[string]any{
+		"title":         tool.Title,
+		"description":   tool.Description,
+		"interface_ref": interfaceRef,
+	}
+	if tool.InputSchema != nil {
+		entry["parameters"] = extractParams(tool.InputSchema)
+	}
+	allTools[canonicalKey] = entry
+}
+
+// buildCrossServerRefs indexes reviewed cross-server mappings from the
+// previous snapshot: for every registry canonical whose interface_ref names a
+// different MCP identity (product_id.rpc_name != canonical), the live key is
+// mapped back to that canonical. One live tool may serve several canonicals,
+// so values are slices, sorted for deterministic merge order.
+func buildCrossServerRefs(prevTools map[string]map[string]any, registryMap map[string]map[string]string) map[string][]string {
+	index := map[string][]string{}
+	for canonical, entry := range prevTools {
+		if _, inRegistry := registryMap[canonical]; !inRegistry {
+			continue
+		}
+		ref, ok := entry["interface_ref"].(map[string]any)
+		if !ok {
+			continue
+		}
+		productID, _ := ref["product_id"].(string)
+		rpcName, _ := ref["rpc_name"].(string)
+		if productID == "" || rpcName == "" {
+			continue
+		}
+		liveKey := productID + "." + rpcName
+		if liveKey == canonical {
+			continue
+		}
+		index[liveKey] = append(index[liveKey], canonical)
+	}
+	for _, canonicals := range index {
+		sort.Strings(canonicals)
+	}
+	return index
+}
+
+// collectedIdentityInterfaceRefs collects command identity from the live
+// command tree — the replacement for the retired reviewed CommandRegistry —
+// and derives the canonical_path → {product_id, rpc_name} mapping used for
+// interface_ref injection.
+func collectedIdentityInterfaceRefs() (map[string]map[string]string, error) {
+	root := app.NewSchemaSourceRootCommand()
+	specs, _, err := collectIdentitySpecs(root)
+	if err != nil {
+		return nil, fmt.Errorf("collect command identity: %w", err)
+	}
+	out := make(map[string]map[string]string, len(specs))
+	for _, spec := range specs {
+		cp := strings.TrimSpace(spec.CanonicalPath)
+		if cp == "" || !strings.Contains(cp, ".") {
+			continue
+		}
+		parts := strings.SplitN(cp, ".", 2)
+		out[cp] = map[string]string{
+			"product_id": parts[0],
+			"rpc_name":   parts[1],
+		}
+	}
+	return out, nil
+}
+
+// loadRegistryInterfaceRefs builds the canonical_path → interface_ref mapping
+// from the collected command identity. 与旧实现同等告警：静默返回空映射会让
+// 所有 live tool 被丢弃、产出 stub-only 快照且零提示（P1#1 的故障模式）。
+func loadRegistryInterfaceRefs(stderr io.Writer) map[string]map[string]string {
+	refs, err := registrySource()
+	if err != nil {
+		fmt.Fprintf(stderr, "fetch_mcp_metadata: warning: cannot collect command identity: %v\n", err)
+		return map[string]map[string]string{}
+	}
+	return refs
+}
+
+func retiredPinnedMCPMetadataPath(path string) bool {
+	cleaned := strings.ReplaceAll(strings.TrimSpace(path), "\\", "/")
+	return cleaned == "internal/cli/schema_mcp_metadata.json" ||
+		strings.HasSuffix(cleaned, "/internal/cli/schema_mcp_metadata.json")
+}
+
+// extractParams converts a JSON Schema inputSchema (from MCP tools/list) into
+// the flat param-name → metadata map used by diagnostic dumps.
+func extractParams(inputSchema map[string]any) map[string]map[string]any {
+	if inputSchema == nil {
+		return nil
+	}
+	properties, ok := inputSchema["properties"].(map[string]any)
+	if !ok {
+		return nil
+	}
+	requiredSet := map[string]bool{}
+	if req, ok := inputSchema["required"].([]any); ok {
+		for _, r := range req {
+			if s, ok := r.(string); ok {
+				requiredSet[s] = true
+			}
+		}
+	}
+
+	params := make(map[string]map[string]any, len(properties))
+	for name, raw := range properties {
+		prop, ok := raw.(map[string]any)
+		if !ok {
+			continue
+		}
+		meta := map[string]any{}
+		if t, ok := prop["type"].(string); ok {
+			meta["type"] = t
+		}
+		if d, ok := prop["description"].(string); ok {
+			meta["description"] = d
+		}
+		if d, ok := prop["default"].(string); ok {
+			meta["default"] = d
+		}
+		if e, ok := prop["enum"].([]any); ok {
+			enums := make([]string, 0, len(e))
+			for _, v := range e {
+				if s, ok := v.(string); ok {
+					enums = append(enums, s)
+				}
+			}
+			if len(enums) > 0 {
+				meta["enum"] = enums
+			}
+		}
+		meta["required"] = requiredSet[name]
+		params[name] = meta
+	}
+	return params
+}

cmd/fetch_mcp_metadata/main_test.go

@@ -0,0 +1,612 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0 (the "License");
+
+package main
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"math"
+	"os"
+	"path/filepath"
+	"reflect"
+	"strings"
+	"testing"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/auth"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/cli"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/syncdata"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/transport"
+	"github.com/spf13/cobra"
+)
+
+func TestCrossPlatformCoverageLoadRegistryInterfaceRefsCollectsIdentity(t *testing.T) {
+	var stderr bytes.Buffer
+	refs := loadRegistryInterfaceRefs(&stderr)
+	if len(refs) == 0 {
+		t.Fatal("loadRegistryInterfaceRefs() returned no collected commands")
+	}
+
+	got, ok := refs["calendar.list_calendars"]
+	if !ok {
+		t.Fatal("calendar.list_calendars missing from collected command identity")
+	}
+	if got["product_id"] != "calendar" || got["rpc_name"] != "list_calendars" {
+		t.Fatalf("calendar.list_calendars ref = %#v", got)
+	}
+
+	direct, err := collectedIdentityInterfaceRefs()
+	if err != nil {
+		t.Fatalf("collectedIdentityInterfaceRefs() error = %v", err)
+	}
+	if len(direct) == 0 || direct["calendar.list_calendars"]["rpc_name"] != "list_calendars" {
+		t.Fatalf("collectedIdentityInterfaceRefs() = %#v", direct["calendar.list_calendars"])
+	}
+
+	prevRegistry := registrySource
+	prevCollect := collectIdentitySpecs
+	t.Cleanup(func() {
+		registrySource = prevRegistry
+		collectIdentitySpecs = prevCollect
+	})
+	registrySource = func() (map[string]map[string]string, error) {
+		return nil, errors.New("collect boom")
+	}
+	stderr.Reset()
+	if got := loadRegistryInterfaceRefs(&stderr); len(got) != 0 || !strings.Contains(stderr.String(), "cannot collect command identity") {
+		t.Fatalf("loadRegistryInterfaceRefs error path = %#v stderr=%q", got, stderr.String())
+	}
+
+	collectIdentitySpecs = func(*cobra.Command) ([]cli.CommandSpec, cli.IdentityCollectionReport, error) {
+		return nil, cli.IdentityCollectionReport{}, errors.New("walk boom")
+	}
+	if _, err := collectedIdentityInterfaceRefs(); err == nil || !strings.Contains(err.Error(), "collect command identity") {
+		t.Fatalf("collectedIdentityInterfaceRefs wrap error = %v", err)
+	}
+	collectIdentitySpecs = func(*cobra.Command) ([]cli.CommandSpec, cli.IdentityCollectionReport, error) {
+		return []cli.CommandSpec{
+			{CanonicalPath: ""},
+			{CanonicalPath: "nodot"},
+			{CanonicalPath: "doc.create"},
+		}, cli.IdentityCollectionReport{}, nil
+	}
+	gotRefs, err := collectedIdentityInterfaceRefs()
+	if err != nil || len(gotRefs) != 1 || gotRefs["doc.create"]["rpc_name"] != "create" {
+		t.Fatalf("collectedIdentityInterfaceRefs skip = %#v err=%v", gotRefs, err)
+	}
+}
+
+func TestBuildCrossServerRefs(t *testing.T) {
+	registryMap := map[string]map[string]string{
+		"aitable.advperm_enable":  {"product_id": "aitable", "rpc_name": "advperm_enable"},
+		"aitable.advperm_disable": {"product_id": "aitable", "rpc_name": "advperm_disable"},
+		"doc.copy_document":       {"product_id": "doc", "rpc_name": "copy_document"},
+	}
+	prevTools := map[string]map[string]any{
+		// Fan-out: two canonicals share one live tool; insertion order must
+		// not affect the sorted result.
+		"aitable.advperm_enable": {
+			"interface_ref": map[string]any{"product_id": "aitable-helper", "rpc_name": "set_advanced_permission"},
+		},
+		"aitable.advperm_disable": {
+			"interface_ref": map[string]any{"product_id": "aitable-helper", "rpc_name": "set_advanced_permission"},
+		},
+		// Identity ref (live key == canonical) needs no cross entry.
+		"doc.copy_document": {
+			"interface_ref": map[string]any{"product_id": "doc", "rpc_name": "copy_document"},
+		},
+		// Not in the registry: must be ignored.
+		"ghost.tool": {
+			"interface_ref": map[string]any{"product_id": "ghost-helper", "rpc_name": "haunt"},
+		},
+	}
+	got := buildCrossServerRefs(prevTools, registryMap)
+	want := map[string][]string{
+		"aitable-helper.set_advanced_permission": {"aitable.advperm_disable", "aitable.advperm_enable"},
+	}
+	if len(got) != len(want) {
+		t.Fatalf("index = %#v, want %#v", got, want)
+	}
+	for k, v := range want {
+		if gv := got[k]; len(gv) != len(v) || gv[0] != v[0] || gv[1] != v[1] {
+			t.Fatalf("index[%q] = %v, want %v", k, gv, v)
+		}
+	}
+}
+
+func TestBuildCrossServerRefsSkipsMalformedRefs(t *testing.T) {
+	registryMap := map[string]map[string]string{
+		"a.x": {"product_id": "a", "rpc_name": "x"},
+		"a.y": {"product_id": "a", "rpc_name": "y"},
+		"a.z": {"product_id": "a", "rpc_name": "z"},
+	}
+	prevTools := map[string]map[string]any{
+		"a.x": {"interface_ref": "not-a-map"},
+		"a.y": {"interface_ref": map[string]any{"product_id": "", "rpc_name": "r"}},
+		"a.z": {"title": "no ref at all"},
+	}
+	if got := buildCrossServerRefs(prevTools, registryMap); len(got) != 0 {
+		t.Fatalf("index = %#v, want empty", got)
+	}
+}
+
+func TestRunRefreshesCrossServerTools(t *testing.T) {
+	registry := func() (map[string]map[string]string, error) {
+		return map[string]map[string]string{
+			"aitable.advperm_enable":  {"product_id": "aitable", "rpc_name": "advperm_enable"},
+			"aitable.advperm_disable": {"product_id": "aitable", "rpc_name": "advperm_disable"},
+		}, nil
+	}
+	servers := []syncdata.ServerInfo{{ID: "aitable-helper", Endpoint: "https://helper.example"}}
+	lister := &fakeLister{
+		results: map[string]transport.ToolsListResult{
+			"https://helper.example": {Tools: []transport.ToolDescriptor{
+				{Name: "set_advanced_permission", Title: "live title", Description: "live desc"},
+			}},
+		},
+	}
+	stubDeps(t, "env-token", nil, servers, lister, registry)
+
+	output := filepath.Join(t.TempDir(), "snapshot.json")
+	prev := `{"tools":{
+		"aitable.advperm_enable":{"title":"stale","interface_ref":{"product_id":"aitable-helper","rpc_name":"set_advanced_permission"}},
+		"aitable.advperm_disable":{"title":"stale","interface_ref":{"product_id":"aitable-helper","rpc_name":"set_advanced_permission"}}
+	}}`
+	if err := os.WriteFile(output, []byte(prev), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	var stderr bytes.Buffer
+	if code := run([]string{"--output", output}, &stderr); code != 0 {
+		t.Fatalf("run() = %d, stderr=%s", code, stderr.String())
+	}
+	if !strings.Contains(stderr.String(), "cross-server ref index: 1 live keys") {
+		t.Fatalf("stderr = %q, want cross-server index log", stderr.String())
+	}
+
+	data, err := os.ReadFile(output)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var snapshot struct {
+		Tools map[string]map[string]any `json:"tools"`
+	}
+	if err := json.Unmarshal(data, &snapshot); err != nil {
+		t.Fatal(err)
+	}
+	for _, canonical := range []string{"aitable.advperm_enable", "aitable.advperm_disable"} {
+		entry := snapshot.Tools[canonical]
+		if entry["title"] != "live title" || entry["description"] != "live desc" {
+			t.Fatalf("%s = %#v, want live refresh", canonical, entry)
+		}
+		ref := entry["interface_ref"].(map[string]any)
+		if ref["product_id"] != "aitable-helper" || ref["rpc_name"] != "set_advanced_permission" {
+			t.Fatalf("%s reviewed ref lost: %#v", canonical, ref)
+		}
+	}
+}
+
+// TestRunCrossOwnedCanonicalIgnoresNameCoincidence：canonical 拥有评审过的
+// 跨 server 身份时，另一 server 上恰好同名的工具不得直连覆盖其元数据——
+// 数据源只能是评审身份指向的 live 工具。
+func TestRunCrossOwnedCanonicalIgnoresNameCoincidence(t *testing.T) {
+	registry := func() (map[string]map[string]string, error) {
+		return map[string]map[string]string{
+			"aitable.advperm_enable": {"product_id": "aitable", "rpc_name": "advperm_enable"},
+		}, nil
+	}
+	servers := []syncdata.ServerInfo{
+		{ID: "aitable", Endpoint: "https://aitable.example"},
+		{ID: "aitable-helper", Endpoint: "https://helper.example"},
+	}
+	lister := &fakeLister{
+		results: map[string]transport.ToolsListResult{
+			// 同名巧合：aitable server 上恰好也有 advperm_enable。
+			"https://aitable.example": {Tools: []transport.ToolDescriptor{
+				{Name: "advperm_enable", Title: "coincidence title", Description: "coincidence desc"},
+			}},
+			"https://helper.example": {Tools: []transport.ToolDescriptor{
+				{Name: "set_advanced_permission", Title: "owner title", Description: "owner desc"},
+			}},
+		},
+	}
+	stubDeps(t, "env-token", nil, servers, lister, registry)
+
+	output := filepath.Join(t.TempDir(), "snapshot.json")
+	prev := `{"tools":{"aitable.advperm_enable":{"title":"stale","interface_ref":{"product_id":"aitable-helper","rpc_name":"set_advanced_permission"}}}}`
+	if err := os.WriteFile(output, []byte(prev), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	var stderr bytes.Buffer
+	if code := run([]string{"--output", output}, &stderr); code != 0 {
+		t.Fatalf("run() = %d, stderr=%s", code, stderr.String())
+	}
+
+	data, err := os.ReadFile(output)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var snapshot struct {
+		Tools map[string]map[string]any `json:"tools"`
+	}
+	if err := json.Unmarshal(data, &snapshot); err != nil {
+		t.Fatal(err)
+	}
+	entry := snapshot.Tools["aitable.advperm_enable"]
+	if entry["title"] != "owner title" || entry["description"] != "owner desc" {
+		t.Fatalf("entry = %#v, want reviewed-identity source to win over name coincidence", entry)
+	}
+}
+
+func TestMergeLiveMCPToolRefreshesExistingMetadata(t *testing.T) {
+	const canonical = "calendar.list_calendars"
+	reviewedRef := map[string]any{
+		"product_id": "calendar-helper",
+		"rpc_name":   "list_user_calendars",
+	}
+	allTools := map[string]map[string]any{
+		canonical: {
+			"title":         "old title",
+			"description":   "old description",
+			"interface_ref": reviewedRef,
+			"parameters": map[string]any{
+				"stale": map[string]any{"type": "string"},
+			},
+		},
+	}
+	live := transport.ToolDescriptor{
+		Name:        "list_calendars",
+		Title:       "new title",
+		Description: "new description",
+		InputSchema: map[string]any{
+			"type": "object",
+			"properties": map[string]any{
+				"cursor": map[string]any{
+					"type":        "string",
+					"description": "next page cursor",
+				},
+			},
+			"required": []any{"cursor"},
+		},
+	}
+	fallbackRef := map[string]string{
+		"product_id": "calendar",
+		"rpc_name":   "list_calendars",
+	}
+
+	mergeLiveMCPTool(allTools, canonical, live, fallbackRef)
+
+	got := allTools[canonical]
+	if got["title"] != "new title" || got["description"] != "new description" {
+		t.Fatalf("live metadata was not refreshed: %#v", got)
+	}
+	if !reflect.DeepEqual(got["interface_ref"], reviewedRef) {
+		t.Fatalf("interface_ref = %#v, want reviewed mapping %#v", got["interface_ref"], reviewedRef)
+	}
+	params, ok := got["parameters"].(map[string]map[string]any)
+	if !ok {
+		t.Fatalf("parameters type = %T, want refreshed parameter map", got["parameters"])
+	}
+	if _, stale := params["stale"]; stale {
+		t.Fatalf("stale parameter survived refresh: %#v", params)
+	}
+	if cursor := params["cursor"]; cursor["type"] != "string" || cursor["description"] != "next page cursor" || cursor["required"] != true {
+		t.Fatalf("cursor parameter = %#v", cursor)
+	}
+}
+
+func TestBuildCoverageReportsFailedServices(t *testing.T) {
+	got := buildCoverage(26, []string{"doc", "sheet"}, 800, 813, 40)
+	if got["source_services"] != 26 {
+		t.Fatalf("source_services = %v, want 26", got["source_services"])
+	}
+	if got["snapshot_services"] != 24 {
+		t.Fatalf("snapshot_services = %v, want 24 (26 sources - 2 failures)", got["snapshot_services"])
+	}
+	if !reflect.DeepEqual(got["missing_services"], []string{"doc", "sheet"}) {
+		t.Fatalf("missing_services = %#v, want failed service IDs", got["missing_services"])
+	}
+	// matched 必须剔除 stub 占位，unmatched 据实等于 stub 数。
+	if got["matched_tools"] != 773 || got["unmatched_tools"] != 40 {
+		t.Fatalf("matched/unmatched = %v/%v, want 773/40 (813 surface - 40 stubs)", got["matched_tools"], got["unmatched_tools"])
+	}
+	if got["source_tools"] != 800 || got["surface_tools"] != 813 {
+		t.Fatalf("tool counts = %#v", got)
+	}
+}
+
+func TestBuildCoverageFullSnapshotHasNoMissingServices(t *testing.T) {
+	got := buildCoverage(26, nil, 813, 813, 0)
+	if got["snapshot_services"] != 26 {
+		t.Fatalf("snapshot_services = %v, want 26", got["snapshot_services"])
+	}
+	if !reflect.DeepEqual(got["missing_services"], []string{}) {
+		t.Fatalf("missing_services = %#v, want empty non-nil slice", got["missing_services"])
+	}
+	if got["matched_tools"] != 813 || got["unmatched_tools"] != 0 {
+		t.Fatalf("matched/unmatched = %v/%v, want 813/0 for stub-free snapshot", got["matched_tools"], got["unmatched_tools"])
+	}
+}
+
+// fakeLister returns canned tools/list results per endpoint.
+type fakeLister struct {
+	results map[string]transport.ToolsListResult
+	errs    map[string]error
+}
+
+func (f *fakeLister) ListTools(_ context.Context, endpoint string) (transport.ToolsListResult, error) {
+	if err := f.errs[endpoint]; err != nil {
+		return transport.ToolsListResult{}, err
+	}
+	return f.results[endpoint], nil
+}
+
+// stubDeps swaps every injection point for the duration of one test.
+func stubDeps(t *testing.T, token string, keychain func() (*auth.TokenData, error), servers []syncdata.ServerInfo, lister toolLister, registry func() (map[string]map[string]string, error)) {
+	t.Helper()
+	origGetenv, origLoad, origServers, origNew, origRegistry := getenv, loadTokenData, staticServers, newToolLister, registrySource
+	t.Cleanup(func() {
+		getenv, loadTokenData, staticServers, newToolLister, registrySource = origGetenv, origLoad, origServers, origNew, origRegistry
+	})
+	getenv = func(key string) string {
+		if key == "DWS_ACCESS_TOKEN" {
+			return token
+		}
+		return ""
+	}
+	loadTokenData = keychain
+	staticServers = func() []syncdata.ServerInfo { return servers }
+	newToolLister = func(string) toolLister { return lister }
+	registrySource = registry
+}
+
+func testRegistryRefs() (map[string]map[string]string, error) {
+	return map[string]map[string]string{
+		"doc.copy_document": {"product_id": "doc", "rpc_name": "copy_document"},
+		"doc.get_document":  {"product_id": "doc", "rpc_name": "get_document"},
+	}, nil
+}
+
+func TestCrossPlatformCoverageRunRefusesRetiredPinnedMCPMetadataPath(t *testing.T) {
+	stubDeps(t, "env-token", nil, nil, &fakeLister{}, testRegistryRefs)
+	var stderr bytes.Buffer
+	code := run([]string{"--output", "internal/cli/schema_mcp_metadata.json"}, &stderr)
+	if code != 2 {
+		t.Fatalf("run(retired pin) = %d, want 2", code)
+	}
+	if !strings.Contains(stderr.String(), "refusing to write retired Schema pin") {
+		t.Fatalf("stderr = %q, want retired-pin refusal", stderr.String())
+	}
+	if !retiredPinnedMCPMetadataPath("internal/cli/schema_mcp_metadata.json") ||
+		!retiredPinnedMCPMetadataPath("/tmp/repo/internal/cli/schema_mcp_metadata.json") ||
+		retiredPinnedMCPMetadataPath("artifacts/mcp_metadata_diagnostic.json") {
+		t.Fatal("retiredPinnedMCPMetadataPath classification is incorrect")
+	}
+}
+
+func TestRunNoTokenFails(t *testing.T) {
+	stubDeps(t, "", func() (*auth.TokenData, error) { return nil, errors.New("no keychain") }, nil, &fakeLister{}, testRegistryRefs)
+	var stderr bytes.Buffer
+	if code := run(nil, &stderr); code != 1 {
+		t.Fatalf("run() = %d, want 1", code)
+	}
+	if !strings.Contains(stderr.String(), "no auth token") {
+		t.Fatalf("stderr = %q, want no-auth-token hint", stderr.String())
+	}
+}
+
+func TestRunInvalidFlagFails(t *testing.T) {
+	stubDeps(t, "tok", nil, nil, &fakeLister{}, testRegistryRefs)
+	var stderr bytes.Buffer
+	if code := run([]string{"--nonexistent"}, &stderr); code != 2 {
+		t.Fatalf("run() = %d, want 2", code)
+	}
+}
+
+func TestResolveTokenKeychainFallback(t *testing.T) {
+	stubDeps(t, "", func() (*auth.TokenData, error) {
+		return &auth.TokenData{AccessToken: "kc-token"}, nil
+	}, nil, &fakeLister{}, testRegistryRefs)
+	var stderr bytes.Buffer
+	if got := resolveToken(&stderr); got != "kc-token" {
+		t.Fatalf("resolveToken() = %q, want kc-token", got)
+	}
+	if !strings.Contains(stderr.String(), "loaded token from keychain") {
+		t.Fatalf("stderr = %q, want keychain log", stderr.String())
+	}
+}
+
+func TestResolveTokenEmptyKeychainToken(t *testing.T) {
+	stubDeps(t, "", func() (*auth.TokenData, error) { return &auth.TokenData{}, nil }, nil, &fakeLister{}, testRegistryRefs)
+	var stderr bytes.Buffer
+	if got := resolveToken(&stderr); got != "" {
+		t.Fatalf("resolveToken() = %q, want empty", got)
+	}
+}
+
+func TestCrossPlatformCoverageRunWritesSnapshotWithHonestCoverage(t *testing.T) {
+	servers := []syncdata.ServerInfo{
+		{ID: "doc", Endpoint: "https://doc.example"},
+		{ID: "sheet", Endpoint: "https://sheet.example"},
+		{ID: "blank", Endpoint: "   "},
+	}
+	lister := &fakeLister{
+		results: map[string]transport.ToolsListResult{
+			"https://doc.example": {Tools: []transport.ToolDescriptor{
+				{Name: "copy_document", Title: "复制文档", Description: "copy", InputSchema: map[string]any{
+					"type": "object",
+					"properties": map[string]any{
+						"doc_id": map[string]any{"type": "string", "description": "文档 ID", "default": "d", "enum": []any{"a", "b", 3}},
+						"bogus":  "not-a-map",
+					},
+					"required": []any{"doc_id", 42},
+				}},
+				{Name: "   "},
+				{Name: "not_in_registry"},
+			}},
+		},
+		errs: map[string]error{"https://sheet.example": errors.New("boom")},
+	}
+	stubDeps(t, "env-token", nil, servers, lister, testRegistryRefs)
+
+	dir := t.TempDir()
+	output := filepath.Join(dir, "snapshot.json")
+	prev := `{"tools":{"doc.get_document":{"interface_ref":{"product_id":"doc-helper","rpc_name":"fetch_document"}}}}`
+	if err := os.WriteFile(output, []byte(prev), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	var stderr bytes.Buffer
+	if code := run([]string{"--output", output}, &stderr); code != 0 {
+		t.Fatalf("run() = %d, stderr=%s", code, stderr.String())
+	}
+
+	data, err := os.ReadFile(output)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var snapshot struct {
+		Version  int            `json:"version"`
+		Coverage map[string]any `json:"coverage"`
+		Tools    map[string]map[string]any
+	}
+	if err := json.Unmarshal(data, &snapshot); err != nil {
+		t.Fatal(err)
+	}
+	if snapshot.Version != 1 {
+		t.Fatalf("version = %d", snapshot.Version)
+	}
+	if got := snapshot.Coverage["snapshot_services"].(float64); got != 2 {
+		t.Fatalf("snapshot_services = %v, want 2 (3 servers - 1 failed; blank endpoint not counted as failed)", got)
+	}
+	if got := snapshot.Coverage["missing_services"].([]any); len(got) != 1 || got[0] != "sheet" {
+		t.Fatalf("missing_services = %v, want [sheet]", got)
+	}
+	live := snapshot.Tools["doc.copy_document"]
+	if live == nil || live["title"] != "复制文档" {
+		t.Fatalf("doc.copy_document = %#v, want live metadata", live)
+	}
+	params := live["parameters"].(map[string]any)
+	docID := params["doc_id"].(map[string]any)
+	if docID["type"] != "string" || docID["required"] != true || docID["default"] != "d" {
+		t.Fatalf("doc_id = %#v", docID)
+	}
+	if enum := docID["enum"].([]any); len(enum) != 2 {
+		t.Fatalf("enum = %v, want the 2 string members only", enum)
+	}
+	if _, ok := params["bogus"]; ok {
+		t.Fatal("non-map property should be skipped")
+	}
+	prevRef := snapshot.Tools["doc.get_document"]["interface_ref"].(map[string]any)
+	if prevRef["product_id"] != "doc-helper" {
+		t.Fatalf("previous reviewed ref lost: %#v", prevRef)
+	}
+	if _, ok := snapshot.Tools["not_in_registry"]; ok {
+		t.Fatal("tools outside the registry must be dropped")
+	}
+	if !strings.Contains(stderr.String(), "services unreachable: sheet") {
+		t.Fatalf("stderr = %q, want unreachable log", stderr.String())
+	}
+}
+
+func TestRunIgnoresCorruptPreviousSnapshot(t *testing.T) {
+	stubDeps(t, "env-token", nil, nil, &fakeLister{}, testRegistryRefs)
+	output := filepath.Join(t.TempDir(), "snapshot.json")
+	if err := os.WriteFile(output, []byte("{corrupt"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	var stderr bytes.Buffer
+	if code := run([]string{"--output", output}, &stderr); code != 0 {
+		t.Fatalf("run() = %d, stderr=%s", code, stderr.String())
+	}
+}
+
+func TestCrossPlatformCoverageRunRegistryLoadFailureStillWritesStublessSnapshot(t *testing.T) {
+	stubDeps(t, "env-token", nil, nil, &fakeLister{}, func() (map[string]map[string]string, error) { return nil, errors.New("no identity") })
+	output := filepath.Join(t.TempDir(), "snapshot.json")
+	var stderr bytes.Buffer
+	if code := run([]string{"--output", output}, &stderr); code != 0 {
+		t.Fatalf("run() = %d, stderr=%s", code, stderr.String())
+	}
+	if !strings.Contains(stderr.String(), "cannot collect command identity") {
+		t.Fatalf("stderr = %q, want identity collection warning", stderr.String())
+	}
+}
+
+func TestRunWriteFailure(t *testing.T) {
+	stubDeps(t, "env-token", nil, nil, &fakeLister{}, testRegistryRefs)
+	var stderr bytes.Buffer
+	badPath := filepath.Join(t.TempDir(), "missing-dir", "snapshot.json")
+	if code := run([]string{"--output", badPath}, &stderr); code != 1 {
+		t.Fatalf("run() = %d, want 1 on write failure", code)
+	}
+}
+
+func TestWriteMetadataMarshalFailure(t *testing.T) {
+	err := writeMetadata(filepath.Join(t.TempDir(), "out.json"), map[string]any{"bad": math.NaN()})
+	if err == nil || !strings.Contains(err.Error(), "marshal failed") {
+		t.Fatalf("err = %v, want marshal failure", err)
+	}
+}
+
+func TestMainDelegatesToRun(t *testing.T) {
+	stubDeps(t, "env-token", nil, nil, &fakeLister{}, testRegistryRefs)
+	origExit, origArgs := osExit, os.Args
+	t.Cleanup(func() { osExit, os.Args = origExit, origArgs })
+	exitCode := -1
+	osExit = func(code int) { exitCode = code }
+	os.Args = []string{"fetch_mcp_metadata", "--output", filepath.Join(t.TempDir(), "snapshot.json")}
+	main()
+	if exitCode != 0 {
+		t.Fatalf("main() exited with %d, want 0", exitCode)
+	}
+}
+
+func TestExtractParamsNilAndNonObjectSchemas(t *testing.T) {
+	if got := extractParams(nil); got != nil {
+		t.Fatalf("extractParams(nil) = %v, want nil", got)
+	}
+	if got := extractParams(map[string]any{"type": "object"}); got != nil {
+		t.Fatalf("extractParams(no properties) = %v, want nil", got)
+	}
+}
+
+func TestCrossPlatformCoverageNewToolListerBuildsAuthedClient(t *testing.T) {
+	if lister := newToolLister("tok"); lister == nil {
+		t.Fatal("newToolLister returned nil")
+	}
+}
+
+func TestRunRecordsSourceRevision(t *testing.T) {
+	stubDeps(t, "env-token", nil, nil, &fakeLister{}, testRegistryRefs)
+	dir := t.TempDir()
+	head := filepath.Join(dir, "HEAD")
+	if err := os.WriteFile(head, []byte("ref: refs/heads/feature\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	origHead := gitHeadPath
+	t.Cleanup(func() { gitHeadPath = origHead })
+	gitHeadPath = head
+
+	output := filepath.Join(dir, "snapshot.json")
+	var stderr bytes.Buffer
+	if code := run([]string{"--output", output}, &stderr); code != 0 {
+		t.Fatalf("run() = %d, stderr=%s", code, stderr.String())
+	}
+	data, err := os.ReadFile(output)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var snapshot struct {
+		SourceRevision string `json:"source_revision"`
+	}
+	if err := json.Unmarshal(data, &snapshot); err != nil {
+		t.Fatal(err)
+	}
+	if snapshot.SourceRevision != "ref: refs/heads/feature" {
+		t.Fatalf("source_revision = %q", snapshot.SourceRevision)
+	}
+}

cmd/interface-snapshot/main.go

@@ -0,0 +1,284 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// interface-snapshot is an internal CI helper. It is intentionally a separate
+// binary so it can be copied into a temporary worktree and compiled against an
+// older revision's real Cobra root.
+package main
+
+import (
+	"bytes"
+	"encoding/json"
+	"flag"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/app"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/i18n"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/interfacesnapshot"
+	"github.com/spf13/cobra"
+)
+
+var newRootCommand = func() *cobra.Command { return app.NewRootCommand() }
+
+func main() {
+	os.Exit(run(os.Args[1:], os.Stdout, os.Stderr))
+}
+
+func run(args []string, stdout, stderr io.Writer) int {
+	if len(args) == 0 {
+		printUsage(stderr)
+		return 2
+	}
+
+	switch args[0] {
+	case "generate":
+		if err := runGenerate(args[1:], stdout, stderr); err != nil {
+			fmt.Fprintln(stderr, err)
+			return 2
+		}
+		return 0
+	case "compare":
+		compatible, err := runCompare(args[1:], stdout, stderr)
+		if err != nil {
+			fmt.Fprintln(stderr, err)
+			return 2
+		}
+		if !compatible {
+			return 1
+		}
+		return 0
+	default:
+		fmt.Fprintf(stderr, "unknown command %q\n", args[0])
+		printUsage(stderr)
+		return 2
+	}
+}
+
+func runGenerate(args []string, stdout, stderr io.Writer) error {
+	flags := flag.NewFlagSet("generate", flag.ContinueOnError)
+	flags.SetOutput(stderr)
+	output := flags.String("output", "-", "snapshot output path, or - for stdout")
+	if err := flags.Parse(args); err != nil {
+		return err
+	}
+	if flags.NArg() != 0 {
+		return fmt.Errorf("generate accepts no positional arguments")
+	}
+
+	home, err := os.MkdirTemp("", "dws-interface-snapshot-*")
+	if err != nil {
+		return fmt.Errorf("create isolated home: %w", err)
+	}
+	defer os.RemoveAll(home)
+
+	environment := map[string]string{
+		"DWS_CONFIG_DIR": home,
+		"DWS_LANG":       "en",
+		"HOME":           home,
+		"NO_COLOR":       "1",
+		"USERPROFILE":    home,
+	}
+	type previousEnv struct {
+		value string
+		set   bool
+	}
+	previous := make(map[string]previousEnv, len(environment))
+	for key, value := range environment {
+		oldValue, wasSet := os.LookupEnv(key)
+		previous[key] = previousEnv{value: oldValue, set: wasSet}
+		if err := os.Setenv(key, value); err != nil {
+			return fmt.Errorf("set %s: %w", key, err)
+		}
+	}
+	defer func() {
+		for key, old := range previous {
+			if old.set {
+				_ = os.Setenv(key, old.value)
+			} else {
+				_ = os.Unsetenv(key)
+			}
+		}
+	}()
+	previousLang := i18n.Lang()
+	defer i18n.SetLang(previousLang)
+	i18n.SetLang("en")
+
+	root := newRootCommand()
+	snapshot := interfacesnapshot.Capture(root)
+	if err := validateHelpRendering(root, snapshot); err != nil {
+		return err
+	}
+	if *output == "-" {
+		return interfacesnapshot.Write(stdout, snapshot)
+	}
+
+	file, err := os.Create(filepath.Clean(*output))
+	if err != nil {
+		return fmt.Errorf("create snapshot %q: %w", *output, err)
+	}
+	writeErr := interfacesnapshot.Write(file, snapshot)
+	closeErr := file.Close()
+	if writeErr != nil {
+		return fmt.Errorf("write snapshot %q: %w", *output, writeErr)
+	}
+	if closeErr != nil {
+		return fmt.Errorf("close snapshot %q: %w", *output, closeErr)
+	}
+	return nil
+}
+
+func runCompare(args []string, stdout, stderr io.Writer) (bool, error) {
+	flags := flag.NewFlagSet("compare", flag.ContinueOnError)
+	flags.SetOutput(stderr)
+	currentPath := flags.String("current", "", "candidate snapshot path")
+	basePath := flags.String("base", "", "target main/development baseline snapshot path")
+	stablePath := flags.String("stable", "", "latest stable GA snapshot path")
+	approvedMigrationsPath := flags.String(
+		"approved-flag-migrations",
+		"",
+		"merge-base-owned approved flag migration manifest",
+	)
+	candidateMigrationsPath := flags.String(
+		"candidate-flag-migrations",
+		"",
+		"candidate flag migration manifest",
+	)
+	if err := flags.Parse(args); err != nil {
+		return false, err
+	}
+	if flags.NArg() != 0 {
+		return false, fmt.Errorf("compare accepts no positional arguments")
+	}
+	if *currentPath == "" {
+		return false, fmt.Errorf("compare requires --current")
+	}
+	if *basePath == "" && *stablePath == "" {
+		return false, fmt.Errorf("compare requires --base, --stable, or both")
+	}
+	if (*approvedMigrationsPath == "") != (*candidateMigrationsPath == "") {
+		return false, fmt.Errorf(
+			"--approved-flag-migrations and --candidate-flag-migrations must be provided together",
+		)
+	}
+	if *approvedMigrationsPath != "" && (*basePath == "" || *stablePath == "") {
+		return false, fmt.Errorf("flag migration compare requires both --base and --stable")
+	}
+
+	current, err := readSnapshot(*currentPath)
+	if err != nil {
+		return false, fmt.Errorf("read current snapshot: %w", err)
+	}
+	references := make(map[string]interfacesnapshot.Snapshot, 2)
+	if *basePath != "" {
+		references["main"], err = readSnapshot(*basePath)
+		if err != nil {
+			return false, fmt.Errorf("read main/development baseline snapshot: %w", err)
+		}
+	}
+	if *stablePath != "" {
+		references["stable"], err = readSnapshot(*stablePath)
+		if err != nil {
+			return false, fmt.Errorf("read stable snapshot: %w", err)
+		}
+	}
+
+	report := interfacesnapshot.CompareAll(current, references)
+	if *approvedMigrationsPath != "" {
+		approved, readErr := readFlagMigrationManifest(*approvedMigrationsPath)
+		if readErr != nil {
+			return false, fmt.Errorf("read approved flag migrations: %w", readErr)
+		}
+		candidate, readErr := readFlagMigrationManifest(*candidateMigrationsPath)
+		if readErr != nil {
+			return false, fmt.Errorf("read candidate flag migrations: %w", readErr)
+		}
+		report, err = interfacesnapshot.CompareAllWithFlagMigrations(
+			current,
+			references,
+			approved,
+			candidate,
+		)
+		if err != nil {
+			return false, fmt.Errorf("validate flag migration lifecycle: %w", err)
+		}
+	}
+	encoder := json.NewEncoder(stdout)
+	encoder.SetEscapeHTML(false)
+	encoder.SetIndent("", "  ")
+	if err := encoder.Encode(report); err != nil {
+		return false, fmt.Errorf("write comparison report: %w", err)
+	}
+	return report.Compatible, nil
+}
+
+func readFlagMigrationManifest(path string) (interfacesnapshot.FlagMigrationManifest, error) {
+	file, err := os.Open(filepath.Clean(path))
+	if err != nil {
+		return interfacesnapshot.FlagMigrationManifest{}, err
+	}
+	defer file.Close()
+	return interfacesnapshot.ReadFlagMigrationManifest(file)
+}
+
+func validateHelpRendering(root *cobra.Command, snapshot interfacesnapshot.Snapshot) error {
+	for _, command := range snapshot.Commands {
+		path := strings.TrimPrefix(command.Path, "dws")
+		resolved, remaining, err := root.Find(strings.Fields(path))
+		if err != nil || len(remaining) != 0 || resolved == nil {
+			return fmt.Errorf("resolve %q before help rendering: remaining=%v error=%v", command.Path, remaining, err)
+		}
+		if err := renderCommandHelp(resolved); err != nil {
+			return fmt.Errorf("render %q help: %w", command.Path, err)
+		}
+	}
+	return nil
+}
+
+func renderCommandHelp(command *cobra.Command) (err error) {
+	var stdout, stderr bytes.Buffer
+	command.InitDefaultHelpFlag()
+	command.SetOut(&stdout)
+	command.SetErr(&stderr)
+	defer func() {
+		if recovered := recover(); recovered != nil {
+			err = fmt.Errorf("help renderer panicked: %v", recovered)
+		}
+	}()
+	command.HelpFunc()(command, []string{})
+	if stderr.Len() > 0 {
+		return fmt.Errorf("help renderer wrote an error: %s", strings.TrimSpace(stderr.String()))
+	}
+	if stdout.Len() == 0 {
+		return fmt.Errorf("help renderer produced empty output")
+	}
+	return nil
+}
+
+func readSnapshot(path string) (interfacesnapshot.Snapshot, error) {
+	file, err := os.Open(filepath.Clean(path))
+	if err != nil {
+		return interfacesnapshot.Snapshot{}, err
+	}
+	defer file.Close()
+	return interfacesnapshot.Read(file)
+}
+
+func printUsage(w io.Writer) {
+	fmt.Fprintln(w, "usage:")
+	fmt.Fprintln(w, "  interface-snapshot generate [--output FILE]")
+	fmt.Fprintln(w, "  interface-snapshot compare --current FILE [--base FILE] [--stable FILE] [--approved-flag-migrations FILE --candidate-flag-migrations FILE]")
+}

cmd/interface-snapshot/main_test.go

@@ -0,0 +1,577 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package main
+
+import (
+	"bytes"
+	"errors"
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/interfacesnapshot"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/testseam"
+	"github.com/spf13/cobra"
+)
+
+func TestCrossPlatformCoverageRunGenerateCapturesActualRootOffline(t *testing.T) {
+	var stdout, stderr bytes.Buffer
+	if exitCode := run([]string{"generate"}, &stdout, &stderr); exitCode != 0 {
+		t.Fatalf("run(generate) exit=%d stderr=%s", exitCode, stderr.String())
+	}
+	snapshot, err := interfacesnapshot.Read(bytes.NewReader(stdout.Bytes()))
+	if err != nil {
+		t.Fatalf("decode generated snapshot: %v", err)
+	}
+
+	commands := make(map[string]interfacesnapshot.Command, len(snapshot.Commands))
+	for _, command := range snapshot.Commands {
+		commands[command.Path] = command
+	}
+	for _, path := range []string{"dws", "dws chat", "dws dev app create"} {
+		if _, ok := commands[path]; !ok {
+			t.Errorf("actual root snapshot is missing %q", path)
+		}
+	}
+	for _, path := range []string{"dws completion", "dws help"} {
+		if _, ok := commands[path]; ok {
+			t.Errorf("framework-noise path %q leaked into snapshot", path)
+		}
+	}
+
+	create := commands["dws dev app create"]
+	if !hasFlag(create.LocalFlags, "name", "string") {
+		t.Errorf("dev app create local flags do not contain --name string: %#v", create.LocalFlags)
+	}
+	if !hasFlag(create.InheritedFlags, "profile", "string") {
+		t.Errorf("dev app create inherited flags do not contain --profile string: %#v", create.InheritedFlags)
+	}
+}
+
+func TestCrossPlatformCoverageRunGenerateRejectsHelpRenderingFailure(t *testing.T) {
+	testseam.Swap(t, &newRootCommand, func() *cobra.Command {
+		root := &cobra.Command{Use: "dws"}
+		root.SetHelpFunc(func(command *cobra.Command, _ []string) {
+			_, _ = io.WriteString(command.ErrOrStderr(), "injected help failure")
+		})
+		return root
+	})
+
+	var stdout, stderr bytes.Buffer
+	if exitCode := run([]string{"generate"}, &stdout, &stderr); exitCode != 2 {
+		t.Fatalf("run(generate) exit=%d stderr=%s", exitCode, stderr.String())
+	}
+	if !strings.Contains(stderr.String(), "injected help failure") {
+		t.Fatalf("run(generate) stderr=%q", stderr.String())
+	}
+}
+
+func TestCrossPlatformCoverageRunCompareUsesBothSnapshotInputsAndExitCode(t *testing.T) {
+	current := commandSnapshot("dws")
+	mergeBase := commandSnapshot("dws")
+	stable := commandSnapshot("dws", "dws legacy")
+
+	dir := t.TempDir()
+	currentPath := writeSnapshot(t, dir, "current.json", current)
+	mergeBasePath := writeSnapshot(t, dir, "base.json", mergeBase)
+	stablePath := writeSnapshot(t, dir, "stable.json", stable)
+
+	var stdout, stderr bytes.Buffer
+	exitCode := run([]string{
+		"compare",
+		"--current", currentPath,
+		"--base", mergeBasePath,
+		"--stable", stablePath,
+	}, &stdout, &stderr)
+	if exitCode != 1 {
+		t.Fatalf("run(compare) exit=%d, want 1; stdout=%s stderr=%s", exitCode, stdout.String(), stderr.String())
+	}
+	if !bytes.Contains(stdout.Bytes(), []byte(`"reference": "main"`)) ||
+		!bytes.Contains(stdout.Bytes(), []byte(`"reference": "stable"`)) ||
+		!bytes.Contains(stdout.Bytes(), []byte(`"kind": "command_removed"`)) {
+		t.Fatalf("comparison report does not contain both references and the blocking change:\n%s", stdout.String())
+	}
+}
+
+func TestCrossPlatformCoverageRunCompareEnforcesBaseOwnedFlagMigrationLifecycle(t *testing.T) {
+	dir := t.TempDir()
+	before := flagMigrationSnapshot(false)
+	after := flagMigrationSnapshot(true)
+	currentPath := writeSnapshot(t, dir, "current.json", after)
+	basePath := writeSnapshot(t, dir, "base.json", before)
+	stablePath := writeSnapshot(t, dir, "stable.json", before)
+	approvedPath := writeManifest(t, dir, "approved.json", flagMigrationManifestJSON("pending"))
+	candidatePath := writeManifest(t, dir, "candidate.json", flagMigrationManifestJSON("consumed"))
+
+	var stdout, stderr bytes.Buffer
+	exitCode := run([]string{
+		"compare",
+		"--current", currentPath,
+		"--base", basePath,
+		"--stable", stablePath,
+		"--approved-flag-migrations", approvedPath,
+		"--candidate-flag-migrations", candidatePath,
+	}, &stdout, &stderr)
+	if exitCode != 0 {
+		t.Fatalf("exact base-owned migration exit=%d stderr=%s", exitCode, stderr.String())
+	}
+	if !bytes.Contains(stdout.Bytes(), []byte(`"compatible": true`)) {
+		t.Fatalf("exact migration report is not compatible:\n%s", stdout.String())
+	}
+
+	stdout.Reset()
+	stderr.Reset()
+	emptyApproved := writeManifest(t, dir, "empty-approved.json", `{"version":1,"migrations":[]}`)
+	exitCode = run([]string{
+		"compare",
+		"--current", currentPath,
+		"--base", basePath,
+		"--stable", stablePath,
+		"--approved-flag-migrations", emptyApproved,
+		"--candidate-flag-migrations", candidatePath,
+	}, &stdout, &stderr)
+	if exitCode != 2 || !strings.Contains(stderr.String(), "must start pending") {
+		t.Fatalf("candidate self-approval exit=%d stdout=%s stderr=%s", exitCode, stdout.String(), stderr.String())
+	}
+}
+
+func TestCrossPlatformCoverageRunCompareRequiresBothFlagMigrationInputs(t *testing.T) {
+	dir := t.TempDir()
+	currentPath := writeSnapshot(t, dir, "current.json", commandSnapshot("dws"))
+	basePath := writeSnapshot(t, dir, "base.json", commandSnapshot("dws"))
+	approvedPath := writeManifest(t, dir, "approved.json", `{"version":1,"migrations":[]}`)
+
+	var stdout, stderr bytes.Buffer
+	exitCode := run([]string{
+		"compare",
+		"--current", currentPath,
+		"--base", basePath,
+		"--approved-flag-migrations", approvedPath,
+	}, &stdout, &stderr)
+	if exitCode != 2 || !strings.Contains(stderr.String(), "must be provided together") {
+		t.Fatalf("one-sided migration input exit=%d stdout=%s stderr=%s", exitCode, stdout.String(), stderr.String())
+	}
+}
+
+func TestCrossPlatformCoverageRunCompareRequiresBothReferencesForFlagMigrations(t *testing.T) {
+	dir := t.TempDir()
+	currentPath := writeSnapshot(t, dir, "current.json", commandSnapshot("dws"))
+	basePath := writeSnapshot(t, dir, "base.json", commandSnapshot("dws"))
+	stablePath := writeSnapshot(t, dir, "stable.json", commandSnapshot("dws"))
+	approvedPath := writeManifest(t, dir, "approved.json", `{"version":1,"migrations":[]}`)
+	candidatePath := writeManifest(t, dir, "candidate.json", `{"version":1,"migrations":[]}`)
+
+	tests := []struct {
+		name string
+		args []string
+	}{
+		{
+			name: "missing stable",
+			args: []string{"--base", basePath},
+		},
+		{
+			name: "missing base",
+			args: []string{"--stable", stablePath},
+		},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			args := []string{"compare", "--current", currentPath}
+			args = append(args, test.args...)
+			args = append(args,
+				"--approved-flag-migrations", approvedPath,
+				"--candidate-flag-migrations", candidatePath,
+			)
+			var stdout, stderr bytes.Buffer
+			exitCode := run(args, &stdout, &stderr)
+			if exitCode != 2 || !strings.Contains(stderr.String(), "requires both --base and --stable") {
+				t.Fatalf("one-reference migration compare exit=%d stdout=%s stderr=%s", exitCode, stdout.String(), stderr.String())
+			}
+		})
+	}
+}
+
+func TestCrossPlatformCoverageRunPrintsUsageForMissingAndUnknownCommands(t *testing.T) {
+	tests := []struct {
+		name       string
+		args       []string
+		wantStderr []string
+	}{
+		{
+			name:       "missing command",
+			args:       nil,
+			wantStderr: []string{"usage:", "interface-snapshot generate", "--approved-flag-migrations"},
+		},
+		{
+			name:       "unknown command",
+			args:       []string{"unknown"},
+			wantStderr: []string{`unknown command "unknown"`, "usage:", "interface-snapshot compare"},
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			var stdout, stderr bytes.Buffer
+			if exitCode := run(test.args, &stdout, &stderr); exitCode != 2 {
+				t.Fatalf("run(%v) exit=%d, want 2", test.args, exitCode)
+			}
+			if stdout.Len() != 0 {
+				t.Fatalf("run(%v) unexpectedly wrote stdout: %s", test.args, stdout.String())
+			}
+			for _, want := range test.wantStderr {
+				if !strings.Contains(stderr.String(), want) {
+					t.Errorf("run(%v) stderr missing %q:\n%s", test.args, want, stderr.String())
+				}
+			}
+		})
+	}
+}
+
+func TestCrossPlatformCoverageRunRejectsInvalidSubcommandArguments(t *testing.T) {
+	tests := []struct {
+		name string
+		args []string
+		want string
+	}{
+		{name: "generate unknown flag", args: []string{"generate", "--unknown"}, want: "flag provided but not defined"},
+		{name: "generate positional", args: []string{"generate", "unexpected"}, want: "generate accepts no positional arguments"},
+		{name: "compare unknown flag", args: []string{"compare", "--unknown"}, want: "flag provided but not defined"},
+		{name: "compare positional", args: []string{"compare", "unexpected"}, want: "compare accepts no positional arguments"},
+		{name: "compare missing current", args: []string{"compare", "--base", "base.json"}, want: "compare requires --current"},
+		{name: "compare missing reference", args: []string{"compare", "--current", "current.json"}, want: "compare requires --base, --stable, or both"},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			var stdout, stderr bytes.Buffer
+			if exitCode := run(test.args, &stdout, &stderr); exitCode != 2 {
+				t.Fatalf("run(%v) exit=%d, want 2", test.args, exitCode)
+			}
+			if !strings.Contains(stderr.String(), test.want) {
+				t.Fatalf("run(%v) stderr missing %q:\n%s", test.args, test.want, stderr.String())
+			}
+		})
+	}
+}
+
+func TestCrossPlatformCoverageRunGenerateRejectsUnsafeOutputPath(t *testing.T) {
+	outputDirectory := t.TempDir()
+	var stdout, stderr bytes.Buffer
+	exitCode := run([]string{"generate", "--output", outputDirectory}, &stdout, &stderr)
+	if exitCode != 2 || !strings.Contains(stderr.String(), "create snapshot") {
+		t.Fatalf("directory output exit=%d stdout=%s stderr=%s", exitCode, stdout.String(), stderr.String())
+	}
+}
+
+func TestCrossPlatformCoverageRunGenerateReportsTemporaryDirectoryFailure(t *testing.T) {
+	missingTempRoot := filepath.Join(t.TempDir(), "missing")
+	for _, name := range []string{"TMPDIR", "TMP", "TEMP"} {
+		t.Setenv(name, missingTempRoot)
+	}
+
+	var stdout, stderr bytes.Buffer
+	exitCode := run([]string{"generate"}, &stdout, &stderr)
+	if exitCode != 2 || !strings.Contains(stderr.String(), "create isolated home") {
+		t.Fatalf("invalid temporary root exit=%d stdout=%s stderr=%s", exitCode, stdout.String(), stderr.String())
+	}
+}
+
+func TestCrossPlatformCoverageRunCompareReportsSnapshotReadFailures(t *testing.T) {
+	dir := t.TempDir()
+	validPath := writeSnapshot(t, dir, "valid.json", commandSnapshot("dws"))
+	missingPath := filepath.Join(dir, "missing.json")
+	tests := []struct {
+		name string
+		args []string
+		want string
+	}{
+		{
+			name: "current",
+			args: []string{"compare", "--current", missingPath, "--base", validPath},
+			want: "read current snapshot",
+		},
+		{
+			name: "main",
+			args: []string{"compare", "--current", validPath, "--base", missingPath},
+			want: "read main/development baseline snapshot",
+		},
+		{
+			name: "stable",
+			args: []string{"compare", "--current", validPath, "--stable", missingPath},
+			want: "read stable snapshot",
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			var stdout, stderr bytes.Buffer
+			if exitCode := run(test.args, &stdout, &stderr); exitCode != 2 {
+				t.Fatalf("run(compare) exit=%d, want 2", exitCode)
+			}
+			if !strings.Contains(stderr.String(), test.want) {
+				t.Fatalf("stderr missing %q:\n%s", test.want, stderr.String())
+			}
+		})
+	}
+}
+
+func TestCrossPlatformCoverageRunCompareReportsEachManifestReadFailure(t *testing.T) {
+	dir := t.TempDir()
+	snapshotPath := writeSnapshot(t, dir, "snapshot.json", commandSnapshot("dws"))
+	validManifest := writeManifest(t, dir, "valid-manifest.json", `{"version":1,"migrations":[]}`)
+	invalidManifest := writeManifest(t, dir, "invalid-manifest.json", `{`)
+	tests := []struct {
+		name      string
+		approved  string
+		candidate string
+		want      string
+	}{
+		{
+			name:      "approved manifest",
+			approved:  invalidManifest,
+			candidate: validManifest,
+			want:      "read approved flag migrations",
+		},
+		{
+			name:      "candidate manifest",
+			approved:  validManifest,
+			candidate: invalidManifest,
+			want:      "read candidate flag migrations",
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			var stdout, stderr bytes.Buffer
+			exitCode := run([]string{
+				"compare",
+				"--current", snapshotPath,
+				"--base", snapshotPath,
+				"--stable", snapshotPath,
+				"--approved-flag-migrations", test.approved,
+				"--candidate-flag-migrations", test.candidate,
+			}, &stdout, &stderr)
+			if exitCode != 2 || !strings.Contains(stderr.String(), test.want) {
+				t.Fatalf("%s exit=%d stdout=%s stderr=%s", test.name, exitCode, stdout.String(), stderr.String())
+			}
+		})
+	}
+}
+
+func TestCrossPlatformCoverageRunCompareReportsOutputFailure(t *testing.T) {
+	dir := t.TempDir()
+	snapshotPath := writeSnapshot(t, dir, "snapshot.json", commandSnapshot("dws"))
+	var stderr bytes.Buffer
+	exitCode := run([]string{
+		"compare",
+		"--current", snapshotPath,
+		"--base", snapshotPath,
+	}, failingWriter{}, &stderr)
+	if exitCode != 2 || !strings.Contains(stderr.String(), "write comparison report") {
+		t.Fatalf("comparison output failure exit=%d stderr=%s", exitCode, stderr.String())
+	}
+}
+
+func TestCrossPlatformCoverageReadHelpersRejectMissingAndInvalidInputs(t *testing.T) {
+	dir := t.TempDir()
+	missingPath := filepath.Join(dir, "missing.json")
+	invalidPath := filepath.Join(dir, "invalid.json")
+	if err := os.WriteFile(invalidPath, []byte(`{`), 0o600); err != nil {
+		t.Fatalf("write invalid fixture: %v", err)
+	}
+
+	if _, err := readSnapshot(missingPath); err == nil {
+		t.Fatal("readSnapshot(missing) unexpectedly succeeded")
+	}
+	if _, err := readSnapshot(invalidPath); err == nil {
+		t.Fatal("readSnapshot(invalid) unexpectedly succeeded")
+	}
+	if _, err := readFlagMigrationManifest(missingPath); err == nil {
+		t.Fatal("readFlagMigrationManifest(missing) unexpectedly succeeded")
+	}
+	if _, err := readFlagMigrationManifest(invalidPath); err == nil {
+		t.Fatal("readFlagMigrationManifest(invalid) unexpectedly succeeded")
+	}
+}
+
+func TestCrossPlatformCoverageValidateHelpRenderingReportsResolveError(t *testing.T) {
+	root := &cobra.Command{Use: "dws"}
+	err := validateHelpRendering(root, commandSnapshot("dws missing"))
+	if err == nil || !strings.Contains(err.Error(), `resolve "dws missing" before help rendering`) {
+		t.Fatalf("validateHelpRendering resolve error = %v", err)
+	}
+}
+
+func TestCrossPlatformCoverageValidateHelpRenderingReportsTemplateError(t *testing.T) {
+	root := &cobra.Command{Use: "dws"}
+	root.SetHelpTemplate(`{{index .Commands 99}}`)
+	err := validateHelpRendering(root, commandSnapshot("dws"))
+	if err == nil || !strings.Contains(err.Error(), `render "dws" help`) {
+		t.Fatalf("validateHelpRendering template error = %v", err)
+	}
+}
+
+func TestCrossPlatformCoverageValidateHelpRenderingRecoversTemplatePanic(t *testing.T) {
+	root := &cobra.Command{Use: "dws"}
+	root.SetHelpTemplate("{{")
+	err := validateHelpRendering(root, commandSnapshot("dws"))
+	if err == nil || !strings.Contains(err.Error(), `render "dws" help`) {
+		t.Fatalf("validateHelpRendering template panic = %v", err)
+	}
+}
+
+func TestCrossPlatformCoverageValidateHelpRenderingRejectsCustomHelpStderr(t *testing.T) {
+	root := &cobra.Command{Use: "dws"}
+	root.SetHelpFunc(func(command *cobra.Command, _ []string) {
+		_, _ = io.WriteString(command.ErrOrStderr(), "injected help failure")
+	})
+	err := validateHelpRendering(root, commandSnapshot("dws"))
+	if err == nil || !strings.Contains(err.Error(), "injected help failure") {
+		t.Fatalf("validateHelpRendering custom stderr = %v", err)
+	}
+}
+
+func TestCrossPlatformCoverageValidateHelpRenderingRejectsEmptyOutput(t *testing.T) {
+	root := &cobra.Command{Use: "dws"}
+	root.SetHelpFunc(func(*cobra.Command, []string) {})
+	err := validateHelpRendering(root, commandSnapshot("dws"))
+	if err == nil || !strings.Contains(err.Error(), "empty") {
+		t.Fatalf("validateHelpRendering empty output = %v", err)
+	}
+}
+
+func TestCrossPlatformCoverageValidateHelpRenderingAcceptsNormalOutput(t *testing.T) {
+	root := &cobra.Command{Use: "dws", Short: "root command"}
+	if err := validateHelpRendering(root, commandSnapshot("dws")); err != nil {
+		t.Fatalf("validateHelpRendering normal output: %v", err)
+	}
+}
+
+type failingWriter struct{}
+
+func (failingWriter) Write([]byte) (int, error) {
+	return 0, errors.New("injected write failure")
+}
+
+var _ io.Writer = failingWriter{}
+
+func commandSnapshot(paths ...string) interfacesnapshot.Snapshot {
+	commands := make([]interfacesnapshot.Command, 0, len(paths))
+	for _, path := range paths {
+		commands = append(commands, interfacesnapshot.Command{
+			Path:           path,
+			Aliases:        []string{},
+			LocalFlags:     []interfacesnapshot.Flag{},
+			InheritedFlags: []interfacesnapshot.Flag{},
+		})
+	}
+	return interfacesnapshot.Snapshot{
+		SchemaVersion: interfacesnapshot.SchemaVersion,
+		Rules: interfacesnapshot.Rules{
+			ExcludedCommandSubtrees: []string{},
+			ExcludedFlags:           []string{},
+		},
+		Commands: commands,
+	}
+}
+
+func writeSnapshot(t *testing.T, dir, name string, snapshot interfacesnapshot.Snapshot) string {
+	t.Helper()
+	path := filepath.Join(dir, name)
+	file, err := os.Create(path)
+	if err != nil {
+		t.Fatalf("create %s: %v", path, err)
+	}
+	if err := interfacesnapshot.Write(file, snapshot); err != nil {
+		file.Close()
+		t.Fatalf("write %s: %v", path, err)
+	}
+	if err := file.Close(); err != nil {
+		t.Fatalf("close %s: %v", path, err)
+	}
+	return path
+}
+
+func writeManifest(t *testing.T, dir, name, contents string) string {
+	t.Helper()
+	path := filepath.Join(dir, name)
+	if err := os.WriteFile(path, []byte(contents), 0o600); err != nil {
+		t.Fatalf("write %s: %v", path, err)
+	}
+	return path
+}
+
+func flagMigrationSnapshot(after bool) interfacesnapshot.Snapshot {
+	legacy := interfacesnapshot.Flag{
+		Name:      "legacy-id",
+		Shorthand: "l",
+		Type:      "string",
+		Default:   "",
+		NoOpt:     "auto",
+		Required:  true,
+	}
+	flags := []interfacesnapshot.Flag{legacy}
+	if after {
+		legacy.Required = false
+		legacy.Hidden = true
+		legacy.AliasOf = "message-id"
+		flags = []interfacesnapshot.Flag{
+			legacy,
+			{Name: "message-id", Type: "string", Default: "", Required: true},
+		}
+	}
+	return interfacesnapshot.Snapshot{
+		SchemaVersion: interfacesnapshot.SchemaVersion,
+		Rules: interfacesnapshot.Rules{
+			ExcludedCommandSubtrees: []string{},
+			ExcludedFlags:           []string{},
+		},
+		Commands: []interfacesnapshot.Command{
+			{Path: "dws", Runnable: true, Aliases: []string{}, LocalFlags: []interfacesnapshot.Flag{}, InheritedFlags: []interfacesnapshot.Flag{}},
+			{Path: "dws chat send", Runnable: true, Aliases: []string{}, LocalFlags: flags, InheritedFlags: []interfacesnapshot.Flag{}},
+		},
+	}
+}
+
+func flagMigrationManifestJSON(state string) string {
+	return strings.Replace(`{
+  "version": 1,
+  "migrations": [{
+    "command": "dws chat send",
+    "legacy": {
+      "name": "legacy-id",
+      "before": {"present": true, "type": "string", "required": true, "shorthand": "l", "no_opt": "auto", "scope": "local"},
+      "after": {"present": true, "type": "string", "hidden": true, "shorthand": "l", "no_opt": "auto", "scope": "local", "alias_of": "message-id"}
+    },
+    "canonical": {
+      "name": "message-id",
+      "before": {"present": false},
+      "after": {"present": true, "type": "string", "required": true, "scope": "local"}
+    },
+    "state": "STATE",
+    "reason": "reviewed exact migration"
+  }]
+}`, "STATE", state, 1)
+}
+
+func hasFlag(flags []interfacesnapshot.Flag, name, flagType string) bool {
+	for _, flag := range flags {
+		if flag.Name == name && flag.Type == flagType {
+			return true
+		}
+	}
+	return false
+}

cmd/main.go

@@ -19,6 +19,8 @@ import (
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/app"
 )
 
+var exit = os.Exit
+
 func main() {
-	os.Exit(app.Execute())
+	exit(app.Execute())
 }

cmd/main_test.go

@@ -0,0 +1,27 @@
+package main
+
+import (
+	"os"
+	"testing"
+)
+
+func TestCrossPlatformCoverageMainExitsWithSuccessfulVersionCommand(t *testing.T) {
+	previousExit := exit
+	previousArgs := os.Args
+	t.Cleanup(func() {
+		exit = previousExit
+		os.Args = previousArgs
+	})
+
+	called := false
+	code := -1
+	exit = func(value int) {
+		called = true
+		code = value
+	}
+	os.Args = []string{"dws", "version"}
+	main()
+	if !called || code != 0 {
+		t.Fatalf("main exit = called %v, code %d", called, code)
+	}
+}

docs/architecture.md

@@ -1,22 +1,26 @@
 # Architecture
 
-`dws` is a Go CLI that turns DingTalk MCP metadata into a command-line surface for both humans and AI agents.
+`dws` is a Go CLI with a versioned, static command surface for DingTalk MCP capabilities. Cobra help serves humans; runtime-assembled Schema (`ResolveSchemaBuild`) serves AI agents.
 
 ## High-Level Flow
 
 1. `cmd` is the CLI entrypoint, invoking `internal/app` to build the root Cobra command tree.
-2. `internal/app` wires static utility commands (`auth`, `audit`, `schema`, `completion`), product helper commands, and plugin commands.
+2. `internal/app` wires static utility commands (`auth`, `audit`, `schema`, `completion`), product helpers, and versioned plugin descriptors.
 3. `internal/helpers` contains the main command handlers for all product surfaces (`dev`, `chat`, `calendar`, `contact`, `aitable`, etc.).
 4. `internal/executor` and `internal/transport` execute MCP JSON-RPC calls; `internal/output` formats responses.
 5. `internal/auth` manages login state, PAT tokens, and agent-code detection.
+6. Schema assembly (`ResolveSchemaBuild`) starts from the reviewed `CommandRegistry`, binds each identity to the exact current Cobra leaf, and then resolves typed constraints, sanitized MCP snapshots, and leaf ContractFinal / ProductDecl into one `SchemaRegistry`. Startup and Schema queries do not call MCP `tools/list`. There is no generate-written Catalog delivery step.
+7. Production Catalog / `ResolveMeta` consume the lazily assembled registry via `RegisterSchemaSourceRoot` → `ResolveSchemaBuild` / `deliverySchemaCatalog` (声明即 Catalog; lazy `sync.Once`). `ResolveMeta` projects Identity/Safety/Selection from that assembly into an in-process map cache — not a committed `schema_catalog/` or `schema_meta_index.*` fixture. Flag-to-interface property delivery is owned by leaf `ParamDecl.Property` (native annotations). `schema_parameter_mapping_ledger.go` holds reviewed `mapping_exclusions` / `removals` (the empty `schema_parameter_bindings.json` audit table is retired). CLI `required` and constraints come from the resolved typed contract, while MCP `required` remains interface-only metadata.
+8. Agent selection results are fixed in versioned review inputs. Every public tool has explicit use/avoid/example and interface disposition metadata; Skill references that are not current leaves require an explicit alias/group/stale/out-of-surface review instead of fuzzy runtime matching.
 
 ## Repository Structure
 
 - `cmd`: CLI entrypoint
 - `internal/app`: root command wiring, static utility commands, and plugin loading
 - `internal/helpers`: product command handlers (dev, chat, calendar, contact, etc.)
-- `internal/plugin`: plugin-based dynamic command loader
-- `internal/cli`: catalog types and endpoint loader (static endpoint mode)
+- `internal/plugin`: versioned plugin manifest, hook, skill, and transport descriptor loading
+- `internal/cli`: Schema assembly, `dws schema` query, and catalog contracts
+- `internal/generator`: CI/determinism tools (`cmd_schema_catalog` dump) and param-alias generate
 - `internal/executor`: invocation dispatch and result handling
 - `internal/transport`: MCP HTTP client and request signing
 - `internal/auth`: login, token management, agent-code detection, identity
@@ -26,11 +30,16 @@
 - `internal/security`: endpoint allowlist and domain trust
 - `internal/safety`: runtime safety checks (confirm prompts, dry-run guards)
 - `internal/cobracmd`: shared Cobra command builders
+- `internal/corecmd`: dispatch-agnostic leaf-command base — flag registration,
+  alias/env/default value resolution, required and cross-flag constraint
+  validation, Risk write confirmation, toolArgs assembly, Runtime Schema
+  projection. Distinct from `internal/cobracmd` (generic tree helpers): it owns
+  the declarative leaf contract (`corecmd.Spec`) that the LeafSpec framework is
+  built on and that the Shortcut adapter projects into.
 - `internal/pat`: PAT (Personal Access Token) authorization flow
 - `internal/output`: response formatting (json, table, raw, pretty)
 - `internal/logging`: structured logging and argument sanitization
 - `internal/tui`: terminal UI helpers
-- `internal/recovery`: panic recovery and graceful degradation
 - `pkg/configmeta`: environment variable registry and documentation
 - `pkg/config`: configuration constants and paths
 - `pkg/edition`: edition detection (oss vs enterprise)
@@ -39,3 +48,51 @@
 - `skills/`: bundled agent skills (mono/ and multi/ layouts)
 - `test/`: CLI, integration, contract, unit, and skill E2E tests
 - `scripts/`: install scripts, policy checks, and CI helpers
+
+## Quality Pipeline
+
+Quality enforcement is layered so a pull request receives fast, deterministic
+admission feedback without pretending that downstream integration has already
+run.
+
+```mermaid
+flowchart TB
+  PR["Pull request"] --> CLASSIFY["Fail-closed risk classification"]
+  CLASSIFY --> DOCS["Documentation-only<br/>asset/content validation"]
+  CLASSIFY --> STANDARD["Standard<br/>affected + reverse-dependent race<br/>scope-matched HEAD/base coverage"]
+  CLASSIFY --> HIGH["High-risk / main<br/>full race + native tests"]
+  DOCS --> CA["CI"]
+  STANDARD --> CA
+  HIGH --> CA
+  subgraph CA_CHECKS["Nine required contexts"]
+    L["Lint"]
+    T["Test"]
+    C["Coverage"]
+    P["Policy"]
+    E["Edition"]
+    I["Interface Integrity"]
+    A["AI Behavior"]
+    S["CLI Smoke"]
+    M["Mock MCP"]
+  end
+  CA --> CA_CHECKS
+  CA_CHECKS --> MAIN["Protected main"]
+  MAIN --> MP["Main Integration — 主干集成<br/>Multi-profile E2E"]
+  MAIN --> PLATFORM["Risk-selected / release native platform validation"]
+  MP --> RELEASE["Release delivery"]
+  PLATFORM --> RELEASE
+```
+
+All nine named contexts are produced for every tier. Domain-specific helpers
+run when their owned surface is affected; otherwise the corresponding context
+records an explicit unaffected success. Standard code changes still receive
+representative Darwin/Windows compilation. High-risk PRs and protected `main`
+run the complete race and native test suites, while platform-sensitive diffs
+also receive native changed-code coverage.
+
+Review orchestration is also base-owned: it requests one eligible peer without
+executing PR code, re-routes an updated head when needed, and auto-merge
+completes only after the latest push has peer approval plus the current
+revision's nine strict contexts. Complete Multi-profile E2E remains downstream
+of PR admission. See [`docs/ci-pr-gates.md`](ci-pr-gates.md) for the exact
+classification, context, reviewer, and ruleset contract.

docs/automation.md

@@ -43,7 +43,7 @@ repository root while preserving repo-local guidance for automation.
 - Error message or category issues: inspect `internal/errors`
 - Audit log issues: inspect `internal/audit`
 - Plugin loading or command surface: inspect `internal/plugin`
-- Failure or degraded mode: inspect `internal/errors`, `internal/recovery`
+- Failure or degraded mode: inspect `internal/errors`
 
 ## Policy Checks
 
@@ -62,6 +62,74 @@ make lint
 git diff --check
 ```
 
+## Homebrew Formula Delivery
+
+Official releases use the Release workflow's built-in `GITHUB_TOKEN` to update
+exactly one tracked Formula after the immutable GitHub assets and their
+checksums have passed verification. The publisher validates the rendered Ruby,
+commits only the configured Formula path, never force-pushes `main`, and retries
+from a fresh clone up to three times when `main` advances concurrently. Normal
+stable and beta releases do not create a Formula PR or run a permission
+canary. The workflow uses the existing repository-scoped
+`HOMEBREW_PR_TOKEN` release identity because GitHub does not allow its built-in
+Actions App to bypass this repository's rulesets. That identity is the sole
+user bypass actor on the two default-branch rulesets. The workflow creates the
+nine Code Admission checks for the Formula-only commit only after proving its
+sole parent already has all nine successful checks and the committed Formula
+exactly matches this release's verified bytes.
+
+Keep `HOMEBREW_PR_TOKEN` repository-scoped with `Contents: write` and
+`Pull requests: write` (the latter remains necessary for withdrawal rollback),
+keep its owner as the designated ruleset bypass actor, and do not reuse
+`RELEASE_GOVERNANCE_TOKEN`. The workflow and publisher provide the Formula-only
+path restriction; GitHub rulesets do not infer that restriction from the token.
+
+## Release Governance and Recovery
+
+Store `RELEASE_GOVERNANCE_TOKEN` as a dedicated Actions secret with only
+repository `Administration: read`. The immutable-releases REST endpoint is an
+administration setting and cannot be read by the workflow's built-in
+`GITHUB_TOKEN`. Both the default-branch governance preflight and the tag
+contract use this same credential so a missing or expired identity is detected
+before an irreversible tag is created.
+
+Recovery is restricted to an existing annotated tag whose exact tag object,
+commit, sealed metadata, original failed run/attempt, requester identity and
+Release state all match; it then reuses the normal release jobs without a
+second-person environment approval. A same-run “Re-run failed jobs” is even
+lighter: the seal job may adopt an existing tag only when its complete
+authority matches that run and its original attempt is not newer than the
+current attempt. Do not put publication secrets in temporary branches or
+create ad-hoc recovery workflows.
+
+Cloud-sealed releases mirror to OSS only when the repository variable
+`ENABLE_OSS_MIRROR` is exactly `true`. Leave the variable unset while no Bucket
+is provisioned; GitHub, npm, and Homebrew delivery can then complete without
+running the OSS step. Once enabled, missing credentials, an invalid Bucket, or
+an upload failure remains fail-closed. The cloud tag immutably records the
+decision as `OSS-Mirror: enabled|deferred`; publication and withdrawal consume
+that sealed value instead of the variable's later state. Deferred releases
+cannot use `repair_oss_version`; enabling OSS applies to later release tags
+until an audited immutable repair marker is implemented.
+
+If an immutable GitHub Release and npm package were delivered but an enabled
+downstream China mirror failed, dispatch the normal `Release` workflow from the
+protected default branch with exactly one of `repair_gitee_version` or
+`repair_oss_version`. Channel repair accepts a fully successful exact release,
+or a failed exact-tag run only when its latest attempt completed the release
+contract, build, Apple signature, immutable GitHub publication, and npm
+delivery checks for the exact tagged commit. OSS repair additionally requires
+the tag's sealed policy to be `enabled`. It then downloads and re-verifies the
+immutable assets before invoking only the selected mirror. For a failed
+release, an OSS repair requires the OSS step itself to be the recorded failure.
+A Gitee repair accepts either a failed Gitee job or a Gitee job that was
+skipped behind that OSS failure; the latter is an explicit Gitee backfill and
+does not claim that OSS has been repaired. Gitee repair requires `GITEE_TOKEN`,
+`GITEE_USER`, and `GITEE_REPO`; OSS repair requires `OSS_ACCESS_KEY_ID`,
+`OSS_ACCESS_KEY_SECRET`, `OSS_ENDPOINT`, and `OSS_BUCKET` (with optional
+`OSS_PREFIX`) as Actions secrets. Missing credentials fail the selected repair
+closed.
+
 ## Handoff Checklist
 
 Before handoff, include:

docs/ci-pr-gates.md

@@ -0,0 +1,249 @@
+# CI — PR 合入门禁
+
+The pull-request admission layer has exactly nine required external contexts:
+
+| Required context | Contract |
+|---|---|
+| `Lint` | Stable PR revision/risk classification plus applicable formatting, `go vet`, and Actionlint |
+| `Test` | Tier-selected race/unit/release-script tests plus representative cross-platform compilation |
+| `Coverage` | Scope-matched overall non-regression and 100% changed-code coverage |
+| `Policy` | Repository policy and the fail-closed CHANGELOG contract |
+| `Edition` | Edition contract tests |
+| `Interface Integrity` | CLI, Schema, Skill, and stable-release compatibility |
+| `AI Behavior` | Base-owned policy for PRs labeled `ai-generated` |
+| `CLI Smoke` | Offline help for every public top-level command |
+| `Mock MCP` | HTTP and stdio MCP lifecycle smoke tests |
+
+The workflow display name is `CI`. Parallel helper
+jobs may implement `Test` and `Coverage`, but they are not ruleset contexts.
+Do not require an aggregate alias or a downstream integration check in place of
+the nine contracts above.
+
+`AI Behavior` is evaluated by a `pull_request_target` workflow that never
+checks out or executes PR code. It writes the exact `AI Behavior` status to the
+current PR head. Its Files API read is bracketed by base/head revision checks,
+so a synchronize race fails closed. The same workflow supplies a successful
+`AI Behavior` check run on protected `main` pushes for release governance.
+
+## Exact CHANGELOG-only fast path
+
+A pull request qualifies only when GitHub reports exactly one changed file,
+that file is an in-place modification of `CHANGELOG.md`, and the base and head
+both retain it as a regular non-executable `100644` blob. Add, delete, rename,
+symlink, executable-mode, and second-file changes do not qualify.
+
+`Lint` classifies the Files API result only after verifying that the API's base
+and head equal the event revision both before and after pagination. `Policy`
+checks out GitHub's PR merge ref and verifies its parents:
+
+```text
+HEAD^1 = pull_request.base.sha
+HEAD^2 = pull_request.head.sha
+```
+
+It then runs:
+
+```sh
+./scripts/policy/check-changelog-pr.sh \
+  --fast-path "$PR_BASE_SHA" HEAD
+```
+
+Because the verified PR diff contains only `CHANGELOG.md`, the validator and
+its policy dependencies in that merge tree are byte-for-byte the current base
+versions. Validation targets the synthetic merge tree, not the feature-branch
+tree, so a stale branch cannot supply an older validator or combine with newer
+base notes into an invalid final CHANGELOG.
+
+All nine admission contexts are still emitted and must succeed. Expensive
+implementation helpers are skipped; the named contexts record that their code
+surface is unaffected.
+
+The protected `main` push keeps that fast path only when all of these
+fail-closed conditions hold:
+
+- the event is a non-forced update of the existing `refs/heads/main`;
+- the event `after` SHA is the exact workflow SHA, and both event SHAs are
+  complete, non-zero commit IDs;
+- GitHub's comparison reports the previous main tip as the unique linear merge
+  base, with no commits behind it;
+- the complete resulting tree diff is exactly one in-place modification of
+  `CHANGELOG.md`;
+- the previous main tip already has successful GitHub Actions checks for all
+  nine Code Admission contexts.
+
+`Policy` then independently checks out the pushed revision and runs the same
+`check-changelog-pr.sh --fast-path` contract from the event's `before` SHA to
+its `after` SHA. If identity, ancestry, file scope, tree mode, CHANGELOG
+content, or predecessor admission cannot be proved, classification falls back
+to the complete main admission suite. A source change can therefore never
+inherit the CHANGELOG-only result.
+
+Any PR that touches `CHANGELOG.md` but also changes another file runs the same
+content contract in `Policy` with `--content-only`. That mode permits the
+second file but still rejects invalid dates or versions, missing bullets,
+placeholder `TODO`/`TBD`, unmanaged-section changes, and unsafe tree modes.
+Adding a second file therefore cannot bypass CHANGELOG validation.
+
+## Risk tiers and downstream boundaries
+
+`Lint` resolves the complete base/head diff before any helper is skipped.
+Unknown or truncated input fails closed into the high-risk tier.
+
+| Tier | Selection | Admission work |
+|---|---|---|
+| Documentation-only | Only prose/documentation assets; no executable, generated, workflow, packaging, or interface surface | Documentation and repository-asset validation; expensive code helpers skip while every required context still succeeds |
+| Standard | Ordinary code change with a stable package graph | Race tests for changed Go packages and their reverse dependencies; candidate and merge-base coverage over the same impacted scope and `coverpkg`; representative Darwin/Windows compilation |
+| High-risk / protected `main` | Workflow/policy, package add/remove/rename, generated Schema/registry, platform, auth/keychain, installer, packaging, release, transport, recovery, or an unprovable infrastructure classification | Complete race suite and full native macOS/Windows tests, plus every affected domain gate |
+
+Domain helpers (`Edition`, `Interface Integrity`, `CLI Smoke`, and `Mock MCP`,
+for example) execute their substantive suites when the diff can affect that
+contract or when the high-risk tier is selected. Otherwise their stable named
+contexts still report a successful, explicit unaffected result. Release-script
+tests follow the same impact rule. This preserves the ruleset contract without
+charging every developer for unrelated work.
+
+Platform-sensitive changes additionally run native changed-code coverage.
+Protected `main` always runs native tests; generic portable changes are held to
+the Linux changed-code gate rather than being forced to manufacture
+platform-only coverage.
+
+Complete `Multi-profile E2E` is not a PR admission context. It belongs to the
+`Main Integration — 主干集成` workflow and runs only after a push to `main` (or
+an explicit manual dispatch). A failing downstream run remains a real
+regression and must be repaired, but it must not be represented by a synthetic
+successful PR check.
+
+```mermaid
+flowchart TB
+  PR["Pull request"] --> ADMISSION["CI"]
+  ADMISSION --> L["Lint"]
+  ADMISSION --> T["Test"]
+  ADMISSION --> C["Coverage"]
+  ADMISSION --> P["Policy"]
+  ADMISSION --> E["Edition"]
+  ADMISSION --> I["Interface Integrity"]
+  ADMISSION --> A["AI Behavior"]
+  ADMISSION --> S["CLI Smoke"]
+  ADMISSION --> M["Mock MCP"]
+  ADMISSION --> MAIN["Protected main"]
+  MAIN --> NATIVE["Full native platform matrix"]
+  MAIN --> E2E["Multi-profile E2E"]
+  MAIN --> RELEASE["Release delivery"]
+```
+
+## Review ownership and auto-merge
+
+A base-owned `pull_request_target` workflow routes newly opened, updated,
+reopened, or newly ready PRs targeting `main` to one eligible peer reviewer. It
+does not check out or execute PR code, excludes both the author and the known
+latest pusher, and balances the open requested-review load across the reviewed
+maintainer pool. A current-head approval or change request is preserved; after
+a new push, stale activity does not suppress a fresh request, and an
+outstanding change requester is preferred for continuity.
+
+The branch ruleset keeps one human approval and all nine strict required
+contexts, and requires someone other than the latest pusher to approve after
+the most recent head update. Repository auto-merge is enabled for ready PRs,
+so a PR merges after that approval and the current revision's nine checks are
+green. If `main` advances, strict checks rerun before merge. The reviewer
+router is orchestration, not a quality context, and must not be added to the
+ruleset.
+
+## Running focused gates locally
+
+Run the contracts relevant to the change. Ordinary contributors are not
+expected to repeat every CI job locally:
+
+```sh
+make build
+make policy
+make interface-integrity BASE_REF=<merge-base> STABLE_REF=<stable-tag> CANDIDATE_REF=<candidate-sha>
+make schema-compatibility BASE_REF=<merge-base> STABLE_REF=<stable-tag> CANDIDATE_REF=<candidate-sha>
+make skill-command-integrity
+make cli-smoke
+make mock-mcp-smoke
+go test -v -count=1 ./pkg/editiontest/...
+```
+
+CI 先解析并核对精确的 merge-base、最近可达且未撤回的 stable GA tag 和已提交的 candidate
+SHA，再调用 `make authoritative-interface-integrity`。本地 `make interface-integrity`
+与该 CI target 都只委托给同一个 modern authoritative wrapper，不存在第二个比较
+入口。省略 `BASE_REF` 时本地 target 默认比较 `origin/main`，省略 `STABLE_REF` 时自动
+选择该 base 可达且未撤回的最近 stable GA tag，省略 `CANDIDATE_REF` 时比较已提交的 `HEAD`。
+需要逐字复现某次 CI 时，应显式传入该次运行记录的 merge-base、stable tag 和
+candidate SHA。
+
+`make update-interface-baseline` / `make reset-interface-baseline` 只维护
+`test/fixtures/cli-interface-baseline.txt` 这一份非权威 CLI Smoke fixture。底层旧
+`check-interface-baseline.sh` 不再作为本地或 CI 的兼容性审批入口，也不能用于批准
+flag 迁移。
+
+Schema compatibility 使用同一组 base、stable、candidate refs 和同一份 base-owned flag
+migration ledger。merge-base-owned checker 分别规范化 merge-base 与 stable 的完整
+Schema，并让 candidate 对两份历史 contract 独立执行检查；它只把已通过 Interface
+lifecycle 的 exact rename 规范化到当前历史副本，不会维护第二份 allowlist，也不会
+放宽其他 Schema 历史字段。
+
+For an exact CHANGELOG-only branch:
+
+```sh
+base_ref=$(git merge-base HEAD origin/main)
+./scripts/policy/check-changelog-pr.sh --fast-path "$base_ref" HEAD
+```
+
+`make coverage-gate` is an enforcement step, not a profile generator. For a
+standard PR, CI derives changed packages and their reverse-dependency test
+closure, then generates candidate and merge-base profiles with the same test
+scope and `coverpkg`. High-risk and protected-main runs use the complete
+profiles. Supporting and (when platform-selected) native profiles are
+generated before the aggregate `Coverage` context evaluates them. The
+aggregate and native gates require 100% coverage for changed executable Go
+statements. Overall coverage remains an unrounded, zero-tolerance,
+scope-matched merge-base non-regression check. Candidate and baseline profiles
+are evaluated by the same block-deduplicating checker; supporting policy and
+shortcut profiles contribute to changed-code coverage only. The checked-in
+badge is presentation only and is never read as a gate input.
+
+CLI 兼容检查只使用 modern Interface Snapshot 这一处权威比较 seam，并从 PR
+merge-base 和最近的可达 stable release 生成权威快照。本治理机制合入后，
+merge-base 拥有生成器、比较器和已审批迁移清单，因此 candidate 不能通过修改
+helper、fixture 或在同一 PR 新增 self-approval 记录来放行 breaking change。首次
+bootstrap 仍由 merge-base 已有的 modern helper 做无豁免比较，并只接受 candidate
+提交中的规范空清单；完整边界见下方治理文档。
+
+精确的两阶段 flag 迁移生命周期见
+[CLI flag 兼容迁移治理](cli-interface-flag-migrations.md)。治理 PR 只能在
+surface 未变化时新增 `pending`；后续产品 PR 达到审批的精确 surface 后，才能
+消费 base-owned 记录并改为 `consumed`。在 main 与 stable 都达到 after 状态前
+必须保留该回执，之后再由单独 PR 清理。机制只放行记录中的 legacy
+visible-to-hidden，以及 canonical required 新增或提升；删除、type、scope、
+shorthand、no-opt 和任何无关漂移仍然阻塞。Schema 可以新增；历史 product、
+tool、parameter、mapping、positional execution、constraint 与 safety 语义继续
+受保护。`alias_of` 只是一项由 `FlagSpec.Aliases` 产生的框架关系证据，不是 payload
+等价证明；产品 PR 仍须证明 canonical 与 legacy 的最终运行 payload 等价并在 transport
+前拒绝冲突输入。当前迁移清单为空，不授权 PR #904。
+
+## Required GitHub repository settings
+
+The `main` quality ruleset must enable strict required-status-check policy
+(`strict_required_status_checks_policy=true`) so a PR is revalidated whenever
+`main` advances. It must require these exact contexts and no legacy aliases:
+
+- `Lint`
+- `Test`
+- `Coverage`
+- `Policy`
+- `Edition`
+- `Interface Integrity`
+- `AI Behavior`
+- `CLI Smoke`
+- `Mock MCP`
+
+Do not require helper jobs, `Multi-profile E2E`, or an aggregate admission
+alias. Update ruleset contexts only after the new names have appeared on the
+protected branch, so a rename cannot silently remove enforcement or leave an
+unproducible required context.
+
+The branch ruleset also requires one approval after the latest push. Enable
+repository auto-merge and automatic head-branch deletion; keep the base-owned
+reviewer router outside the required-context list.

docs/cli-interface-flag-migrations.md

@@ -0,0 +1,176 @@
+# CLI flag 兼容迁移治理
+
+本文定义一种受控迁移：保留旧 flag 的可执行兼容性，但把它从 Help 与 Agent Schema 中隐藏，并将新的规范 flag 提升为必填。它只解决这一种精确变更，不是通用 breaking-change 豁免。
+
+同名 flag 的精确类型迁移属于另一类评审机制，只能进入
+`internal/interfacesnapshot/reviewed.go` 与 legacy smoke helper 的镜像表；flag rename
+只能进入本文的 JSON lifecycle ledger。一项迁移不得跨两种机制组合授权。
+
+## 唯一比较入口与信任边界
+
+PR 与本地兼容性审批的唯一权威比较入口是 modern Interface Snapshot：
+
+- `cmd/interface-snapshot` 生成和比较快照；
+- `internal/interfacesnapshot` 实现兼容规则和迁移生命周期；
+- `scripts/policy/check-command-compatibility.sh` 组装 candidate、PR merge-base 和最近可达且未撤回的 stable GA 三份快照；
+- `scripts/policy/check-authoritative-interface-baselines.sh` 只保留为 Makefile 的兼容包装，不再维护第二套判断逻辑。
+
+紧随其后的 Schema compatibility 不是第二份审批清单。它从同一 merge-base-owned
+ledger 和同一组三方 Interface Snapshot 取得已经完成 lifecycle 校验的
+authorization。merge-base-owned checker 会分别规范化 merge-base 与 stable 的完整
+Schema，并让 candidate 对两份历史 contract 独立执行检查；授权的 flag rename 只会
+精确投影到当前被检查的历史副本。candidate 不能为 CLI 与 Schema 分别提供两套例外。
+
+`make interface-integrity` 也调用上述 authoritative wrapper；默认 base 为
+`origin/main`，stable 可由包装脚本自动解析，candidate 默认为已提交的 `HEAD`。旧
+`scripts/policy/check-interface-baseline.sh` 只供
+`make update-interface-baseline` / `make reset-interface-baseline` 维护非权威 CLI
+Smoke fixture，不参与迁移审批。
+
+直接调用 `interface-snapshot compare` 时，只要提供 migration manifest 参数，就必须
+同时提供 `--base` 与 `--stable`；核心 lifecycle 也拒绝缺失 stable 的非空清单，避免
+调用方因漏传历史参考而提前清理 consumed receipt。
+
+PR merge-base 同时拥有快照生成器、比较器和已审批清单。门禁用这套 base-owned helper 检查同一个已提交 candidate revision、merge-base 与 stable，candidate 不能通过修改自己的 Go 比较 helper 来放宽规则。candidate 中的清单只参与迁移状态流转，不能批准同一个 PR 引入的接口变化。首次引入本机制时，merge-base 尚无迁移解析器；bootstrap 会用 merge-base 已有的 modern Interface Snapshot 做不带豁免的普通比较，并只接受 candidate 中逐字匹配的空清单，不会让 candidate 新增的 comparator 决定本 PR 是否兼容。bootstrap 无法让旧 helper 证明新治理实现本身正确，因此本治理 PR 的新 parser、lifecycle、launcher 与 hostile tests 仍是必须由真人评审的受保护策略变更；它们合入后才成为后续 PR 的 base-owned authority。
+
+这条边界保护比较规则和审批数据，不是任意代码沙箱。GitHub workflow / launcher 的变更仍由仓库保护规则和真人评审负责；candidate Cobra 构建也会执行 candidate 代码，因此对同一 runner 上的主动恶意代码，需要独立进程或文件系统隔离，不能把本门禁描述成已经解决。
+
+已审批清单固定为：
+
+```text
+scripts/policy/interface-migrations/approved-flag-migrations-v1.json
+```
+
+清单使用严格 JSON 解析：版本、字段名大小写、JSON 值类型、命令路径和 flag 名都必须精确；拒绝重复键、未知键、scalar `null` 与尾随 JSON 值，`reason` 不能为空；禁止 `*`、`?`、前缀规则或其他 wildcard。当前清单为空，因此本治理 PR **不授权 PR #904 或任何产品接口变化**。
+
+## 两阶段迁移与回执清理
+
+每条迁移以 `(command, legacy flag, canonical flag)` 为唯一精确键，并经历以下生命周期：
+
+| 阶段 | PR 可以做什么 | 必须满足的快照状态 |
+|---|---|---|
+| 1. 治理审批 | 新增 `state: pending` 的精确记录；不得在同一个 PR 修改产品 surface | candidate 和 merge-base 都与记录中的 `before` 完全一致；该记录不改变 stable 的判断 |
+| 2. 产品迁移 | merge-base 已拥有 `pending` 后，按记录一次性切到精确 `after`，并把记录改为 `state: consumed` | legacy 仍存在但由 visible 变 hidden，且声明 `alias_of`；canonical 达到记录的必填状态 |
+| 3. 保留回执 | 产品 PR 合入后，如果 stable 仍是 `before`，继续保留 `consumed` | merge-base 或 stable 仍有任一份尚未达到 `after` |
+| 4. 单独清理 | 当 merge-base 和 stable 都已经是 `after`，在后续 PR 删除该记录 | 两份参考快照均精确匹配 `after`；继续保留过期回执会被门禁拒绝 |
+
+因此，新增 `pending` 和修改产品 surface 不能发生在同一个 PR；candidate 自己新增的记录不能 self-approve。迁移也不能部分执行：legacy、canonical、`alias_of` 或状态只要有一项不匹配，门禁即失败。
+
+下面只是清单结构示例，不代表已审批命令；实际字段必须从 Interface Snapshot 核对：
+
+```json
+{
+  "version": 1,
+  "migrations": [
+    {
+      "command": "dws chat message recall",
+      "legacy": {
+        "name": "msg-id",
+        "before": {
+          "present": true,
+          "type": "string",
+          "required": true,
+          "scope": "local"
+        },
+        "after": {
+          "present": true,
+          "type": "string",
+          "hidden": true,
+          "scope": "local",
+          "alias_of": "message-id"
+        }
+      },
+      "canonical": {
+        "name": "message-id",
+        "before": { "present": false },
+        "after": {
+          "present": true,
+          "type": "string",
+          "required": true,
+          "scope": "local"
+        }
+      },
+      "state": "pending",
+      "reason": "保留旧 argv 兼容性，并将规范 flag 设为唯一可见入口"
+    }
+  ]
+}
+```
+
+产品迁移 PR 必须保持同一条记录的命令、flag、before/after 和 reason 不变，只把 `pending` 改成 `consumed`。
+
+## `alias_of` 是框架来源的受评审关系证据
+
+`alias_of` 不是 Schema 同义词、参数概念词典或任意文字声明。它只能由 `FlagSpec.Aliases` 写入，并与内部 origin `corecmd.flag_spec_aliases.v1` 成对出现；每次 Interface Integrity 都会在已提交的 detached candidate 上执行源码门禁，禁止其他生产文件写入或复刻这些 evidence token。Interface Snapshot 会验证：
+
+- legacy 与 canonical 位于同一个可执行命令；
+- canonical flag 确实存在；
+- legacy 与 canonical 类型一致；
+- legacy 不是指向自身，也不存在 alias chain；
+- legacy 的 after 状态精确指向该记录中的 canonical flag。
+
+通过命令框架声明 `FlagSpec.Aliases` 时，框架会自动注册隐藏的兼容 flag，并写入两项 relation annotation；仅手写 `alias_of`、伪造 origin、重复值或不精确值都会让快照生成失败。不要用 Schema overlay、迁移清单或手写 Cobra annotation 伪造关系。
+
+这项关系证据只证明 legacy/canonical 经过受控框架路径建立关系，不证明最终 transport payload 等价，也不会替产品代码实现命令特有的值同步。当前框架还禁止把
+`MarkRequired` 与 `FlagSpec.Aliases` 直接组合，因为 Cobra 的 hard-required
+校验只识别 canonical spelling。若产品迁移同时需要 canonical 的 Cobra required
+标记和 legacy spelling，产品 PR 必须提供明确的运行时方案，并通过 canonical / legacy
+最终 payload 等价、同值输入一致、冲突输入在 transport 前失败、legacy 仍可调用但 Help 隐藏等测试；迁移清单和 relation evidence 都不能替代这些证明。
+
+## 豁免边界
+
+一条 base-owned、状态正确且前后快照精确匹配的记录，只会从普通兼容报告中移除以下两类预期 finding：
+
+1. legacy flag 的 `flag_became_hidden`（visible → hidden）；
+2. canonical flag 的 `required_flag_added`（新增时即必填）或 `flag_became_required`（已有 flag 从可选变必填）。
+
+以下变化仍按普通兼容规则阻塞，不能被迁移记录掩盖：
+
+- 删除 legacy、canonical、命令或其他 flag；
+- flag 类型或迁移记录中的 scope、shorthand、`no_opt` 漂移；
+- `alias_of` 缺失、指向变化或 alias chain；
+- 命令路径及任何无关的阻塞性接口变化；
+- 不精确、部分完成、超出记录范围的 surface 变化。
+
+## Schema 投影边界
+
+Agent-visible command 会把 visible Cobra flag 投影为 Schema parameter，因此合法的
+legacy hidden 迁移会同时表现为历史 parameter 消失，constraint member 也可能从
+legacy 名改为 canonical 名。Schema adapter 只接受已经由三方 Interface Snapshot
+判定为 authorized 的迁移，并按 tool 的精确 `primary_cli_path` 绑定：
+
+- reference 仍处于 `before` 且该 flag 有 Schema surface 时，baseline legacy parameter
+  必须存在，candidate legacy parameter 必须消失，candidate canonical parameter 必须存在；
+  如果 baseline 只有 canonical、没有 legacy，则 adapter 不得借 CLI ledger 提升
+  `required` / `cli_required` 或重写 constraint；
+- rename 前后的 `type`、`property`、`interface_type`、default、format、enum 与
+  `required_when` 必须完全一致；
+- `required` / `cli_required` 只能保持不变或按审批从 `false` 提升为 `true`，禁止降低；
+- constraint 只允许在同一 tool 内按已枚举的 legacy → canonical map 做 member 替换、
+  排序与去重；group kind、非迁移 member 或 group 增删仍然阻塞；
+- 多个 legacy 指向同一 canonical 时，所有历史 parameter signature 必须一致，否则
+  fail closed。
+
+adapter 先构造经过上述验证的历史 contract 副本，再调用原 Schema checker；它不会按
+错误字符串删除 finding。这样既能处理纯 rename，也能阻止“旧 required 参数改名后意外
+变为 optional”或 property 漂移等伪兼容。`consumed` 回执在 merge-base Schema 已经处于
+canonical-only `after` 状态时不需要再次投影；adapter 保持 baseline 不变，由原 checker
+验证 candidate 是否仍与该 canonical contract 兼容。
+
+## 本地验证
+
+先确保 merge-base 和 stable tag 已在本地，然后运行与 CI 相同的权威门禁：
+
+```sh
+make interface-integrity \
+  BASE_REF=<merge-base> \
+  STABLE_REF=<stable-tag> \
+  CANDIDATE_REF=<candidate-sha>
+
+make schema-compatibility \
+  BASE_REF=<merge-base> \
+  STABLE_REF=<stable-tag> \
+  CANDIDATE_REF=<candidate-sha>
+```
+
+`STABLE_REF` 必须解析到从该 merge-base 可达的最高未撤回 stable GA tag；primary checker 会按 release contract 独立核对，不能用任意 after commit 或已撤回版本提前清理回执。包装脚本可以在省略时自动解析。`CANDIDATE_REF` 省略时固定为命令启动时的已提交 `HEAD`；评审和复现 CI 时应显式传入 candidate SHA，避免 surface 与清单来自不同 revision。

docs/command-domain-model.md

@@ -0,0 +1,209 @@
+# command 领域模型
+
+本文档描述 `internal/corecmd` 包的领域模型——类型、概念及其关系。
+
+## 核心模型图
+
+```
+┌─────────────────────────────────────────────────────────────────────┐
+│                         corecmd.Spec                                  │
+│                    (一个叶子命令的完整契约)                             │
+├─────────────────────────────────────────────────────────────────────┤
+│                                                                     │
+│  ┌─── CLI 表面 ───┐   ┌─── 参数声明 ───────────────────────────┐   │
+│  │ Use            │   │ FlagSpec[]                              │   │
+│  │ Short          │   │  ├─ Name / Kind / Default              │   │
+│  │ Long           │   │  ├─ Required / MarkRequired            │   │
+│  │ Example        │   │  ├─ Aliases[] / EnvVar (回退链)         │   │
+│  └────────────────┘   │  ├─ Bind / Transform / OmitEmpty       │   │
+│                       │  └─ Enum / Format / SchemaDescription  │   │
+│                       │                                        │   │
+│                       │ Constraint[]                            │   │
+│                       │  ├─ at_least_one                       │   │
+│                       │  ├─ exactly_one                        │   │
+│                       │  └─ mutually_exclusive                 │   │
+│                       │                                        │   │
+│                       │ ConstParams map[string]any             │   │
+│                       └────────────────────────────────────────┘   │
+│                                                                     │
+│  ┌─── 安全模型 ──────────────────────────────────────────────────┐  │
+│  │  Safety contract.SafetySpec                                       │  │
+│  │  ├─ Effect       (read / write / destructive)                │  │
+│  │  ├─ Risk         (low / medium / high)                       │  │
+│  │  ├─ Confirmation (not_required / user_required) ──▶ 运行时门 │  │
+│  │  └─ Idempotency  (idempotent / retryable / …)                │  │
+│  │                                                               │  │
+│  │  四字段彼此独立；同一值同时供运行时与 Schema 使用              │  │
+│  │  ConfirmFirst: bool (只控制确认门顺序)                         │  │
+│  └───────────────────────────────────────────────────────────────┘  │
+│                                                                     │
+│  ┌─── Contract 声明 (Agent 可见的元数据) ────────────────────────┐  │
+│  │  ContractDecl                                                   │  │
+│  │  ├─ Title / Description                                      │  │
+│  │  ├─ contract.DryRunSpec     {PreviewKind, RemoteReads}       │  │
+│  │  ├─ contract.InterfaceSpec  {Mode, Availability, Reason, Ref}│  │
+│  │  ├─ contract.SelectionSpec  {AgentSummary, UseWhen, AvoidWhen│  │
+│  │  │                          Prerequisites, Tips, Examples}    │  │
+│  │  ├─ contract.ToolIdentitySpec {ProductID, CanonicalPath, …}  │  │
+│  │  └─ Positionals[]  {Name, Type, Required, Variadic}          │  │
+│  └───────────────────────────────────────────────────────────────┘  │
+│                                                                     │
+│  ┌─── 执行体 (恰好一个) ─────────────────────────────────────────┐  │
+│  │  Invoke(Ctx, toolArgs)     ← #830 过渡：单步派发（目标 mcpbind）│  │
+│  │  Orchestrate(Ctx)          ← #830 过渡：多步编排（目标 Handler）│  │
+│  │  RunE(cmd, args)           ← 逃生舱：完全自定义               │  │
+│  └───────────────────────────────────────────────────────────────┘  │
+│                                                                     │
+│  ┌─── 钩子 ─────────────────────────────────────────────────────┐  │
+│  │  Validate(cmd, args)   ← 条件式业务校验（约束表达不了的）      │  │
+│  │  PostMount(cmd)        ← 挂载收尾（设置 Args 等 cobra 属性）   │  │
+│  └───────────────────────────────────────────────────────────────┘  │
+└─────────────────────────────────────────────────────────────────────┘
+```
+
+## 构建与执行流
+
+```
+corecmd.Spec ──── corecmd.New() ────▶ cobra.Command ──── 用户执行 ────▶ Ctx
+                     │                    │                            │
+              构建时检查:              注册产物:                    执行上下文:
+              • validateDispatchDecl   • Flags + Aliases            • Str(name)
+              • validateSafetySpec     • Annotations (Schema)       • Int(name)
+              • validateContractDecl     • Long (约束 help)           • Bool(name)
+              • RegisterFlags          • RunE (管线)                • StrSlice(name)
+              • ValidateConstraintDecls                            • Changed(name)
+              • embedContractIntoSchema                            • DryRun() / Yes()
+              • AnnotateConstraints
+              • PostMount
+```
+
+## 领域概念
+
+| 概念 | 类型 | 职责 |
+|------|------|------|
+| **corecmd.Spec** | struct | 一个命令的完整契约（声明 + 执行） |
+| **FlagSpec** | struct | 一个参数的注册、回退链、绑定规则 |
+| **Constraint** | struct | 参数间的关系约束 |
+| **Safety** | contract.SafetySpec | 运行时与 Schema 共用的安全契约 |
+| **ContractDecl** | struct | Agent 可见的完整工具规格声明 |
+| **contract.SelectionSpec** | struct | Agent 选择该工具的语义指引 |
+| **contract.InterfaceSpec** | struct | 工具的接口模式与可用性 |
+| **contract.DryRunSpec** | struct | dry-run 能力声明 |
+| **contract.ToolIdentitySpec** | struct | 工具在注册表中的身份标识 |
+| **contract.RuntimeSchemaPositional** | struct | 有序位置参数声明 |
+| **Ctx** | struct | 执行上下文（类型安全的 flag 读取） |
+| **New** | func | 统一构建器（`corecmd.Spec` → `*cobra.Command`） |
+
+## SafetySpec
+
+`corecmd.Spec.Safety` 使用 `corecmd/contract.SafetySpec`（无 cli 类型别名），没有 command 自定义 Risk/Safety 枚举，也没有 `SafetyDecl` 覆盖层：
+
+```go
+Safety: contract.SafetySpec{
+    Effect:       "write",
+    Risk:         "high",
+    Confirmation: "user_required",
+    Idempotency:  "unknown",
+}
+```
+
+- `Confirmation == "user_required"` 时执行确认；`--yes` 和 `--dry-run` 可跳过交互。
+- `Effect`、`Risk`、`Idempotency` 不参与确认决策，也不会改写 `Confirmation`。
+- 任意一个字段非空时，四个字段必须全部显式声明；构建时拒绝部分声明。
+- 完全空值仅作为历史只读默认，最终发布为 `read/low/not_required/idempotent`。
+
+## FlagSpec 有效值回退链
+
+框架统一的 flag 值解析顺序：
+
+```
+显式主 flag (Changed)
+    │ 空?
+    ▼
+隐藏别名 (Changed, 按声明序)
+    │ 空?
+    ▼
+环境变量 (EnvVar)
+    │ 空?
+    ▼
+注册默认值 (Default)
+    │ 空?
+    ▼
+ArgDefault (兜底)
+```
+
+各 Kind 的特殊行为：
+
+| Kind | 入参条件 | 回退链 |
+|------|----------|--------|
+| KindString | 有效值非空（或 !OmitEmpty） | 完整参与 |
+| KindInt | 值 ≠ 0（putInt 语义） | 完整参与 |
+| KindBool | Changed 时入参（显式 false 也下发） | 不参与别名/env 回退 |
+| KindStringSlice | 存在非空元素 | 仅 Changed 的主 flag/alias |
+
+## Constraint 约束
+
+声明式跨 flag 关系，构建时校验合法性，运行时统一执行：
+
+| Kind | 语义 | 错误文案示例 |
+|------|------|-------------|
+| `at_least_one` | 至少提供一个 | "请至少指定 --a、--b 之一" |
+| `exactly_one` | 恰好提供一个 | "请指定 --a、--b 之一" / "只能指定其一" |
+| `mutually_exclusive` | 最多提供一个 | "参数 --a、--b 互斥，只能指定其一" |
+
+"是否提供"的判定复用有效值回退链（显式主 flag → 别名 → env），注册默认值不算作已提供。
+
+## ContractDecl 子结构
+
+### contract.SelectionSpec（Agent 选择指引）
+
+```go
+contract.SelectionSpec{
+    AgentSummary: "一句话描述工具做什么",
+    UseWhen:      []string{"在什么场景下应该选择这个工具"},
+    AvoidWhen:    []string{"什么场景不应该用，应该用什么替代"},
+    Prerequisites: []string{"使用前提条件"},
+    Tips:          []string{"使用技巧"},
+    Examples:      []string{"dws dev app create --name Bot --dry-run"},
+}
+```
+
+### contract.InterfaceSpec（接口模式）
+
+```go
+contract.InterfaceSpec{
+    Mode:         "composite",  // local / mcp / composite
+    Availability: "available",  // available / unavailable
+    Reason:       "...",        // composite/unavailable 时的原因
+    Ref: &contract.InterfaceRefSpec{ // mcp 时的 ref
+        ProductID: "...",
+        RPCName:   "...",
+    },
+}
+```
+
+### contract.DryRunSpec（dry-run 能力）
+
+```go
+contract.DryRunSpec{
+    PreviewKind: "invocation",  // invocation / request / plan
+    RemoteReads: false,         // dry-run 时是否发起远端读
+}
+```
+
+## 执行体三选一
+
+| 执行体 | 适用场景 | 框架做了什么 |
+|--------|----------|-------------|
+| **Invoke** | #830 过渡单步派发（生产仍用；目标 mcpbind） | 框架完成 required→constraint→validate→buildArgs→confirm，传入装配好的 toolArgs |
+| **Orchestrate** | #830 过渡多步编排（生产仍用；目标 Handler） | 框架完成 required→constraint→validate→confirm，传入 Ctx 自行组装调用 |
+| **RunE** | 逃生舱 | 框架仍执行 Safety 确认，具体业务执行完全自定义 |
+
+## 设计不变量
+
+1. **一个 corecmd.Spec = 一个叶子命令的全部事实**
+2. **声明面绝不调用后端**——command 是 dispatch-agnostic
+3. **执行面绝不发明 CLI 表面**——业务 flag 必须在 Flags 声明
+4. **构建时拦截 > 运行时报错**——声明错误 panic 在注册阶段
+5. **SafetySpec 是单一事实源**——Confirmation 驱动运行时，其余字段原样进入 Schema
+6. **声明即 review**——代码中的 Schema 经 code review 后直接投影，不依赖外部 hint 文件

docs/command-framework-architecture.md

@@ -0,0 +1,286 @@
+# 命令框架架构
+
+本文档描述 `internal/corecmd` 统一命令框架的当前架构，面向框架使用者和维护者。
+
+## 概览
+
+```
+用户输入 → cobra 命令树 → corecmd.New() → 运行时管线 → 后端派发
+```
+
+命令框架将 CLI 命令的**声明**与**执行**分离：
+
+- **声明面** — 数据字段描述命令是什么（flag、约束、SafetySpec、Contract 元数据）
+- **执行面** — 钩子函数描述命令做什么（校验、派发、编排）
+
+框架负责：flag 注册、有效值回退链、required/约束校验、SafetySpec 确认、toolArgs 装配、Agent Runtime Schema 投影。
+
+## 核心类型
+
+### corecmd.Spec
+
+统一的类型化命令规格，是框架的核心数据结构：
+
+```go
+type Spec struct {
+    // 声明面
+    Use         string
+    Short       string
+    Long        string
+    Example     string
+    Flags       []FlagSpec
+    Constraints []Constraint
+    Safety      contract.SafetySpec // 运行时与 Schema 的单一安全来源
+    ConfirmFirst bool         // 确认门先于参数校验
+    ConstParams map[string]any
+    Contract   ContractDecl    // 叶子 Contract 声明（非 Catalog Schema）
+
+    // 执行面（恰好一个；Invoke/Orchestrate 为 #830 过渡派发 API，目标 mcpbind+Handler）
+    Invoke      func(c *Ctx, toolArgs map[string]any) error  // 过渡：单步
+    Orchestrate func(c *Ctx) error                           // 过渡：多步
+    RunE        func(cmd *cobra.Command, args []string) error // 逃生舱
+
+    // 钩子
+    Validate    func(cmd *cobra.Command, args []string) error
+    PostMount   func(cmd *cobra.Command)
+}
+```
+
+### SafetySpec（单一安全来源）
+
+`Spec.Safety` 使用 `corecmd/contract.SafetySpec`：
+
+| 字段 | 职责 |
+|------|------|
+| `Effect` | 操作影响：read / write / destructive |
+| `Risk` | 风险等级：low / medium / high |
+| `Confirmation` | 是否需要用户确认：not_required / user_required |
+| `Idempotency` | 幂等性：idempotent / retryable / non_idempotent / unknown |
+
+四个字段彼此独立。框架只读取 `Confirmation` 决定运行时确认，其余字段原样发布到 Schema，不从一个字段机械推导另一个。非空 SafetySpec 必须一次声明完整：
+
+```go
+Safety: contract.SafetySpec{
+    Effect:       "write",
+    Risk:         "high",
+    Confirmation: "user_required",
+    Idempotency:  "unknown",
+},
+```
+
+完全空值保留历史只读默认 `read/low/not_required/idempotent`；不存在 Risk/Safety 枚举或覆盖优先级链。
+
+### FlagSpec
+
+声明一个 flag 的注册方式、有效值回退链、到 toolArgs 的绑定：
+
+```go
+type FlagSpec struct {
+    Name     string       // flag 名（kebab-case）
+    Usage    string       // --help 文案
+    Kind     FlagKind     // String / Int / Bool / StringSlice
+    Default  string       // 注册默认值
+    Required bool         // 框架校验非空
+    Aliases  []string     // 隐藏别名
+    EnvVar   string       // 环境变量回退
+    Bind     string       // toolArgs 键名（空则用 Name）
+    Transform func(string) (any, error)  // 值转换
+    // ...更多字段见源码
+}
+```
+
+### Constraint
+
+跨 flag 关系约束：
+
+```go
+type Constraint struct {
+    Kind  ConstraintKind  // at_least_one / exactly_one / mutually_exclusive
+    Flags []string
+}
+```
+
+## 有效值回退链
+
+flag 解析按以下顺序取值（先命中先生效）：
+
+```
+显式主 flag (Changed) → 隐藏别名 (Changed) → 环境变量 → 注册默认值
+                                                            │
+                                              ArgDefault ←──┘ (兜底)
+```
+
+- KindBool：仅 Changed 时生效，不参与回退链
+- KindStringSlice：仅主 flag / alias Changed 时生效，元素恒 TrimSpace
+- KindInt：非零才入 toolArgs（putInt 语义）
+
+## 构建时流程
+
+`corecmd.New(spec)` 执行以下构建时检查（失败则 panic）：
+
+1. **validateDispatchDecl** — 恰好一个执行体（Invoke/Orchestrate/RunE）
+2. **validateSafetySpec** — 非空 SafetySpec 的四个独立字段必须完整
+3. **validateContractDecl** — Contract 声明完整性（Description、AgentSummary、UseWhen、AvoidWhen、Examples、Interface）
+4. **RegisterFlags** — flag + alias 注册到 cobra
+5. **ValidateConstraintDecls** — 约束引用的 flag 必须存在
+6. **embedContractIntoSchema** — 投影到 dws.schema.* annotations
+7. **AnnotateConstraints** — 约束渲染到 --help
+8. **PostMount** — 调用方的挂载收尾钩子
+
+## 运行时流程
+
+生成的 `RunE` 按以下顺序执行：
+
+```
+[ConfirmFirst? → ConfirmSafety]      ← 可选：先确认后校验
+  │
+  ▼
+ValidateRequired                     ← 有效值回退链校验
+  │
+  ▼
+ValidateConstraints                  ← 互斥/至少一个/恰好一个
+  │
+  ▼
+Validate hook                        ← 条件式业务校验（可选）
+  │
+  ▼
+BuildArgs                            ← flag → toolArgs 装配
+  │
+  ▼
+ConstParams 合并
+  │
+  ▼
+[!ConfirmFirst? → ConfirmSafety]     ← 默认顺序：校验后确认
+  │
+  ▼
+Invoke(ctx, toolArgs)                ← #830 过渡：单步派发
+  或 Orchestrate(ctx)                ← #830 过渡：多步编排
+```
+
+## 消费方式
+
+### LeafSpec（MCP 直连叶子命令）
+
+```go
+func newDevAppCreateCommand(runner executor.Runner) *cobra.Command {
+    return NewLeafCommand(LeafSpec{
+        Use:     "create",
+        Short:   "创建开放平台企业内部应用",
+        Tool:    devAppCreateTool,
+        Safety: contract.SafetySpec{
+            Effect: "write", Risk: "high",
+            Confirmation: "user_required", Idempotency: "unknown",
+        },
+        ConfirmFirst: true,
+        Flags: []LeafFlag{
+            {Name: "name", Usage: "应用名称 (必填)", Bind: "name",
+             Trim: true, Required: true, RequiredHint: "--name 为必填"},
+        },
+        Contract: ContractDecl{
+            Description: "创建开放平台企业内部应用",
+            DryRun:      &contract.DryRunSpec{PreviewKind: "invocation"},
+            Interface:   &contract.InterfaceSpec{Mode: "composite", Availability: "available", Reason: "create then configure"},
+            Selection: contract.SelectionSpec{
+                AgentSummary: "创建钉钉开放平台应用",
+                UseWhen:      []string{"需要新建企业内部应用"},
+                AvoidWhen:    []string{"应用已存在时用 update"},
+                Examples:     []string{`dws dev app create --name "Bot" --dry-run`},
+            },
+        },
+        Call: devAppCall(runner),
+    })
+}
+```
+
+`NewLeafCommand` 经 `FromLeafSpec()` 归一为 `corecmd.Spec`，再交 `corecmd.New()` 构建。这是**完全托管模式**：声明 + 执行都归 command。
+
+### 声明元数据模式（既有命令补 Contract）
+
+执行体必须冻结时，用同一套 `LeafSpec` 词汇只声明元数据，写在命令字面量旁：
+
+```go
+baseListCmd := &cobra.Command{
+    Use: "list", Short: "获取 AI 表格列表",
+    RunE: func(cmd *cobra.Command, args []string) error { /* 原执行体不动 */ },
+}
+DeclareLeafMetadata(baseListCmd, LeafSpec{
+    Safety: aitableSafetyRead(),
+    Contract: ContractDecl{
+        Description: "列出最近访问的 AI 表格 Base。",
+        Interface:   aitableMCPInterface("list_bases"),
+        Selection: contract.SelectionSpec{
+            AgentSummary: "列出最近访问的 AI 表格 Base。",
+            UseWhen:      []string{"只需浏览最近打开过的 Base 时"},
+            AvoidWhen:    []string{"按名称查找优先 base search"},
+            Examples:     []string{"dws aitable base list"},
+        },
+    },
+})
+```
+
+`DeclareLeafMetadata` 调用 `corecmd.AttachContract` 挂 Safety+Contract；不注册 flag、不接管参数投影。可选 `Validate` 与 `ConfirmSafety` 同挂在 **RunE 包装器**内（不是 PreRunE）。当 `Safety.Confirmation=user_required` 时，用**同一份** SafetySpec 包一层 `ConfirmSafety`，保证执行门禁与 Catalog 同源；无 Validate 时确认推迟到 gated `CallTool`，成功返回却未确认则 fail-closed。迁移态入口；新命令仍应走 `NewLeafCommand`。
+
+### 三档路径（当前可接受）
+
+| 档 | 入口 | 说明 |
+|---|---|---|
+| **Tier1** | `corecmd.New` / `NewLeafCommand` | 完全托管：声明 + 执行都归框架 |
+| **Tier2** | `DeclareLeafMetadata` | helpers 迁移态；**Shortcut 也可采用，可接受** |
+| **Tier3** | 裸 Cobra | 应逐步收；新增裸叶需补声明或精确排除 |
+
+长期展望（非当前硬要求）：更多 Shortcut 可收敛到 mcpbind / 减少仅为参数装配的 `Execute`。**不要**把「Shortcut 必须去掉 Execute / 必须 mcpbind」当作当前门禁；也不要否定 Shortcut + `DeclareLeafMetadata`。
+
+### Shortcut（智能快捷方式，已接入 live mount）
+
+```go
+func mount(s Shortcut) *cobra.Command {
+    return corecmd.New(FromShortcut(s))
+}
+
+spec := FromShortcut(Shortcut{
+    Service: "chat",
+    Command: "+demo",
+    Risk:    RiskHighWrite,
+    Flags:   []Flag{...},
+    Execute: func(rt *RuntimeContext) error { ... },
+})
+```
+
+Shortcut 当前仍保留自身的 `Risk`，adapter 只在边界将它展开成完整
+`contract.SafetySpec`；command/Leaf 不再保留该枚举。Shortcut 的 Cobra
+type/default/usage provenance 保持不变，command 统一补充 Required、Enum 和关系约束投影。
+需要补 Agent Schema 且执行体暂不迁入时，Shortcut 也可走 Tier2
+`DeclareLeafMetadata`（与 helpers 同一路径）。
+
+## 文件结构
+
+| 文件 / 包 | 职责 |
+|------|------|
+| `internal/corecmd/corecmd.go` | 核心类型 + `New` 构建器 + 运行时管线 |
+| `internal/corecmd/contract_decl.go` | ContractDecl 载荷类型 + 声明完整性守卫 |
+| `internal/corecmd/contract/` | 契约 DTO（`SafetySpec` / `ParamDecl` / `ProductDecl` / `ContractFinalPayload`）；**无** Cobra-keyed Final store |
+| `internal/corecmd/runtimeannotate/` | `AnnotateRuntime*` 写注解（框架侧；`cli` 薄 re-export） |
+| `internal/corecmd/contractfinal/` | ContractFinal Cobra store + `RegisterRuntimeContractFinal`（框架侧；`cli` 薄 re-export） |
+| `internal/cli/homology/` | flag/help/schema 同源门禁（`HOM-*`） |
+| `internal/helpers/leaf.go` | LeafSpec 门面：`NewLeafCommand`（完全托管）+ `DeclareLeafMetadata`（声明元数据） |
+| `internal/shortcut/adapter.go` | FromShortcut 完整映射与 Risk 兼容边界 |
+| `internal/shortcut/runner.go` | RuntimeContext；live mount 委托 `corecmd.New(FromShortcut(s))` |
+
+## Schema 投影
+
+声明即 review：代码中的 Contract 声明经过 code review 后直接投影为：
+
+- **Agent Runtime Schema**（`dws.schema.*` Cobra annotations；经 `runtimeannotate` / ContractFinal 嵌入）
+- **运行时组装的 SchemaRegistry / Catalog ToolSpec wire**（`RegisterSchemaSourceRoot` → `ResolveSchemaBuild`；`dws schema` / `--all` / 完整 leaf 载荷）
+- **CommandMeta 投影**（装配 Once 同步缓存 `map[cli_path]CommandMeta`；`ResolveMeta` / `SafetyForCLIPath` / leaf `--help` Safety 稳态 O(1) 读缓存，与 SchemaRegistry 同源）
+- **Dry-run Capabilities**（声明自动索引为 reviewed 能力）
+
+生产权威是 leaf `ContractFinal` / `ProductDecl`（经 `RegisterSchemaSourceRoot` → `ResolveSchemaBuild` 装配进 Catalog）；`InstallBuildTimeAgentMetadataJSON` 仅用于 `cmd_schema_catalog` 的 CI/local dump inject，不是生产交付路径。`schema_agent_metadata/` 与 `schema_hints/` 已退役。不再需要外部 hint 文件维护 selection/metadata/dry-run 信息。Catalog/meta-index 路径不得提交。
+
+## 设计原则
+
+1. **声明 vs 执行分离** — Flags/Constraints/Safety/Contract 是声明；Invoke/Validate/PostMount 是执行
+2. **单一数据源** — 一份声明驱动 --help、Schema、catalog、runtime 校验
+3. **安全字段不互推** — Confirmation 单独驱动确认，Effect/Risk/Idempotency 原样发布
+4. **构建时拦截 > 运行时报错** — 声明不完整在命令注册时 panic，不等到用户触发
+5. **边界兼容** — Shortcut 暂由 adapter 转换，Leaf 直接声明 SafetySpec

docs/command-framework-comparison.md

@@ -0,0 +1,236 @@
+# 命令框架对比：DWS command vs lark-cli vs GWS
+
+本文档对比 DWS（钉钉工作区 CLI）、lark-cli（飞书 CLI）和 GWS（Google Workspace CLI / gcloud）三套命令框架的设计差异。
+
+## 总览对比
+
+| 维度 | DWS (command) | lark-cli | GWS (gcloud) |
+|------|---------------|----------|--------------|
+| 语言 | Go | Go | Python (gcloud) / Go (部分) |
+| CLI 框架 | cobra | cobra | argparse + calliope |
+| 调用底座 | MCP JSON-RPC | Lark REST SDK (`CallAPITyped`) | Google API Client |
+| 命令层次 | 2 层：LeafSpec + Shortcut | 3 层：Shortcuts + API Commands + Raw API | 2 层：surface commands + raw |
+| Schema 来源 | 代码声明投影 | 代码声明 + 运行时 introspection | API Discovery 文档自动生成 |
+| Agent 适配 | 内建 (dws.schema.*) | 内建 (--print-schema) | 外挂 (MCP adapter) |
+
+## 架构对比
+
+### DWS command
+
+```
+corecmd.Spec (声明) → corecmd.New() → cobra.Command
+     │
+     ├── contract.SafetySpec (运行时 + Schema 单一安全来源)
+     ├── FlagSpec[] (参数 + 回退链 + 绑定)
+     ├── Constraint[] (互斥/至少一个)
+     ├── ContractDecl (Agent Selection/DryRun/Interface)
+     │
+     └── Invoke / Orchestrate / RunE (执行)
+```
+
+**核心特点**：
+- 声明与执行严格分离
+- SafetySpec 四个独立字段直接对齐 Agent Runtime Schema
+- 有效值回退链：flag → alias → env → default
+- 框架统一校验、装配、确认、投影
+- Schema 从代码声明直接投影，无外部 hint 文件
+
+### lark-cli
+
+```
+Shortcut (声明) → runner.Mount() → cobra.Command
+     │
+     ├── Risk string (确认行为)
+     ├── Scopes / ConditionalScopes (OAuth 权限)
+     ├── Flag[] (参数 + Enum + Input sources)
+     ├── AuthTypes (user/bot)
+     │
+     ├── DryRun hook → DryRunAPI
+     ├── Validate hook
+     └── Execute hook → RuntimeContext → CallAPITyped
+```
+
+**核心特点**：
+- Execute 内直接调 REST API (`CallAPITyped`)
+- DryRun 是独立 hook（返回结构化 API 计划）
+- 内建 OAuth scope 声明与预检
+- `--print-schema --flag-name` 运行时 introspection
+- 无 Schema 投影层，Agent 通过 introspection 动态发现
+
+### GWS (gcloud 风格)
+
+```
+API Discovery → 代码生成 → surface command
+     │
+     ├── arguments (从 JSON Schema 自动生成)
+     ├── request/response 映射
+     └── 自定义 action hook (少量)
+```
+
+**核心特点**：
+- Schema-first：从 API Discovery 文档自动生成命令
+- 参数直接映射 API 字段（flat schema）
+- 人工 surface command 是 thin wrapper
+- Agent 适配通过 MCP 外部 adapter
+
+## 核心设计差异
+
+### 1. 声明粒度
+
+| 能力 | DWS command | lark-cli | GWS |
+|------|-------------|----------|-----|
+| 参数别名 + 环境变量回退 | ✅ FlagSpec.Aliases + EnvVar | ❌ 无 | ❌ 无 |
+| 声明式约束 (互斥/至少一个) | ✅ Constraint[] | ❌ 只有 Validate hook | ✅ argparse group |
+| 安全契约 | ✅ SafetySpec（effect/risk/confirmation/idempotency） | Risk | 无 |
+| Schema 投影 (Agent metadata) | ✅ ContractDecl 内建 | ⚠️ 运行时 introspection | ❌ 外挂 |
+| 参数绑定 (flag name → API key) | ✅ FlagSpec.Bind | ❌ 手写 | ✅ 自动映射 |
+| ConstParams (固定载荷) | ✅ | ❌ 手写在 Execute | ✅ 隐式 |
+| 确认门顺序可配 (ConfirmFirst) | ✅ | ❌ 固定顺序 | ❌ 无确认机制 |
+
+### 2. 执行模型
+
+| 维度 | DWS command | lark-cli | GWS |
+|------|-------------|----------|-----|
+| 参数装配 | 框架自动 (BuildArgs) | 手写 (`runtime.Str()/Bool()`) | 自动映射 |
+| 派发方式 | Invoke(ctx, toolArgs) | Execute(ctx, runtime) | 自动调用 |
+| 多步编排 | Orchestrate(ctx) | Execute 内链式 CallAPITyped | 不支持 |
+| DryRun | 框架统一 (--dry-run flag) | 独立 DryRun hook 返回 API 计划 | 部分命令支持 |
+| 错误分类 | apperrors 类型化 | errs.Problem 类型化 | HTTP status 映射 |
+
+### 3. Agent 适配
+
+| 维度 | DWS command | lark-cli | GWS |
+|------|-------------|----------|-----|
+| 工具发现 | `dws schema --all` (静态 catalog) | `--print-schema` (运行时) | API Discovery |
+| 选择指引 | contract.SelectionSpec (UseWhen/AvoidWhen) | Description + Tips | 无 |
+| 安全声明 | contract.SafetySpec 直接声明 | Risk string | 无 |
+| dry-run 能力声明 | contract.DryRunSpec (reviewed) | DryRun hook 存在性 | 无 |
+| 接口模式 | contract.InterfaceSpec (local/mcp/composite) | 隐式 (全部 REST) | 隐式 (全部 REST) |
+
+### 4. Schema 生命周期
+
+```
+DWS:      代码声明 → code review → cobra annotation → catalog/metadata JSON
+          (单一数据源，构建时验证完整性)
+
+lark-cli: 代码声明 → 运行时 introspection → Agent 动态发现
+          (无离线 catalog，Agent 必须执行命令才能发现)
+
+GWS:      API Discovery JSON → 代码生成 → surface command
+          (Schema-first，但命令行体验受限于 API 形状)
+```
+
+## 设计哲学对比
+
+### DWS command 的选择
+
+| 选择 | 理由 | 对比 |
+|------|------|------|
+| 框架装配参数 | 消除 N 个命令各写一份 toolArgs 装配 | lark-cli 每个 Execute 手动取 flag 值 |
+| SafetySpec 单一来源 | confirmation 驱动运行时，其余字段原样发布且互不推导 | lark-cli 只有 Risk 一个维度 |
+| 声明式约束 | 构建时校验合法性 + 投影到 Schema + 渲染帮助 | lark-cli 约束隐藏在 Validate 逻辑里 |
+| Schema 构建时投影 | 离线 catalog 支持 Agent 批量发现 | lark-cli 需要逐个命令 introspection |
+| 有效值回退链 | flag → alias → env 统一语义 | lark-cli 别名是独立 Flag 手动关联 |
+| ConfirmFirst | 精确建模遗留语义 | lark-cli 确认始终在 Execute 内 |
+
+### lark-cli 的选择
+
+| 选择 | 理由 | 对比 |
+|------|------|------|
+| 直连 REST API | 精确控制请求/响应，可处理分页/重试 | DWS 通过 MCP 间接调用 |
+| DryRun 返回 API 计划 | Agent 可预览将要发出的真实 HTTP 请求 | DWS dry-run 只展示参数 |
+| OAuth scope 声明 | 框架预检权限，失败提前 | DWS 依赖 MCP 层鉴权 |
+| `--print-schema` introspection | 运行时发现，无需维护离线 catalog | DWS 需要 re-generate |
+| Input sources (file/@path/stdin) | 丰富的输入方式声明 | DWS 无此抽象 |
+| PrintFlagSchema | 单 flag 级别的 JSON Schema 暴露 | DWS 只在 catalog 级别 |
+
+### GWS (gcloud) 的选择
+
+| 选择 | 理由 | 对比 |
+|------|------|------|
+| API Discovery 驱动 | 一份 Schema 生成所有：SDK/CLI/文档 | DWS/lark 手写 |
+| Flat parameter 映射 | API 字段 = CLI flag，零转换 | DWS 需要 Bind 映射 |
+| 无 shortcut 层 | API 粒度即用户粒度 | DWS/lark 有精选层 |
+
+## 代码量对比
+
+| 框架 | 核心框架代码 | 单命令声明开销 | 备注 |
+|------|-------------|---------------|------|
+| DWS command | ~1400 行 (command.go + contract_decl.go) | ~20-30 行 (纯声明) | 框架重、单命令轻 |
+| lark-cli | ~800 行 (runner.go + types.go + common.go) | ~50-150 行 (声明 + Execute 逻辑) | 框架轻、单命令重 |
+| GWS gcloud | ~5000+ 行 (calliope 框架) | ~10 行 (多数自动生成) | 框架最重、单命令最轻 |
+
+## DWS 命令声明示例 vs lark-cli
+
+### DWS (command / LeafSpec)
+
+```go
+NewLeafCommand(LeafSpec{
+    Use:    "create",
+    Short:  "创建应用",
+    Tool:   "create_dev_app",
+    Safety: contract.SafetySpec{
+        Effect: "write", Risk: "high",
+        Confirmation: "user_required", Idempotency: "unknown",
+    },
+    ConfirmFirst: true,
+    Flags: []LeafFlag{
+        {Name: "name", Usage: "应用名称", Bind: "name",
+         Trim: true, Required: true, RequiredHint: "--name 为必填"},
+    },
+    Contract: ContractDecl{
+        Description: "创建开放平台企业内部应用",
+        DryRun:    &contract.DryRunSpec{PreviewKind: "invocation"},
+        Interface: &contract.InterfaceSpec{Mode: "composite", Availability: "available", Reason: "create then configure"},
+        Selection: contract.SelectionSpec{
+            AgentSummary: "创建钉钉开放平台应用",
+            UseWhen:      []string{"需要新建企业内部应用"},
+            AvoidWhen:    []string{"应用已存在时用 update"},
+            Examples:     []string{`dws dev app create --name "Bot" --dry-run`},
+        },
+    },
+    Call: devAppCall(runner),
+})
+```
+
+### lark-cli (Shortcut)
+
+```go
+var CalendarCreate = common.Shortcut{
+    Service:     "calendar",
+    Command:     "+create",
+    Description: "Create a new calendar event",
+    Risk:        "write",
+    Scopes:      []string{"calendar:calendar"},
+    Flags: []common.Flag{
+        {Name: "summary", Desc: "Event title", Required: true},
+        {Name: "start", Desc: "Start time (RFC3339)", Required: true},
+        {Name: "end", Desc: "End time (RFC3339)", Required: true},
+        {Name: "attendees", Type: "string_slice", Desc: "Attendee emails"},
+    },
+    DryRun: func(ctx context.Context, rt *common.RuntimeContext) *common.DryRunAPI {
+        return &common.DryRunAPI{
+            Method: "POST",
+            Path:   "/open-apis/calendar/v4/calendars/{id}/events",
+            Body:   buildEventBody(rt),
+        }
+    },
+    Execute: func(ctx context.Context, rt *common.RuntimeContext) error {
+        body := buildEventBody(rt)
+        data, err := rt.CallAPITyped("POST",
+            "/open-apis/calendar/v4/calendars/{id}/events", nil, body)
+        if err != nil { return err }
+        return rt.Output(data)
+    },
+}
+```
+
+## 适用场景总结
+
+| 场景 | 最适合 | 原因 |
+|------|--------|------|
+| MCP 后端 + Agent Schema 投影 | **DWS command** | 内建 Schema 声明、SafetySpec 契约、离线 catalog |
+| REST API 直连 + OAuth scope 管理 | **lark-cli** | CallAPITyped + scope 预检 + DryRun API 计划 |
+| API-first 大规模 surface 生成 | **GWS gcloud** | Discovery 驱动，一份 Schema 生成一切 |
+| 多步编排 (跨服务链式调用) | **lark-cli** / DWS Orchestrate | lark 的 CallAPITyped 链式 + DWS 的 Orchestrate |
+| 遗留系统迁移 (保持行为等价) | **DWS command** | ConfirmFirst + 回退链 + catalog 漂移门禁 |

docs/command-index.md

@@ -40,7 +40,7 @@ Every command inherits these flags (documented here once, not repeated per comma
 - [`dws doc` — DingTalk Doc](#dws-doc) · 21 commands
 - [`dws drive` — DingTalk Drive](#dws-drive) · 6 commands
 - [`dws minutes` — AI Minutes](#dws-minutes) · 19 commands
-- [`dws oa` — OA Approval](#dws-oa) · 9 commands
+- [`dws oa` — OA Approval](#dws-oa) · 12 commands
 - [`dws report` — Reports](#dws-report) · 7 commands
 - [`dws todo` — Todo Tasks](#dws-todo) · 6 commands
 
@@ -147,8 +147,8 @@ _Group chats, conversations, messages, and robot/webhook integrations._
 | `dws chat group members remove` | Remove one or more members from a group chat. | When the agent kicks users who should no longer have access to the group. |
 | `dws chat group rename` | Update the display name of a group chat. | When the agent is rebranding or clarifying the purpose of an existing group. |
 | `dws chat list-top-conversations` | Fetch the list of conversations the current user has pinned to the top of their chat list. | When the agent needs to prioritize the user's most important conversations in a summary or dashboard. |
-| `dws chat message list` | Pull the recent message history of a specific conversation (v2), paginated. | When the agent needs to read what has recently been said in a conversation to summarize or reason about it. |
-| `dws chat message list-all` | Search all messages across the current user's conversations within a time range. | When the agent needs to audit or summarize everything the user saw across chats in a window. |
+| `dws chat message list` | Pull the recent message history of a specific conversation, including quoted-message context for merged forwards and images. | When the agent needs to read what has recently been said in a conversation and retain the context of replies. |
+| `dws chat message list-all` | Search all messages across the current user's conversations within a time range, surfacing any search-entitlement guidance. | When the agent needs to audit or summarize everything the user saw across chats in a window. |
 | `dws chat message list-by-sender` | Fetch messages authored by a specific sender across both single and group chats. | When the agent needs to pull everything a particular colleague said recently. |
 | `dws chat message list-focused` | Fetch messages from users the current user has marked as "special focus" (starred contacts). | When the agent builds a priority-inbox view highlighting messages from important people. |
 | `dws chat message list-mentions` | Fetch messages where the current user was @-mentioned. | When the agent wants to surface items that explicitly require the user's attention. |
@@ -164,16 +164,19 @@ _Group chats, conversations, messages, and robot/webhook integrations._
 
 ## `dws contact` — Contact Directory
 
-_Users, departments, and directory lookups._
+_Users, departments, directory lookups, and enterprise onboarding._
 
-**6 commands**
+**9 commands**
 
 | Command | Description | When to use |
 |---|---|---|
+| `dws contact account create` | Create a dedicated login account in the current enterprise. | When the user explicitly asks for an enterprise account or login account, rather than a new enterprise organization. |
 | `dws contact dept list-members` | List members of a specific department by department ID. | When the agent needs the roster of a department to target communication or build a team overview. |
 | `dws contact dept search` | Search departments in the organization's contact directory by keyword. | When the agent needs to resolve a department name to a department ID. |
+| `dws contact org create` | Create a new DingTalk enterprise organization. | When the user explicitly asks to create or initialize an enterprise and provides its name and creator display name. |
 | `dws contact user get` | Batch-fetch detailed profile information for one or more users by user ID. | When the agent needs names, titles, emails, or departments for a known set of user IDs. |
 | `dws contact user get-self` | Retrieve the profile of the currently authenticated user. | When the agent needs to identify who it is acting on behalf of (user ID, name, org). |
+| `dws contact user invite` | Invite one employee by mobile number into the current enterprise. | When the user explicitly asks to add an employee and has supplied the employee name and mobile number. |
 | `dws contact user search` | Search users in the contact directory by keyword (name, title, etc.). | When the agent needs to resolve a person's display name to a user ID. |
 | `dws contact user search-mobile` | Look up a user by mobile phone number. | When the agent has only a phone number and needs to find the corresponding DingTalk user. |
 
@@ -274,14 +277,17 @@ _AI meeting notes: listing, summary, todos, transcription, recording control, mi
 
 ## `dws oa` — OA Approval
 
-_OA approval workflows: list, approve, reject, revoke, records._
+_OA approval workflows: inspect forms, forecast routes, create instances, approve, reject, revoke, and audit records._
 
-**9 commands**
+**12 commands**
 
 | Command | Description | When to use |
 |---|---|---|
 | `dws oa approval approve` | Approve a pending approval process instance (task) as the current user. | When the agent acts on a pending approval the user has delegated it to handle. |
+| `dws oa approval create-instance` | Create a real approval process instance from validated form values or a complete request payload. | After the agent has inspected the form Schema, forecast the route, resolved any selectable approvers, and obtained explicit user confirmation. |
 | `dws oa approval detail` | Retrieve full details of an approval process instance, including form fields, attachments, and state. | When the agent needs to read the content of an approval ticket before deciding on it or summarizing it. |
+| `dws oa approval form-schema` | Retrieve the form Schema for an approval template by processCode. | Before collecting or validating values for a new approval instance. |
+| `dws oa approval forecast-process` | Forecast the approval route for a template and its proposed form values. | Before creating an instance, especially when the route contains user-selectable approver or notifier nodes. |
 | `dws oa approval list-forms` | List approval process templates (forms) the current user is allowed to initiate. | When the agent needs to pick the right approval form before submitting a new request. |
 | `dws oa approval list-initiated` | List approval process instances the current user has initiated. | When the agent reviews the status of approvals the user submitted. |
 | `dws oa approval list-pending` | List approval process instances currently awaiting action from the current user. | When the agent surfaces "needs your approval" items in the user's inbox. |

docs/dev-agent-test-cases.md

@@ -1,7 +1,7 @@
 # dws dev 命令集 · Agent 人肉手工评测集（10 条复合用例）
 
 > 性质：**人肉手工评测集**——由测评人逐条手工跑、肉眼核对、人工判分，不是自动化脚本。
-> 用途：评测 agent（加载 `dingtalk-dev` 技能后）能否正确处理开放平台 dev 任务。
+> 用途：评测 agent（加载 `dingtalk-misc` 的 `references/devapp.md` 后）能否正确处理开放平台 dev 任务。
 > 特点：10 条**复合用例**，每条串多个子任务，一条覆盖一类完整场景；10 条合起来覆盖全部 34 个子命令 + 8 类横切行为。
 > 约定：所有命令应带 `--format json`；写操作应先 `--dry-run` 预览、用户确认后再 `--yes`；应用定位只用 `--unified-app-id`。
 

docs/devapp-agent-install-guide.md

@@ -6,7 +6,7 @@
 
 ## 一键安装
 
-`dws dev` 能力已经合入主干并随正式版发布。专用安装脚本会下载预编译二进制 + `dingtalk-dev` skill，**只需要 curl + tar，不需要 git / go / make**。
+`dws dev` 能力已经合入主干并随正式版发布。专用安装脚本会下载预编译二进制 + `dingtalk-misc` skill（开放平台应用文档落在 misc），**只需要 curl + tar，不需要 git / go / make**。
 
 ### macOS / Linux
 
@@ -24,7 +24,7 @@ irm https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/ma
 
 1. 从 `DingTalk-Real-AI/dingtalk-workspace-cli` 的最新 Release 下载对应平台的预编译二进制。
 2. 安装 `dws` 到默认目录 `~/.local/bin`。
-3. 从 Release 的 skills 包里安装 `dingtalk-dev` skill 到本机已检测到的 Agent 目录。
+3. 从 Release 的 skills 包里安装 `dingtalk-misc` skill 到本机已检测到的 Agent 目录。
 
 支持这些环境变量（全部可选）：
 
@@ -33,7 +33,7 @@ irm https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/ma
 | `DEVAPP_REPO` | 覆盖发布仓库，默认 `DingTalk-Real-AI/dingtalk-workspace-cli` |
 | `DEVAPP_VERSION` | 钉某个 release tag，默认取最新 release |
 | `DWS_INSTALL_DIR` | 二进制安装目录，默认 `~/.local/bin` |
-| `DWS_NO_SKILLS` | 设为 `1` 跳过 `dingtalk-dev` skill 安装 |
+| `DWS_NO_SKILLS` | 设为 `1` 跳过 `dingtalk-misc` skill 安装 |
 
 > `dws dev` 已在正式版里，所以你也可以直接用标准安装脚本 `install.sh`，二者都会带上 `dws dev`。
 

docs/drafts/dws-wiki-guide.md

@@ -0,0 +1,203 @@
+# 5分钟搭好团队知识库：DWS 让 IT 服务台自己开口答疑，你值得拥有
+
+> 💡 **省流版摘要：**
+> 别再用鼠标一个个建文档了，真的没必要！本文带你用 DWS（钉钉命令行工具），5 分钟搭好一个团队知识库：建空间、搭目录、批量导入 FAQ、配权限，全程命令行一步到位。最后再花 30 秒把它挂到钉钉机器人上——同事在群里 @ 一下，知识库自己开口答疑。知识管理员从"搬运工"升级为"甩手掌柜"，就是这么简单。
+>
+> （温馨提示：本文内容可以直接丢给智能体，让智能体一次性逐步安装和使用）
+
+## 一、痛点吐槽：管知识库有多累？
+
+身为企业 IT / 知识管理员，你肯定经历过这些"九九八十一难"……
+
+- **点击马拉松**：建空间点 5 下、建文件夹点 3 下、建一篇文档再点 4 下。一个季度下来，鼠标点击次数比写的字还多。
+- **搬运工噩梦**：几百篇历史 FAQ 散落在本地 Word / Markdown 里，要搬进钉钉知识库？复制粘贴到天荒地老，格式还经常翻车。
+- **权限苦差**：新同事入职要加权限、转岗要改角色、离职要移除。逐个空间点进去操作，漏一个就是安全隐患。
+- **知识沉睡**：库是建好了，可同事还是习惯私聊问你"VPN 又连不上了怎么办"。知识库躺着吃灰，你继续当人肉客服。😭
+
+今天，DWS（DingTalk Workspace CLI）闪亮登场！🌟
+
+你不需要写一行代码，只要在终端敲几行命令，知识库的"建、搬、管、用"全链路一次搞定。更香的是：搭好的知识库可以直接挂到钉钉机器人上，让知识自己开口答疑。
+
+## 二、DWS 是个啥？知识库的"遥控指挥中心"
+
+DWS 是钉钉能力的原子化封装，把复杂的 OpenAPI 打包成简单指令。管知识库这件事，主要靠它的"三驾马车"：
+
+| 命令族 | 能干什么 |
+|---|---|
+| 🗂️ `dws wiki` | 知识库空间、目录节点、成员权限的全生命周期管理 |
+| 📄 `dws doc` | 文档内容读写、本地文件批量导入、模板套用 |
+| 📁 `dws drive` | 钉盘文件上传下载、全局搜索、归档备份 |
+
+你可以把它想象成知识库的"遥控指挥中心" 🎮——既能你手动按（终端敲命令，比点界面快 10 倍），也能让 AI 帮你按（Claude Code、Qoder 等智能体直接听懂并调用）。
+
+**适合谁用：**
+
+- **企业 IT / 知识管理员**：批量建库、批量导入、批量管权限，脚本化解放双手
+- **开发者 / AI 玩家**：把知识库挂到机器人上，打造 24 小时答疑小能手
+- **重度钉钉用户 / 效率党**：一条命令搜全库，比在界面里翻目录快得多
+
+## 三、搭好的知识库能帮你做什么？
+
+| 场景 | 玩法 |
+|---|---|
+| 🛟 IT 服务台 FAQ 库 | VPN、邮箱、打印机常见问题集中沉淀，机器人自动答疑 |
+| 📜 制度流程库 | 报销、请假、采购制度批量导入，全文秒搜 |
+| 🎓 新人上岗手册 | 按部门建目录，入职即授权限，自助通关 |
+| 🤖 知识库 + 机器人 | `--knowledge-source wiki:<spaceId>` 一挂，群里 @ 它就答 |
+
+所想即所得，拒绝画饼，直接上菜！🍽️
+
+## 四、5分钟倒计时，搭好你的团队知识库
+
+> 以下命令全部经过真实环境跑通验证，放心照抄。
+
+### 0. 安装并登录 DWS（约 1 分钟）
+
+把以下指令复制给你的智能体（Claude Code、Qoder、Codex 等）执行，或手动在终端跑：
+
+macOS / Linux：
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/main/scripts/install.sh | sh
+```
+
+Windows（PowerShell）：
+
+```powershell
+irm https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/main/scripts/install.ps1 | iex
+```
+
+登录（提示授权请扫码）：
+
+```bash
+dws auth login
+```
+
+【截图位：dws auth status 显示 token_valid: true】
+
+### 1. 建一个知识库空间（10 秒）
+
+```bash
+dws wiki space create --name "IT服务台知识库" --desc "IT 常见问题与制度流程"
+```
+
+返回里的 `workspaceId` 就是空间的身份证号，后面每步都要用它。
+
+【截图位：返回 workspaceId 与 spaceUrl】
+
+### 2. 搭目录结构（20 秒）
+
+知识库的结构 = 文件夹节点 + 文档节点。先建分类文件夹：
+
+```bash
+dws wiki node create --workspace <workspaceId> --name "常见问题FAQ" --type folder
+dws wiki node create --workspace <workspaceId> --name "制度流程" --type folder
+```
+
+在文件夹下建一篇空文档（不加 `--folder` 就建在根目录）：
+
+```bash
+dws wiki node create --workspace <workspaceId> --name "VPN连接失败排查指南" --folder <文件夹nodeId>
+```
+
+### 3. 批量导入历史文档（1 分钟，重头戏！）
+
+几百篇本地 FAQ 不用复制粘贴，`doc import` 直接整批灌进知识库，Word、Excel、Markdown、txt 通吃：
+
+```bash
+# 单篇导入到指定文件夹
+dws doc import --file ./vpn-faq.md --workspace <workspaceId> --folder <文件夹nodeId> --name "VPN连接失败排查指南"
+
+# 批量导入整个目录（bash 一把梭）
+for f in ./faq/*.md; do
+  dws doc import --file "$f" --workspace <workspaceId> --folder <文件夹nodeId>
+done
+```
+
+已有在线文档想补内容？Markdown 直接写入：
+
+```bash
+dws doc update --node <文档nodeId> --content-file ./补充内容.md --mode append
+```
+
+【截图位：终端批量导入的滚动输出 + 知识库里齐刷刷的文档列表】
+
+### 4. 配权限：把人拉进来（30 秒）
+
+```bash
+# 先用通讯录查到同事的 userId
+dws contact user search --query "张三"
+
+# 加为编辑者（--users 支持逗号分隔批量加）
+dws wiki member add --workspace <workspaceId> --users <userId1>,<userId2> --role EDITOR
+
+# 随时盘点成员
+dws wiki member list --workspace <workspaceId>
+```
+
+### 5. 验收：搜一下，秒级命中（10 秒）
+
+```bash
+# 库内全文搜索
+dws wiki node search --workspace <workspaceId> --query "VPN"
+
+# 全局搜知识库空间
+dws wiki space search --query "IT服务台"
+```
+
+【截图位：搜索结果命中文档标题】
+
+### 6. 封神一步：挂到机器人上，知识自己开口答疑（30 秒）
+
+如果你已经按《5分钟抱走你的嘴替机器人》建好了钉钉机器人，只需加一个参数：
+
+```bash
+dws dev connect --channel claudecode \
+  --robot-client-id <你的机器人ID> --robot-client-secret <你的机器人密钥> \
+  --knowledge-source wiki:<workspaceId>
+```
+
+机器人会自动从知识库拉取知识并缓存。同事在群里 @ 它问"VPN 连不上怎么办"，它直接引用你刚导入的排查指南回答——你，终于不用当复读机了。😎
+
+## 五、进阶使用技巧
+
+### 知识库管理速查表
+
+| 操作 | 命令 |
+|---|---|
+| 列出我的个人空间 | `dws wiki space list --type myWikiSpace` |
+| 列出组织知识库 | `dws wiki space list --type orgWikiSpace` |
+| 浏览库内节点树 | `dws wiki node list --workspace <ID> [--folder <nodeId>]` |
+| 移动 / 复制节点 | `dws wiki node move` / `dws wiki node copy` |
+| 改成员角色 | `dws wiki member update --users <UID> --role VIEWER` |
+| 移除成员 | `dws wiki member remove --users <UID>` |
+| 删除整个空间 | `dws wiki space delete --workspace <ID>`（进回收站，可恢复） |
+
+### 老手避坑指南 ⛳
+
+- 建在线表格用 `--type axls`，**`asheet` 服务端不支持**，别踩坑。
+- `member list` 只返回姓名和角色、**不返回 userId**；要串联 `update` / `remove`，先用 `dws contact user search --query "<姓名>"` 反查。
+- 搜索关键词的 flag 是 `--query`，`--keyword` 是遗留别名，新脚本请用 `--query`。
+- 所有命令加 `--format json`，配合 `--jq` 过滤字段，写脚本时稳得一批。
+- 破坏性操作（删空间、覆盖写文档）前加 `--dry-run` 先预览，确认无误再执行。
+
+### 让机器人答得更好
+
+| 开关 | 作用 |
+|---|---|
+| `--knowledge-source wiki:<spaceId>` | 从钉钉知识库拉知识作为答疑来源（本文主角） |
+| `--knowledge-dir <目录>` | 挂本地 .md/.txt 知识目录，可与上面并存 |
+| `--allowed-groups / --allowed-users` | 白名单，只让指定群或人触发 |
+| `--daemon` | 后台常驻，关掉终端也不断线 |
+
+## 六、更多 DWS 官方信息
+
+- 钉钉 CLI 官网：https://open.dingtalk.com/dingtalk-cli
+- 开源仓库：https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli
+- 上一篇姊妹篇：《5分钟抱走你的嘴替机器人：启动钉钉DWS，你值得拥有》
+
+## 七、欢迎加入交流群
+
+【截图位：DWS 交流群二维码】
+
+遇到问题来群里喊一声，官方同学在线答疑。下一篇想看什么？批量备份知识库？给知识库做权限审计？留言区点菜！🍻

docs/event-subprocess-contract.md

@@ -0,0 +1,219 @@
+# Event consume — AI subprocess contract
+
+Defines the stable `dws event consume` subprocess contract so an
+orchestrator can determine when the consumer is ready, stop it cleanly,
+and machine-read why it exited.
+
+Scope of this branch: the six **contract** items below. Reconnect
+resilience (keeping the stream alive across a transient upstream drop) is
+tracked separately and intentionally out of scope here.
+
+## Baseline (already present, no work)
+
+- `--max-events N` — stop after N events (exit 0).
+- `--duration D` — wall-clock budget (exit 0). Kept as `--duration`, NOT
+  aliased to `--timeout`: the global `--timeout` is the HTTP request
+  timeout (int seconds) and would collide (different type and meaning).
+- Bus idle-shutdown fires only with **zero** consumers, so a connected
+  consumer is never idle-killed.
+- SIGINT/SIGTERM already cancel the run context and return cleanly.
+
+## Improvements
+
+### 1. Ready marker (standardized)
+
+On connect, emit a fixed stderr line **before** any stdout event:
+
+```
+[event] ready event_key=<key> bus_pid=<pid> subscribe_id=<id>
+```
+
+Parents block on stderr until this line, then read stdout. Suppressed
+under `--quiet`. Replaces the ad-hoc `connected bus pid=...` line (which
+omits `event_key`).
+
+**Verification**
+- T1a: stderr contains a line matching `^\[event\] ready event_key=<key>`.
+- T1b: that line appears before the first stdout event (ordering).
+- T1c: with `--quiet`, the line is absent.
+
+### 2. stdin EOF = graceful exit
+
+`consume` watches stdin; closing stdin is a shutdown signal (wired for AI
+subprocess callers). To stay resident, feed a never-EOF stdin
+(`< <(tail -f /dev/null)`) or run bounded (`--max-events` / `--duration`).
+
+**Verification**
+- T2a: `printf '' | dws event consume <key>` exits ≤2s, code 0, final
+  line `reason: signal` (stdin-eof classified as signal).
+- T2b: `dws event consume <key> < <(tail -f /dev/null)` still alive after
+  5s, connection intact.
+- T2c (unit): a controllable stdin reader hitting EOF makes Run return nil
+  via the cleanup path.
+
+### 3. Exit reason contract + exit codes
+
+On exit, final stderr line:
+
+```
+[event] exited — received N event(s) in Xs (reason: <limit|timeout|signal|bus_shutdown>)
+```
+
+Exit codes: controlled exit (limit/timeout/signal/stdin-eof) = 0; startup
+or runtime failure (permissions, network, params) = non-zero, with no
+`exited` line and an `Error:` line instead.
+
+**Verification**
+- T3a: `--max-events 1` + 1 event → exit 0, reason=`limit`, N=1.
+- T3b: `--duration 2s`, no events → exit 0, reason=`timeout`.
+- T3c: SIGTERM mid-run → exit 0, reason=`signal`.
+- T3d: bad params / permission failure → exit≠0, no `exited` line, has `Error:`.
+- Unit tests assert (reason string, exit code) for each path.
+
+### 4. Cleanup on exit (no `kill -9`)
+
+Ownership-based cleanup:
+- If this run **created** the subscription (no `--subscribe-id`), a clean
+  exit (SIGTERM / SIGINT / stdin-EOF / limit / timeout) **unsubscribes**
+  it server-side and sends Bye.
+- If `--subscribe-id` was passed (reusing an existing subscription), the
+  subscription is **left intact** — the caller owns its lifecycle.
+- `--ephemeral` remains as an explicit "always unsubscribe" override.
+- Help/docs warn: avoid `kill -9` (skips the unsubscribe → leaked
+  server-side subscription: "subscription already exists" on restart,
+  duplicate delivery). Prefer SIGTERM or closing stdin.
+
+**Verification**
+- T4a: start consume (self-created subscription), record subscribe_id;
+  SIGTERM; afterwards `dws event status` no longer lists that subscribe_id
+  and the server-side subscription is gone.
+- T4b: start consume with `--subscribe-id <existing>`; SIGTERM; the
+  subscription is still present (reuse case preserved).
+- T4c (control): `kill -9` leaves subscribe_id lingering (documented risk;
+  we only guarantee SIGTERM is clean, we do not fix kill -9 itself).
+
+### 5. Subscription-create retry orchestration and local guard
+
+This policy covers all 16 public personal-event keys and every logical
+subscription in a multi-event command. It applies only before the ready
+marker; reconnecting an established Stream remains a separate mechanism.
+
+- The `0/2/1` limits below are an **Agent/host orchestration contract**, not
+  a CLI-enforced persisted total-attempt cap. Each `dws event consume`
+  process sends at most one subscription-create HTTP request for a logical
+  subscription and performs no in-process automatic retry. The CLI persists
+  only the `in_flight`, `cooldown`, and `terminal_hold` guard states; it does
+  not persist or enforce the Agent/host attempt count across invocations.
+- ID resolution, `event consume`, and later `event status/stop` must use the
+  same `--profile`. A user or conversation ID resolved under another profile
+  must not be reused for the current subscription.
+- A logical subscription is keyed by the current profile/identity, event key,
+  rule type, target, and filters. A new `subscribe_id`, `trace_id`, or process
+  does not create a new logical operation or reset the Agent/host budget.
+- For the Agent/host, `retryable=false` means
+  `max_additional_attempts=0`.
+- For the Agent/host, `retryable=true` means
+  `max_additional_attempts=2`. It must honor `retry_after_seconds` or
+  `next_retry_at` when present and must not retry early.
+- For the Agent/host, an omitted retryable value
+  (`retryable=unknown`) means `max_additional_attempts=1`; a second unknown
+  failure stops the operation.
+- `in_flight` means the original logical request is still running.
+  `cooldown` and `terminal_hold` mean a guard is already delaying or blocking
+  it. These states must not recursively launch `event consume`, start a
+  parallel equivalent subscription, or bypass the guard with a new subId or
+  trace. The caller waits for the original request/guard or stops, while the
+  Agent/host keeps its own orchestration count.
+- A multi-event command remains one original operation. A caller must not
+  split out a failed event, reorder events, or restart the command to bypass
+  a budget. Existing startup rollback cleans subscriptions created before a
+  later item fails.
+
+#### Local guard state operations
+
+- The default open-edition state file is
+  `~/.dws/events/open/personal_stream/<identity_hash>/personal_subscription_attempts.json`.
+  The config root follows `DWS_CONFIG_DIR` when set, and another edition uses
+  that edition's directory instead of `open`.
+- The identity directory is mode `0700`; both
+  `personal_subscription_attempts.json` and
+  `personal_subscription_attempts.lock` are mode `0600`.
+- A failure streak resets after 24h without another failure. A
+  `terminal_hold` lasts 1h. Prefer waiting until the reported
+  `next_retry_at`; do not clear the file as a normal retry mechanism.
+- For emergency recovery, first ensure that no subscription-create process is
+  running for that identity. Delete only
+  `personal_subscription_attempts.json`, never the lock file. This clears
+  every protection record for that identity, not just one event.
+
+**Verification**
+- T5a (policy): skill/docs tests pin the Agent/host 0/2/1 orchestration
+  contract and explicitly reject describing it as a CLI-persisted hard cap.
+- T5b (CLI): one process issues at most one create request per logical
+  subscription; a changed subId/trace or process restart does not bypass the
+  persisted fingerprint guard.
+- T5c: `in_flight`/`cooldown` does not recursively issue another create.
+- T5d: multi-event startup cannot be split or reordered to bypass the guard,
+  and a partial startup still rolls back earlier subscriptions.
+- T5e: state-store tests cover `0700`/`0600` permissions, 24h reset, 1h
+  `terminal_hold`, and identity-scoped cleanup; skill/docs tests pin the
+  operational recovery instructions.
+
+### 6. Host runtime-token handoff
+
+When the root command carries an explicit host-supplied `--token`, personal
+event control requests and the foreground Stream use that token with higher
+priority than local OAuth. A detached bus receives it only through the
+owner-only local IPC transport:
+
+1. The child starts in runtime-token mode with non-sensitive identity and
+   ticket metadata only; neither its argv nor environment contains the token.
+2. The consumer sends `Hello` with `credential_mode=runtime_token`.
+3. The bus advertises the additive `runtime_token_v1` capability and its
+   in-memory credential generation in `HelloAck`.
+4. Only after that capability is confirmed does the consumer send a bounded
+   `credential_update` frame. The bus applies it with generation CAS, replies
+   with `credential_update_ack`, and registers the consumer only on success.
+
+The bus blocks ticket acquisition until the first runtime credential arrives.
+A later invocation may rotate Token A to Token B on a compatible existing bus;
+the current WebSocket remains connected and the next ticket request or natural
+reconnect uses B. If a 401 rejects the current runtime token, only an already
+installed newer generation is retried; the runtime path never refreshes or
+falls back to a local OAuth profile and never suggests `dws auth login`.
+
+Clients do not send a token to a bus that lacks the capability, do not stop
+other consumers automatically, and fail before printing the ready marker. With
+no explicit `--token`, the original OAuth, refresh, profile, and old-client to
+new-bus protocol behavior remains unchanged.
+
+**Verification**
+- T6a: a stale local Token A and root Token B produce control and ticket
+  requests authenticated only with B.
+- T6b: compatible bus reuse supports A-to-B rotation and generation conflicts;
+  401 retries only an already-installed newer runtime token.
+- T6c: an old bus receives no credential and remains running; the new consumer
+  exits before its ready marker.
+- T6d: a canary credential is absent from child argv/environment, dry-run,
+  stdout/stderr, `bus.meta`, `bus.log`, run state, and returned errors.
+- T6e: no-token OAuth, refresh, multi-profile, marker/cache, and bus-reuse tests
+  continue to pass.
+
+## Out of scope (next branch)
+
+**Reconnect resilience** — today `personal source` retries only
+`retryable` errors (1–30s backoff); a non-retryable error tears the bus
+down and takes consume with it (the likely cause of the observed silent
+drop). Making more drops retryable, keeping the bus alive across a
+reconnect, and emitting `reason: source_lost` only after exhausting the
+budget — tracked on its own branch, since it needs error-classification
+judgement and real flaky-network testing, and would otherwise couple clean
+contract work with resilience work.
+
+## Test surface
+
+- Unit: extend `internal/event/consume/*_test.go` with fake bus conn /
+  stdin / stderr sink for T1c, T2c, T3 (all paths), T4 ownership branch.
+- Integration/e2e: `--foreground` + mock source (or a short real run) for
+  T1a/b, T2a/b, T3a–d, T4a/b/c — assert the stderr contract lines and exit
+  codes.

docs/flag-help-schema-homology.md

@@ -0,0 +1,269 @@
+# flag / help / schema 同源
+
+- **状态**：已决策（路径 A + Contract 嵌入 Schema）
+- **相关**：[`rfc-command-framework-convergence.md`](rfc-command-framework-convergence.md)、[`schema-dynamic-endpoint-design.md`](schema-dynamic-endpoint-design.md)
+- **实现门面**：`LeafSpec` → `corecmd.Spec` → `corecmd.New`
+
+## 1. 决策
+
+采用 **路径 A：Contract / LeafSpec 为 CLI 表面权威**，并且 **Contract 必须嵌入进 Schema**。
+
+「同源」的含义是：同一份 Contract（今日经 LeafSpec / `corecmd.Spec` 表达）同时决定——且门禁能证明——
+
+1. cobra 实际注册的 flags / required / defaults；
+2. `--help` 的 Flags 与「参数约束」段；
+3. 运行时**组装后的** Schema 交付中的 parameters、关系约束和 SafetySpec：
+   - **SchemaRegistry / Catalog ToolSpec wire** 供 `dws schema` / `--all` / 完整 leaf 载荷（lazy；首次 `dws schema` 触发完整装配）。
+   - **`ResolveMeta` / `SafetyForCLIPath` / leaf `--help` Safety** 与装配同源：`deliverySchemaCatalog` sync.Once 装配后同步物化 `map[cli_path]CommandMeta`；稳态为 O(1) map lookup，不为 Meta 单独重做全量 ToolSpec/wire 投影。
+   Agent metadata 在组装期间经内存 inject，不落盘、不 embed；`schema_agent_metadata/` 已退役，若存在则 policy 失败。
+
+嵌入机制（已落地）：
+
+```text
+corecmd.Spec
+  → RegisterFlags + embedContractIntoSchema
+  → cobra annotations:
+       dws.schema.contract=command
+       dws.schema.property / type / required  (per flag)
+       dws.schema.constraints
+  → RegisterRuntimeContractFinal(SafetySpec + ContractDecl)
+  → Schema 组装透传 Contract Final
+    （组装时内存 inject Agent metadata）
+  → RegisterSchemaSourceRoot → ResolveSchemaBuild
+  → SchemaRegistry (+ Meta cache map) / dws schema wire projection
+```
+
+command/Leaf 不再写 `dws.schema.risk`；SafetySpec 走类型化 Final 载荷，不使用字符串枚举注解。
+
+### 1.1 硬规则：声明 = 最终数据源（Schema 透传）
+
+受管命令进入 Schema 的叶子数据由 **Contract 声明**定义最终值；框架（`corecmd.New`）做**类型转换**并注册，Schema 组装**透传**，不得：
+
+- 把声明序列化成 JSON 注解再解析；
+- 在声明体系里再挂「评审字段」并行权威；
+- 用 hints/registry 盖写已声明字段。
+
+**declare-vs-delivery 例外（Title / Description）**：构造期 `ContractDecl.Description` **必填**（声明证据），但这不是「declare = wire 最终值」。Catalog **交付**时：
+
+- **description**：有 Cobra Long → 交付 Long（provenance `cobra_help` / `cobra_help_preferred`）；无 Long → 交付声明（`contract_final`）。**Short 不进入 description**。
+- **title**：声明 `ContractDecl.Title` 优先，否则 Cobra Short，再 MCP；组装 stamp 真实 winner。
+
+这是一条权威链上的显式交付偏好，不是双权威（见 RFC §5.0.4）。
+
+迁移期未迁完的叶子可暂走旧组装路径；**新声明面不含 review_reason / reviewed 字段**。写命令未设 `Safety` 时，过渡期仍可用 `runtime_gate` annotate（`HOM-S2`）。
+
+**不采用**路径 B（以 MCP meta / 已退役的 `schema_mcp_metadata` 生成全部 CLI flag/help/schema）作为主权威。钉钉 MCP meta 不是飞书 OAPI：粒度与 CLI 特有语义（二选一、OmitEmpty、ConstParams、write guard）无法从裸 meta 推出；强行生成会违反「Schema 描述 CLI，不制造 CLI」。
+
+路径 B 仅允许作为 **可选的 1:1 MCP 透传叶子通道**（见 §5），不得覆盖 LeafSpec / Shortcut 主路径。
+
+对外叙事：与飞书 **分层单权威** 同构——API/透传叶可用平台事实，产品 CLI / Shortcut 用手写契约 + 执行体——而不是「全家只有平台 meta」。
+
+### 1.2 声明（declare）：写什么、写在哪、投影到哪
+
+**定义**：声明 = 在 Contract 结构体的**数据字段**上写出事实；`corecmd.New` 据此注册 cobra、渲染 help、写入 `dws.schema.*`。钩子闭包（`Validate` / `Call` / `PostMount` / `RunE`）里的逻辑**不算**声明——即使行为正确，也不能单靠钩子让 Schema/help「猜出」该事实。
+
+命令框架边界与今日/目标对应见 RFC [`rfc-command-framework-convergence.md`](rfc-command-framework-convergence.md) **§5.0**（框架上的「声明」定义）；本节给字段级表与示例。
+
+**唯一写入面**（三选一，语义相同）：
+
+| 入口 | 类型 | 归一 |
+|---|---|---|
+| Leaf 命令 | `helpers.LeafSpec` | `FromLeafSpec` → `corecmd.Spec` |
+| Shortcut | Shortcut 声明（经 `FromShortcut`） | → `corecmd.Spec` |
+| 直接基座 | `corecmd.Spec` | `corecmd.New` |
+
+今日产品 CLI 叶子以 **`LeafSpec` 为声明门面**；字段与 `corecmd.Spec` 契约面一一对应。
+
+#### 1.2.1 契约字段（算声明）
+
+| 字段 | 声明什么 | 运行时 | 嵌入 Schema / help |
+|---|---|---|---|
+| `Flags[]`（`FlagSpec` / `LeafFlag`） | 用户可见参数面：名、类型、默认、必填、usage | 注册 cobra flag；装配 toolArgs | `dws.schema.property` / `type` / `required`；`--help` Flags |
+| `Constraints[]` | 跨 flag 关系：`at_least_one` / `exactly_one` / `mutually_exclusive`；`custom` 记录钩子校验 | 通用关系由 `ValidateConstraints` 执行；`custom` 由 `Validate` 执行 | `dws.schema.constraints`；`--help`「参数约束」 |
+| `Safety`（`contract.SafetySpec`） | effect/risk/confirmation/idempotency 四个独立事实 | `confirmation=user_required` 时 `ConfirmSafety`；`--yes` / `--dry-run` 跳过 | 同一个 SafetySpec 原样进入 Contract Final（`HOM-S1`） |
+| `ConstParams` | 固定载荷（不上 flag 表） | 并入 toolArgs；不满足 Required | **不**投影为用户 parameter |
+| `Use` / `Short` / `Long` / `Example` | 命令身份文案与示例 | cobra 自身 | help；identity 以 collector 收集的 `ContractFinal.Identity` 声明为准（reviewed registry 已退役） |
+
+`FlagSpec` 子字段（声明细节）：
+
+| 子字段 | 作用 | 是否进 Schema parameters |
+|---|---|---|
+| `Name`, `Usage`, `Kind`, `Default` | 注册名/说明/类型/DefValue | 是（name/type；default 与 cobra 对齐） |
+| `Required` / `MarkRequired` | 非空校验 / cobra 硬必填 | 是（`required`） |
+| `RequiredHint`, `Aliases`, `EnvVar` | 校验提示、隐藏别名、环境回退 | 否（执行细节；别名不上主 parameter 表） |
+| `ArgDefault`, `Bind`, `OmitEmpty`, `Trim`, `Transform` | toolArgs 装配语义 | 否（载荷细节；`Bind` 可进 property 映射，但不另造 flag） |
+
+#### 1.2.2 编排 / 执行字段（不算声明）
+
+| 字段 | 角色 | 禁止用来「冒充」的声明 |
+|---|---|---|
+| `Validate` | 条件式/领域校验钩子 | 不得在此 `Flags().String(...)` 注册业务 flag；不得只靠钩子表达「必填/互斥」而不写 `Flags`/`Constraints` |
+| `Call` / `Invoke` / `Orchestrate` | 执行体 | 不得 `params[k]=…` 装配业务参数（应在 `Flags`/`ConstParams`） |
+| `PostMount` | 挂载后收尾（annotate、领域工具注入） | 不得注册业务 flag；分页等横切由领域工具注入并可走 annotate |
+| `RunE` | 逃生舱（整段手写） | 表面事实仍须 Flags 声明；框架仍按 Safety 执行确认 |
+| `Server` / `Tool` | MCP 路由 | 不构成 CLI parameter 声明 |
+
+#### 1.2.3 最小声明示例
+
+```go
+// 读：Safety 四字段显式对齐 Schema
+NewLeafCommand(LeafSpec{
+    Use: "get", Short: "…", Tool: "…",
+    Safety: contract.SafetySpec{
+        Effect: "read", Risk: "low",
+        Confirmation: "not_required", Idempotency: "idempotent",
+    },
+    Flags: []LeafFlag{
+        {Name: "unified-app-id", Usage: "…", Bind: "unifiedAppId", Trim: true, Required: true},
+    },
+    Call: devAppCall(runner), PostMount: devAppMeta(tool),
+})
+
+// 写：同一个 SafetySpec 同时驱动确认与 Schema
+NewLeafCommand(LeafSpec{
+    Use: "publish", Short: "…", Tool: "…",
+    Flags: []LeafFlag{ /* … */ },
+    Safety: contract.SafetySpec{
+        Effect: "write", Risk: "medium",
+        Confirmation: "user_required", Idempotency: "unknown",
+    },
+    Call: devAppCall(runner), PostMount: devAppMeta(tool),
+})
+
+// 迁移期旧写命令：确认走 annotate，见 §1.3 —— 不是新声明面
+NewLeafCommand(LeafSpec{
+    Use: "create", /* Flags… */,
+    Validate: func(cmd *cobra.Command, _ []string) error {
+        return devAppRequireWriteGuard(cmd, "create") // 执行守卫，不是 Contract 声明
+    },
+    Call: devAppCall(runner),
+    PostMount: devAppMetaWrite(tool), // 人工标注 runtime_gate
+})
+```
+
+**空 `Safety` 的含义**：command 为兼容旧只读叶保留 `read/low/not_required/idempotent` 默认。因此「会改状态却留空 Safety」**不是**合法声明；新 Leaf 必须写完整 SafetySpec，未迁移旧路径则按 §1.3 标注 gate。
+
+### 1.3 人工标注（annotate）：声明的补充通道
+
+当事实无法或不愿放进 Contract 字段时，必须**显式**落注解 / 评审源，禁止组装期推断：
+
+| 标注手段 | 典型值 | 何时用 |
+|---|---|---|
+| `cli.AnnotateRuntimeGate` / `devAppMetaWrite` | `dws.schema.runtime_gate=devAppRequireWriteGuard` | 尚未迁移到 SafetySpec 的旧写命令（`HOM-S2`） |
+| `cli.AnnotateRuntimeRisk` | `dws.schema.risk=…` | Shortcut 暂存的旧兼容路径；command/Leaf 禁止新增 |
+| `cli.AnnotateRuntimeFlag` / Constraints | 与 embed 同形 | 手写 cobra 叶补齐表面（长期应迁入 Contract） |
+| reviewed `schema_hints/metadata` Safety（已退役） | effect/risk/confirmation | 已删：`schema_hints/` 不得重现；受管命令以 Contract.Safety / gate 为准 |
+
+标注与声明冲突时：**Contract 声明胜**（路径 A）。标注不得发明未注册的 CLI flag。
+
+### 1.4 Schema 全覆盖（`ToolSpec` 无空洞）
+
+`dws schema` 叶子模型是 `cli.ToolSpec`。命令框架必须为**每一个字段组**指定权威；完整矩阵在 RFC **§5.0.4**，摘要：
+
+| ToolSpec 组 | 权威类 | 框架声明字段 / 其它源 |
+|---|---|---|
+| Identity | 声明源（identity collector 收集 `ContractFinal.Identity`；reviewed `schema_command_registry` 已退役） | `ContractDecl.Identity` 声明 |
+| Display / Title / Description | 声明证据 + 交付偏好（非双权威） | **title**：ContractDecl 优先，否则 Cobra Short，再 MCP；**description**：构造期 Description 必填；Catalog 交付 Long→`cobra_help`，无 Long→`contract_final`；**Short 不进 description**（RFC §5.0.4） |
+| Parameters.`name/type/required/default/property` | **声明**（或同形 annotate） | `Flags` / `Bind` |
+| Parameters.`description` | 声明 usage（`FlagSpec.Usage` / ParamDecl） | `schema_hints/` 已退役；不得用 overlay 改 type/required/default |
+| Parameters.`interface_*` | 评审源 | MCP meta / bindings；**不造 flag** |
+| Constraints | **声明** | `Constraints` |
+| Positionals | **声明** 或显式 annotate | 目标 `Args`；禁止推断 |
+| Safety.`effect/risk/confirmation/idempotency` | **声明**完整 `Safety`，或迁移期 `runtime_gate` / reviewed Safety | 四字段独立；不得互相推导 |
+| DryRun | 评审源 | dry-run capabilities registry |
+| Interface | 评审源 | MCP + 内存 inject 的 Agent metadata |
+| Selection | 声明（ContractFinal / ProductDecl） | `ContractDecl.Selection` / `ProductDecl` |
+| FieldProvenance / Extensions | 组装派生或评审扩展 | 组装器；与 delivered value 一致 |
+| （非 Schema parameter）ConstParams | **声明** | 载荷；不上 parameters 表 |
+
+验收：新增 Schema 字段必须同步改 RFC §5.0.4 + 本表；受管写命令 Safety 不得无主。
+
+## 2. 字段归属（每一类恰好一个写入者）
+
+| 字段类 | 权威 | 投影到 |
+|---|---|---|
+| flags / defaults / required / enum / 关系约束 / 运行时 Risk | Contract（LeafSpec / `corecmd.Spec` 门面） | cobra、`--help`、Schema `parameters` / constraints / confirmation |
+| ConstParams、Bind、OmitEmpty、Transform | 同上（载荷声明，不上 flag 表） | toolArgs；Schema 不把 ConstParams 伪装成用户 flag |
+| canonical path / aliases / navigation / exposure | identity collector 收集的 `ContractFinal.Identity` 声明（reviewed `schema_command_registry` 已退役） | Schema identity |
+| use_when / avoid_when / examples / agent_summary 文案 | `ContractDecl.Selection` / `ProductDecl` | Schema selection |
+| RPC tool 形状、`interface_ref`、interface 描述 | leaf `Contract.Interface` + `ParamDecl`（`schema_parameter_mapping_ledger.go` 仅 mapping_exclusions / removals；`schema_mcp_metadata` 已退役） | Schema `interface_*` 字段；**不得创建 flag** |
+| 参数描述 overlay（可选） | 生产 metadata 壳为空；参数事实走 ParamDecl / FlagSpec | **Contract/cobra 胜** |
+| 遗留 Safety 文案（迁移期） | 生产 metadata 壳为空；Safety 走 Contract | 以 Contract.Safety / runtime_gate 为准（见 §4） |
+| dry-run 正能力 | reviewed dry-run registry | Schema `dry_run` |
+| positionals | Contract Args / 显式 annotate | Schema `positionals` |
+| FieldProvenance | 组装派生 | Schema provenance（与值一致） |
+
+Selection **刻意不**由单命令 Contract 取代（RFC 决策 8 / schema 设计硬规则）；identity 的历史决策是不进 Contract、归 reviewed `CommandRegistry`，该 registry 已退役，现由 identity collector 收集 `ContractFinal.Identity` 声明（声明即 identity）。**完整无空洞表见 RFC §5.0.4。**
+
+## 3. 当前缺口与目标闭环
+
+已具备：
+
+- Flags / ConstParams / Constraints → 注册、校验与 `ConstraintHelp`；SafetySpec → 运行时 `ConfirmSafety`（command）；
+- Call / Execute 作为执行体；业务参数不得在 Call 内装配（helpers 门禁）；
+- **Contract → Schema 嵌入**：参数/约束写原生 annotation，SafetySpec 与 ContractDecl 注册为类型化 Contract Final 并由 Schema 组装透传；
+- Selection 权威为 `ContractDecl.Selection` / `ProductDecl`（`contract_final`）；`schema_hints/` 已退役。
+
+已进 CI（`make policy` → `check-schema-catalog.sh` / `check-runtime-confirmation-truth.sh`）：
+
+| Gate ID | CI 入口（`-run` 白名单 / 脚本） |
+|---|---|
+| `HOM-P1` / `HOM-D1` | `./internal/app`：`TestFinalSchemaParametersMatchExecutableHelpFlags`、`TestDeliverySchemaParametersMatchExecutableHelpFlags` |
+| `HOM-P2`（参数映射/bindings 子集） | `./internal/cli`：`TestSchemaParameterBindingsMatchReviewedBaselineAndDeliveryCatalog`、`TestDeliveryCatalogMCPParameterMappingsAreComplete` 等 bindings 门禁 |
+| `HOM-S1` / `HOM-S2`（confirmation 同源） | `./internal/cli/homology`：`TestUserRequiredSafetyHomologyWithRuntimeGate` + `check-runtime-confirmation-truth.sh`；`./internal/app`：`TestSheetFinalSchemaConfirmationMatchesRuntimeGuards` |
+| 词汇/决策钉扎 | `./internal/cli/homology`：`TestHomologyDecisionDocPinsPathAAndGateIDs`、`TestMCPPassthroughAdmissionExcludesLeafAndShortcut`、`TestHomologyCIEntrypointsPinned` |
+
+仍缺（未宣称全量 CI 覆盖）：
+
+1. 独立可执行的 `HOM-P3`（constraints ≡ AnnotateConstraints）与 `HOM-S3`（read 不得误投影 user_required）全量 gate；
+2. `HOM-I1` 作为单独 gate ID 的显式用例（MCP bindings ⊆ Contract flags 已有映射审计子集，但未钉 `HOM-I1` 标签）。
+
+已落地（写命令确认语义，`HOM-S2`）：
+
+- `AnnotateRuntimeGate` / `dws.schema.runtime_gate`；Leaf `PostMount: devAppMetaWrite`；手写 delete/robot 等同路径显式标注；
+- Schema 组装在无 Contract Safety 但有 gate 时 overlay `confirmation=user_required`（`applyContractGateToSafety`）；
+- AST/mount 测试：新 Leaf 须声明完整 SafetySpec；尚未迁移的旧路径须有 runtime_gate。
+
+## 4. Schema 投影与 Safety 门禁规划
+
+以下门禁 ID 稳定，便于 CI 认领。§3 表标明哪些已挂入 `check-schema-catalog.sh`。
+
+| Gate ID | 断言 | 范围 | CI |
+|---|---|---|---|
+| `HOM-P1` | 受管 leaf 的 schema `parameters[].name` 集合 ≡ cobra 本地 flag 名集合（排除全局 persistent） | LeafSpec / Contract 编译命令 | **已进**（app help↔schema） |
+| `HOM-P2` | schema parameter `type` / `required` / `default` 与 cobra DefValue / MarkFlagRequired / FlagSpec 一致；不得用已退役 hints overlay 改写这三项 | 同上 | **部分**（bindings/mapping 门禁） |
+| `HOM-P3` | schema 关系约束（require_one_of / mutually_exclusive）≡ Contract/Leaf `Constraints` 投影（与 `AnnotateConstraints` 同构） | 声明了 Constraints 的命令 | 规划 |
+| `HOM-S1` | Contract/Leaf `user_required` Safety 与运行时 Confirm/gate 同源，且 help Safety 行同语义 | 受管写/破坏性命令 | **已进** |
+| `HOM-S2` | 若命令走显式 write guard（如 `devAppRequireWriteGuard`）而非完整 SafetySpec，则必须人工标注 `dws.schema.runtime_gate`；Schema 不得呈 `confirmation=not_required`；符合 §1.1 declare OR annotate | 今日 devapp 写命令 | **已进**（同源测试含 gate 路径） |
+| `HOM-S3` | `Risk=read`（或空→read）不得投影为 `user_required`，除非有 reviewed exclusion reason | 受管读命令 | 规划 |
+| `HOM-I1` | `interface_ref` 存在时，bindings 覆盖的 CLI flag ⊆ Contract flags；MCP meta **不**引入额外 CLI flag | 有 MCP 绑定的命令 | **部分**（mapping 审计） |
+| `HOM-D1` | `dws <path> --help` Flags 段与 schema leaf parameters 零未解释增量 | 受管公开 leaf | **已进**（与 HOM-P1 同测） |
+
+落地顺序建议：
+
+1. 保持 `HOM-P1`/`HOM-D1`/`HOM-S1`/`HOM-S2` 在 `check-schema-catalog.sh` 白名单中（勿再只靠词汇钉扎）；
+2. 补独立 `HOM-P3` / `HOM-S3` 可执行 gate，并把 `HOM-P2`/`HOM-I1` 从「部分」升到全量标签；
+3. 新 Leaf 继续走 Contract 嵌入；禁止 `schema_hints/` 回潮。
+
+## 5. 路径 B 子通道：1:1 MCP 透传叶（可选，非主路径）
+
+仅当同时满足以下条件时，才允许「从 MCP meta 生成 flag/help/schema 参数」：
+
+1. CLI path ↔ 单一 MCP tool **严格 1:1**，无多步、无按名解析、无本地 effect；
+2. 无不在 MCP input schema 中的 CLI 特有 flag（含 guard 专用语义 flag 除外的全局 `--yes`/`--dry-run`）；
+3. 无 ConstParams / Transform / 跨 flag 约束 / Call 内业务逻辑；
+4. Risk/confirmation 在 meta 或并列 reviewed Safety 中有显式来源，不靠生成器猜测；
+5. 在 identity 声明面标记 `surface_kind=mcp_passthrough`（名称可调整；原计划标在 reviewed registry，该 registry 已退役），且 **不得**与 LeafSpec/Shortcut 手写定义双注册同一 `cli_path`；
+6. 文档与门禁写明：该通道是子集优化，失败时回退/禁止扩张到产品 CLI。
+
+显式排除（永远走路径 A / Shortcut）：
+
+- 全部 `LeafSpec` 命令（含 `dws dev app …`）；
+- 全部 `+shortcut` 与 smart 编排；
+- 任何需要 `devAppRequireWriteGuard`、cursor 工具注入、或响应投影的命令。
+
+## 6. 非目标
+
+- ~~不把 canonical identity / 导航塞进 Contract（仍归 reviewed `CommandRegistry`）~~ —— 该非目标已被后续演进取代：reviewed `CommandRegistry` 已退役，identity 现由 collector 收集 `ContractFinal.Identity` 声明（声明即 identity，不再是独立评审文件）。selection 文案已由 `ContractDecl.Selection` / `ProductDecl` 声明（非 hints）。
+- 不要求删除 LeafSpec 门面。
+- 不把「生成 catalog 字节一致」当作运行时同源的充分条件（仍需 `HOM-*` 与差分门禁）。

docs/pagination-schema-compatibility-plan.md

@@ -0,0 +1,134 @@
+# 独立 `meta.pagination` Schema 方案
+
+## 1. 目标结构
+
+业务结果与分页控制信息分层：
+
+```json
+{
+  "ok": true,
+  "outcome": "success",
+  "data": {
+    "items": [{"id": "a"}]
+  },
+  "meta": {
+    "pagination": {
+      "endpoint_exhausted": false,
+      "next_token": "cursor-2"
+    }
+  }
+}
+```
+
+对应 compact/full leaf Schema：
+
+```json
+{
+  "result": {
+    "outcomes": ["success", "failure"],
+    "data_schema": {
+      "type": "object",
+      "properties": {
+        "items": {
+          "type": "array",
+          "description": "当前页业务记录",
+          "items": {"type": "object"}
+        }
+      }
+    }
+  },
+  "pagination": {
+    "kind": "cursor",
+    "cursor_parameter": "cursor",
+    "meta_path": "meta.pagination",
+    "endpoint_exhausted_path": "meta.pagination.endpoint_exhausted",
+    "next_token_path": "meta.pagination.next_token"
+  }
+}
+```
+
+`result` 只描述 `data`；`pagination` 是与 `result` 同级的命令能力声明。
+
+## 2. 分页状态
+
+| 状态 | `endpoint_exhausted` | `next_token` | Agent 行为 |
+|---|---:|---|---|
+| 可续跑 | `false` | 必须非空 | 将 token 传给 `--<cursor_parameter>` |
+| 已耗尽 | `true` | 必须省略 | 停止翻页 |
+
+`endpoint_exhausted:true` 只表示观察到 Endpoint 分页耗尽，不表示搜索索引
+健康、数据全量覆盖或业务对象不存在。
+
+## 3. 映射规则
+
+产品 mapper 可以读取服务端原始 `hasMore/nextCursor`、`has_more/page_token`
+等字段，但统一 CLI 输出只公布 `meta.pagination`：
+
+- 服务端表示还有下一页且 cursor 非空 → `endpoint_exhausted:false` + token。
+- 服务端表示没有下一页 → `endpoint_exhausted:true`，不带 token。
+- 表示还有下一页但 cursor 缺失、类型错误或证据冲突 → typed
+  `pagination_inconsistent`，禁止伪装终页。
+- mapper 使用同一份上游响应构造 `data` 与 `meta`，不得重新请求。
+
+原始分页控制字段不进入新的 `result.data_schema`。未迁移命令保持 legacy；
+已迁移命令按命令独立切换和回滚，不通过 Agent 参数选择协议。
+
+## 4. Schema 规则
+
+- `kind` 当前只允许 `cursor`。
+- `cursor_parameter` 是真实 canonical CLI flag 名，不带 `--`，并必须存在于
+  同一 leaf 的 `parameters`。
+- 三个 meta path 由框架固定生成，产品不能覆盖。
+- compact/full leaf 同时包含相同的 `result` 和 `pagination`。
+- product/group 导航摘要不复制分页对象；Agent 需要时查询具体 compact leaf。
+- 没有 `pagination` 表示该命令尚未发布经评审的分页能力，Agent 不得猜测。
+
+## 5. 渐进接入
+
+1. **legacy_only**：保持原输出，不公布分页声明。
+2. **dual_validate**：业务执行一次；影子构造并校验 `meta.pagination`，外部
+   legacy 字节不变。
+3. **unified_active**：输出独立 `meta.pagination`，Schema 公布同级
+   `pagination` 声明。
+4. **unified_stable**：Skill、示例和 Agent 审计均只读取 meta 分页。
+
+不增加 `contract_version`、`--output-contract` 或分页协议别名。
+
+## 6. 验收
+
+每个分页命令至少验证：
+
+1. 有下一页时 `endpoint_exhausted:false` 且 token 非空。
+2. 终页和空终页为 `endpoint_exhausted:true` 且无 token。
+3. 分页矛盾产生 typed failure，不 panic、不静默停止。
+4. `cursor_parameter` 在 Help/Schema 中真实存在。
+5. compact/full 的 `result`、`pagination` 分别 JSON 等价。
+6. `data_schema` 不包含分页控制字段。
+7. 运行时 `data` 不包含迁移后的分页控制字段。
+8. dual validate 与 active 都只消费一次上游响应。
+9. Agent 逐命令扫描结果进入评测台账；不提交生成 Schema JSON fixture。
+
+### DevApp 首批落地
+
+以下 8 个终结命令已发布独立 `pagination` Schema；运行时统一输出只在
+`meta.pagination` 返回分页控制信息：
+
+- `dev app list`
+- `dev app permission list`
+- `dev app event list`
+- `dev app version list`
+- `devapp +list`
+- `devapp +permission-list`
+- `devapp +event-list`
+- `devapp +version-list`
+
+两套既有命令前缀继续保留。原子命令的业务记录字段为 `data.items`；Shortcut
+保留既有业务投影（例如 `data.apps`、`data.permissions`、`data.events`、
+`data.versions` 以及 `data.count`），但两套入口都不再在业务数据中公布
+`hasMore/nextCursor`。
+
+## 7. 对齐依据
+
+GWS 用请求参数和 response schema 描述分页事实；Lark 在统一输出层维护分页
+元数据。DWS 采用更明确的分层：业务 `data` 保真承载记录，框架 `meta` 承载
+续跑状态，Schema 用独立能力把 token 与下一次 CLI 参数连接起来。

docs/reference.md

@@ -5,6 +5,8 @@
 | Variable | Purpose / 用途 |
 |---------|---------|
 | `DWS_CONFIG_DIR` | Override default config directory / 覆盖默认配置目录 |
+| `DWS_AGENT_PRODUCT` | Optional, caller-declared Agent product sent as `x-dws-agent-product` (for example `qwenwork`) for downstream logs/BI and used as the IM `clawType` display label when `--ai-tag` is enabled. `--ai-tag` defaults to `true`, so a configured Product changes the displayed label by default. With `--ai-tag=false`, native `chat message send` / `reply` calls send an empty `clawType`, while shortcut calls omit the argument. Surrounding ASCII spaces/tabs are trimmed; the remaining value must be at most 64 bytes and match `^[A-Za-z0-9][A-Za-z0-9_-]*$`. Unset or empty values omit the Header and use the edition's IM display default. This client never uses Product to change the separate HTTP `claw-type` PAT/routing label. / 可选、由调用方声明的 Agent 产品标识，经校验后作为 `x-dws-agent-product` 发送，并用于 IM 小尾巴；`--ai-tag` 默认为 `true`，因此配置 Product 后默认会改变展示标签。使用 `--ai-tag=false` 时，原生 `chat message send` / `reply` 发送空的 `clawType`，shortcut 调用则省略该参数。未设置时省略请求头且 IM 使用发行版默认值；本客户端不会用 Product 修改独立的 HTTP `claw-type` |
+| `DWS_AGENT_HOST` | Optional, caller-declared Agent runtime form sent as `x-dws-agent-host` (for example `cloud` or `desktop`) for downstream logs/BI. Surrounding ASCII spaces/tabs are trimmed; the remaining value must be at most 64 bytes and match `^[a-z0-9][a-z0-9_-]*$`; unset values are omitted. This client does not use Host for PAT, authentication, Discovery, or MCP endpoint selection. / 可选、由调用方声明的 Agent 运行形态，经校验后作为 `x-dws-agent-host` 发送给下游日志/BI；本客户端不使用该值进行 PAT、鉴权、Discovery 或 MCP 端点选择，未设置时省略 |
 | `DWS_<PRODUCT>_MCP_URL` | Override a product MCP endpoint for local development / 本地开发时覆盖指定产品 MCP endpoint |
 | `DWS_CLIENT_ID` | OAuth client ID (DingTalk AppKey) |
 | `DWS_CLIENT_SECRET` | OAuth client secret (DingTalk AppSecret) |
@@ -12,6 +14,64 @@
 | `DWS_ALLOW_HTTP_ENDPOINTS` | Set `1` to allow HTTP for loopback during dev / 设为 `1` 允许回环地址 HTTP，仅用于开发调试 |
 | `DWS_DISABLE_KEYCHAIN` | macOS only. Set `1` to skip system Keychain for the encryption key and use file-based storage (same scheme as Linux). For sandboxed runtimes (e.g. Codex App) that block Keychain APIs. Weakens at-rest protection — DEK and ciphertext live in the same directory. / 仅 macOS。设为 `1` 时跳过系统 Keychain，密钥以文件形式存储（与 Linux 一致）。用于 Keychain API 被拦截的沙盒环境（如 Codex App）。代价是 DEK 与密文同目录，保护强度低于默认方案 |
 
+### Agent Product, Host, and `claw-type` / Agent 产品、运行形态与 `claw-type`
+
+`DWS_AGENT_PRODUCT` and `DWS_AGENT_HOST` are caller-declared observation
+signals. They are not credentials, attestations, or proof of the calling
+host's identity. The CLI validates and emits `x-dws-agent-product` and
+`x-dws-agent-host`, but does not use either value to derive its authentication,
+PAT mode, Discovery behaviour, or ordinary MCP endpoint selection. Downstream
+services own and must document their own contracts for these caller-declared
+Headers.
+
+Service integrators should treat both Headers as untrusted input, allowlist
+expected values, and should not grant access, bypass authentication, or skip
+authorization solely because a Header claims a particular Product or Host.
+
+The HTTP `claw-type` Header is a separate, edition-fixed PAT/routing label:
+`openClaw` in the open-source build. `DWS_AGENT_PRODUCT` never changes it or
+PAT `hostControl.clawType`. On IM send/reply operations with `--ai-tag`,
+however, a valid non-empty Product value is used as the `clawType` tool
+argument so the delivered message carries the matching “Send from AI” label.
+Because `--ai-tag` defaults to `true`, this display change is enabled by
+default for callers that set Product. With `--ai-tag=false`, native
+`chat message send` / `reply` calls serialize `clawType: ""`, while shortcut
+calls omit the argument; this client does not assume downstream services treat
+an empty value and an absent key as equivalent. The display-value precedence
+when the tag is enabled is valid non-empty `DWS_AGENT_PRODUCT`, then the active
+edition's `ClawTypeValue`, then `openClaw`.
+
+Do not set arbitrary Product values that the target downstream and IM services
+have not explicitly enabled; an unknown value may be ignored or may not render
+the expected label.
+
+For QwenWork, report the dimensions separately:
+
+```bash
+DWS_AGENT_PRODUCT=qwenwork
+DWS_AGENT_HOST=cloud       # or desktop
+```
+
+Older combined Host labels such as `qwenwork_cloud` still satisfy the generic
+syntax for compatibility, but new integrations should use the two-dimensional
+convention above.
+
+`DWS_AGENT_PRODUCT` 和 `DWS_AGENT_HOST` 均由调用方声明，不是认证凭据，也不能证明
+真实宿主身份。CLI 只负责校验并发送 `x-dws-agent-product` 与 `x-dws-agent-host`，
+不会用它们派生本客户端的鉴权、PAT 模式、Discovery 行为或 MCP 端点；下游服务的
+使用契约由对应服务自行定义和说明。HTTP `claw-type` 是发行版固定的 PAT/路由标签，
+开源版固定为 `openClaw`，不受 `DWS_AGENT_PRODUCT` 影响。
+
+服务集成方应将这两个请求头视为不可信输入并对白名单值做校验，不应仅因请求头声明了
+某个 Product 或 Host 就授予访问、绕过认证或跳过鉴权。
+
+`--ai-tag` 默认为 `true`，因此配置合法非空 Product 后，默认发送的 IM 工具参数
+`clawType` 及小尾巴会随之改变。传入 `--ai-tag=false` 时，原生
+`chat message send` / `reply` 会发送 `clawType: ""`，shortcut 调用则省略该参数；
+本客户端不假定下游会将空值与键缺失等价处理。启用小尾巴时，展示值优先级依次为
+`DWS_AGENT_PRODUCT`、当前发行版的 `ClawTypeValue`、`openClaw`。不要传入目标下游及
+IM 服务未明确支持的 Product 值，否则可能被忽略或无法展示预期标签。
+
 ## Exit Codes / 退出码
 
 | Code | Category | Description / 描述 |
@@ -34,7 +94,7 @@ With `-f json`, error responses include structured payloads: `category`, `reason
 dws contact user search --query "Alice" -f table   # Table (default, human-friendly / 表格，默认)
 dws contact user search --query "Alice" -f json    # JSON (for agents and piping / 适合 agent)
 dws contact user search --query "Alice" -f raw     # Raw API response / 原始响应
-dws schema -f pretty "dev app create"                # Pretty helper-only schema view / helper-only schema 彩色分区展示
+dws schema -f pretty "calendar event create" --compact # Pretty Agent schema view / Agent Schema 彩色查看
 ```
 
 ## Dry Run / 试运行
@@ -51,40 +111,60 @@ dws contact user search --query "Alice" -o result.json
 
 ## Schema Introspection / Schema 查询
 
-静态端点模式下，产品命令和 flag 以当前二进制的 `--help` 与内置 Skill 为准。`dws schema` 仅保留 helper-only 子树（如 `dev.*`）的 schema 查询。
+`--help` 展示当前二进制的 Cobra 命令和可接受 flag，`dws schema` 查询同版本运行时组装的 Agent 命令契约。Schema 查询不访问 MCP endpoint、不执行 `tools/list`，也不搜索钉钉文档或任何业务数据。
+
+Schema 的稳定 `canonical_path`、主 CLI 路径和 aliases 收集自命令树叶节点上的 `ContractFinal.Identity`（`CollectIdentitySpecs`），并在发布时逐项绑定当前 Cobra tree。原 reviewed `schema_command_registry/` 已退役，身份变化通过编辑叶节点声明完成。Native annotation 只做实现一致性校验；Catalog 是该统一强类型契约的发布输出，不作为命令发现或下一轮生成的输入。
 
 ### 路径写法
 
 ```bash
-dws schema                                  # 静态端点模式提示
-dws schema "dev app create"                 # CLI 空格路径
-dws schema --cli-path "dev app create"      # 显式 flag（脚本友好，免转义）
-dws schema -f pretty "dev app create"       # ANSI 着色分区展示（人肉查看最舒服）
+dws schema                                      # 当前公开产品面的紧凑概览
+dws schema calendar --compact                   # Agent 产品视图
+dws schema "calendar event" --compact          # Agent 分组视图
+dws schema "calendar event create" --compact   # Agent leaf（CLI 空格路径）
+dws schema calendar.create_calendar_event --compact     # Agent leaf（canonical path）
+dws schema --cli-path "calendar event create" --compact # Agent leaf（显式 CLI path）
+dws schema "calendar event create"             # full leaf，仅用于映射/provenance 审计
+dws schema --all                                # 全部工具的完整 leaf Schema，用于审计/CI/baseline
 ```
 
-helper-only schema 以 CLI 路径为准；普通产品命令请使用 `dws <path> --help` 查看参数。
+兼容入口 `dws schema list` 等价于根概览。`schema --all` 是完整导出：每个工具都包含完整 leaf 参数、约束和安全语义。它输出很大，只用于明确要求的全量导出、审计、CI 或参数 baseline；普通 Agent 任务应按概览、产品/分组、leaf 渐进查询，不要把 `--all` 直接注入上下文。`schema --all --compact` 虽受支持，但会裁掉 provenance 和接口映射字段，不能作为完整 baseline。
+
+省略 `--compact` 的 full leaf、`--all` 中对应工具和 Catalog full tool 均由同一个 resolved `ToolSpec` 投影，内容必须一致；compact leaf 仅做字段白名单投影，不重新解析语义。概览、产品/分组和 Catalog summary 也来自同一 `ToolSpec`。通过 alias 查询时，只允许路径视图发生变化，参数、安全和接口契约不得变化。
+
+`--compact` 是 Schema 的稳定 Agent 字段白名单，也是普通 Agent 查询的规范选项。它保留 CLI 参数、组合约束、选择和安全语义，但有意省略 `interface_ref`、参数 `property/interface_type` 与 provenance。检查这些映射/审计字段时，使用 full leaf 并通过 `--jq` / `--fields` 精确投影。若兼容旧二进制时收到 `unknown_flag: --compact`，用同一个 Schema 查询去掉 `--compact` 重试；这只降低输出裁剪能力，不表示 leaf 缺失。
+
+### Schema、Help 与业务数据的边界
+
+| 问题 | 事实源 |
+|------|--------|
+| 命令是否由当前二进制暴露、Cobra 接受哪些 flags | `dws <path> --help` |
+| Agent 选哪个命令、CLI 参数与组合约束、risk/confirmation | 对应的 Agent leaf `dws schema "<path>" --compact` |
+| CLI↔RPC 参数映射、接口绑定与 provenance | full leaf 配合 `--jq` / `--fields` 精确投影 |
+| 当前钉钉中的文档、文件、日程、消息等业务数据 | 实际执行 `dws doc read`、`dws drive search` 等 read/search/list 命令 |
+
+Schema 与 Help 冲突表示发布契约漂移，不能静默猜测。执行参数必须以 Cobra 实际接受的 flag 为准；安全语义冲突时采用更保守的处理（例如先确认）或停止执行并报告漂移。完成命令发现后，仍必须执行真实业务命令；`dws schema` 本身不会读取或搜索业务内容。
 
 ### 单工具输出字段
 
 | 字段 | 说明 |
 |------|------|
-| `name` / `cli_name` / `canonical_path` | MCP RPC 名 / CLI 叶子名 / helper-only canonical path |
-| `group` | CLI 父级 group 路径（dot-separated） |
-| `title` / `description` | 工具名/说明（overlay 优先） |
-| `parameters` / `required` | MCP 输入 JSON Schema 的 properties / required |
-| `output_schema` | MCP 输出 Schema（上游下发时才有） |
-| `sensitive` | 敏感写操作，需 `--yes` 确认 |
-| `auth` | DingTalk 授权元数据，包括 `requiredScopes` / `requiredPermissions` / `recommendedScopes` / `grantProductCodes` / `riskAction` / `confirmationRequired` |
-| `annotations.destructive_hint` | 对齐 MCP 2025+ annotations，目前从 `sensitive` 映射 |
-| `flag_overlay[param]` | CLI 层对 MCP 参数的改写：`alias` / `transform` / `transform_args` / `env_default` / `default` / `hidden` |
+| `canonical_path` / `primary_cli_path` / `aliases` | 稳定工具 ID、主 CLI 路径和兼容路径 |
+| `product_id` / `interface_ref` | CLI 产品与实际 MCP product/RPC binding |
+| `title` / `description` / `agent_summary` | 人类说明、接口说明和 Agent 摘要 |
+| `parameters.<flag>` | CLI flag 的类型、属性名、required、默认值、格式、枚举和条件必填 |
+| `constraints` | one-of、互斥、联动等组合约束 |
+| `effect` / `risk` / `confirmation` / `idempotency` | Agent 执行与安全策略 |
+| `use_when` / `avoid_when` / `examples` | Agent 选择提示和示例 |
+| `reviewed` / `agent_source_refs` | 语义审核状态与来源追踪 |
 
-**调试 `--flag` 行为的第一站**是 `flag_overlay` —— 比如 `--users 0232...` 能不能直接用，看 `receiverUserIdList.transform == "csv_to_array"` 即可判断。
+`parameters.<flag>.required` 是按来源 precedence 解析后的 Agent 参数契约；`cli_required=true` 才表示 Cobra 将该 flag 标记为硬必填。条件必填或别名选择通过 `required_when` 和 `constraints.require_one_of` 表达。`required` 不直接复制 MCP input schema，也不取代 Cobra 的实际执行校验。
 
 ### 筛选输出
 
 ```bash
-dws schema "dev app create" --jq '.tool.parameters'               # 只看参数 schema
-dws schema "dev app create" --jq '.tool.required'                 # 只看必填字段
+dws schema "calendar event create" --jq '.parameters'                              # 只看参数
+dws schema "calendar event create" --jq '[.parameters | to_entries[] | select(.value.required)]'  # 只看 Agent required 参数
 ```
 
 ## Shell Completion / 自动补全

docs/releasing.md

@@ -0,0 +1,202 @@
+# 发布手册（预发 / 正式）
+
+发布只走一条受控链路：GitHub Actions 的 `Release` workflow 负责版本分配、封板、构建、签名和下游发布；Homebrew Formula 在不可变 Release 资产及 checksum 通过校验后，由同一 workflow 直接写入 `main`，不再创建二次 PR。本地 `dws-release` 仍是兼容入口，但不再要求某一台固定电脑承担打包；不要直接运行 `goreleaser release`，也不要手工补打、移动或复用 tag。
+
+发布前必须完成平台治理：目标 GitHub 仓库已启用 immutable releases，`main` 精确要求 `CI` workflow 的九个 context：`Lint`、`Test`、`Coverage`、`Policy`、`Edition`、`Interface Integrity`、`AI Behavior`、`CLI Smoke`、`Mock MCP`。云端入口会在封 tag 前检查 immutable releases、当前 SHA 的全部九个 context、Environment 保护规则和在途 Release；`v*` tag ruleset 仍需仓库管理员预先配置。
+
+## 推荐入口：GitHub 云端发布
+
+入口页面是 [GitHub Actions → Release](https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/actions/workflows/release.yml)。在仓库页面依次点击 `Actions` → `Release` → `Run workflow` 即可操作；不需要通过 Agent 或本地机器触发。
+
+只有得到该仓库明确授权、最终权限为 `write`、`maintain` 或 `admin` 的协作者可以运行发布操作，workflow 还会对发起人和重新运行者做同样的权限复核。没有仓库写权限的外部贡献者以及仅有 `read` / `triage` 权限的成员不能规划或发布版本。两个渠道的授权边界如下：
+
+- beta：上述任一内部成员都可以直接规划和发布，不需要人工审批。
+- stable：上述任一内部成员都可以规划并发起发布；完成只读规划和治理检查后，workflow 会在 `release-stable` Environment 等待另一名仓库管理员通过 `Review deployments` 签收。申请人不能批准自己的请求，批准后原 run 自动继续，无需重新触发。
+
+基于当时最新的 `main` 发起发布：
+
+1. 在上述 `Release` 页面选择 `Run workflow`，分支必须是默认分支 `main`。
+2. `release_operation=plan`，选择 `release_channel=beta|stable`；仅在开始新 beta 线时选择 `release_bump=patch|minor|major`。
+3. workflow summary 会给出唯一的下一版本。把对应的精确 `CHANGELOG.md` 章节通过 PR 合入 `main`。
+4. 再次运行，改为 `release_operation=publish`。beta 会直接进入自动化发布；stable 会在封 tag 前等待管理员签收。
+
+`plan` 是纯只读操作，不创建 tag、预留版本号或生成包。CHANGELOG 合入期间若另一个发布先占用了该版本，`publish` 会重新分配并因 CHANGELOG 章节不匹配而拒绝，需要重新 plan。`publish` 会先再次确认 dispatch SHA 仍是当前 `main`、Code Admission 和平台治理均通过，再由唯一的 write job 使用 GitHub API 原子创建 annotated tag；同一次 run 随即进入既有的跨平台构建、GitHub/npm、可选 OSS/Gitee 发布和 Homebrew 直交付 DAG。内置 `GITHUB_TOKEN` 创建的 tag 不依赖第二条 workflow 被再次触发。
+
+为缩短封板前后的关键路径，`publish` 的只读版本规划会与平台治理检查并行，seal 仍严格等待二者成功；plan 在 candidate annotated tag 上验证过的 contract 和 stable/beta baseline 会绑定进 seal，并由 seal 后的 tag authority 检查复用。Code Admission 状态与 immutable-releases 治理仍会在 seal 后再次读取，避免 preflight 与发布之间的状态变化被忽略。随后三类只读门禁（release automation、命令兼容性、multi-profile E2E）与 GoReleaser 构建并行；Node/archive 等仅供后处理使用的工具也延后到构建完成后安装。并行和已验证结果复用只改变调度，不降低发布门禁：任何一条验证失败都会阻止 GitHub Release、npm、镜像和 Homebrew 发布，delivery proof 也要求三条验证 job 全部成功。
+
+OSS 镜像默认不参与发布 DAG，适用于尚未创建 Bucket 的仓库。云端封板会把当时的仓库变量 `ENABLE_OSS_MIRROR=true` 记录为不可变 tag 元数据 `OSS-Mirror: enabled`，否则记录为 `deferred`；后续发布和撤回只读取该 sealed policy，不读取变量的当前值。`enabled` 继续对缺失凭据、无效 Bucket、上传、pointer 和撤回失败保持 fail-closed；`deferred` 明确跳过不存在的渠道。为避免补发后撤回遗漏，deferred 版本暂不接受 `repair_oss_version`，启用 OSS 只影响后续新 tag，直到补齐可审计的不可变 repair 证明。
+
+## 自动版本规则
+
+- beta：如果存在尚未封正式版的最高版本线，自动取 `beta.N+1`；否则从最新已分配正式版按所选 patch/minor/major 开新线并取 `beta.1`。
+- stable：先锁定最高开放版本线上的最新已分配 beta，再要求它已成功交付且未撤回；不会跳过失败/撤回的最新 beta 去选择更早版本。正式版 core 与该 beta 完全相同。
+- `vX.Y.Z`、`vX.Y.Z-beta.N` 一经分配就永久占用。撤回时创建 `withdrawn/v...` 墓碑，原编号永不复用。
+- 例如撤回 `v1.0.53-beta.5` 后，下一 beta 是 `v1.0.53-beta.6`；撤回正式版 `v1.0.53` 后，下一 patch 修复线是 `v1.0.54-beta.1`，验证后再发布 `v1.0.54`。
+- 如果最新 beta 已撤回，禁止直接用更早 beta 晋级正式版；必须先构建下一个 beta。
+
+## 全平台撤回与回滚
+
+已公开版本出现问题时，在 GitHub Actions 运行 `Withdraw release`，分支必须选择当前默认分支 `main`，并填写：
+
+- `version`：精确版本，例如 `v1.0.53` 或 `v1.0.53-beta.5`。
+- `reason`：8–300 字符的单行公开原因。
+- `confirmation`：精确输入 `WITHDRAW <version>`，例如 `WITHDRAW v1.0.53`。
+
+该 workflow 使用与发布相同的串行 publication lock，并进入受保护的 `release-withdrawal` environment。它只接受已经由 Release workflow 完整交付的 public immutable release，自动选择同一渠道中最新的、更早且未撤回的完整版本作为回退目标，然后按以下顺序执行：
+
+1. 先创建永久 annotated tag `withdrawn/<version>`，记录原 tag object、commit、原因、申请人和 workflow run。这个墓碑是版本号永久占用记录，永不移动、永不删除。
+2. 先验证 Homebrew Formula；若它仍指向问题版本，先创建回退 PR，再继续其他渠道撤回。这样 PR 创建失败时只留下可安全续跑的墓碑，不会先造成渠道分裂。若 Formula 尚未指向问题版本或已经处于安全版本，则直接校验。
+3. GitHub Release 先标记为 withdrawn；npm 精确版本执行 `deprecate`，并把 `latest` / `beta` dist-tag 回退；只有目标 tag 封存了 `OSS-Mirror: enabled` 时，OSS 才会先补齐回退版本资产，再移动 `latest.txt` / `beta.txt` 并删除问题版本目录；启用 Gitee 时同样先补齐回退 Release，再删除问题 Release 和 tag。
+4. npm 以及目标 tag 启用或发布时配置的镜像渠道均已验证安全后，删除 GitHub 上的问题 Release 和原 `v...` tag，并验证 `/releases/latest` 对正式版回到安全版本。若本次创建了 Homebrew PR，run 最后故意保持失败，直到另一名维护者审核合入；合入后，从新的 `main` 使用完全相同的 version、reason 和 confirmation 重跑并完成。永久 `withdrawn/v...` 墓碑始终保留。
+
+GitHub、npm、OSS、Gitee 和 Homebrew 的“回滚”指新的安装、升级和渠道解析不再拿到问题版本。已经装到用户电脑上的二进制无法被服务端强制降级；用户必须重新安装回退版本、安装后续修复版，或使用 CLI 自带的本地 rollback 能力。npm 不执行 `unpublish`：问题版本保留明确的弃用警告，但 `latest` / `beta` 不再指向它；即使 registry 允许删除，已发布过的版本号也不会重新使用。
+
+撤回前必须存在同一渠道中更早、完整交付且未撤回的安全版本；若目标是该渠道第一个版本、没有安全候选，workflow 会在创建墓碑或修改任何渠道前 fail closed，需要先决定明确的替代策略。CLI 本地 rollback 也只有在本机仍保留上一次升级备份时可用。
+
+撤回以“精确版本”为单位，不会因为正式版曾由某个 beta 晋级就隐式级联修改另一个渠道。若同一缺陷同时存在于正式版及其 beta，应先撤回正式版，再撤回对应 beta，并分别使用各自的精确确认串；每次都只会把该渠道回退到自己的安全候选。
+
+撤回正式版 `v1.0.53` 后，`v1.0.53` 仍被墓碑视为已分配。下一次 patch 发布从 `v1.0.54-beta.1` 开始，验证后晋级 `v1.0.54`。撤回 `v1.0.53-beta.5` 后，同一开放版本线继续为 `v1.0.53-beta.6`；不会退回或复用 `beta.5`。
+
+## 兼容入口：本地发布
+
+安装发布 Skill 后直接运行：
+
+```bash
+dws-release
+```
+
+零参数会进入引导模式。仓库内的等价入口是 `./scripts/release/dws-release.sh`。第一次使用只需配置一次生产发布远端，命令会把远端名及其规范化仓库身份一起保存在当前 Git 仓库中：
+
+```bash
+dws-release config --remote origin
+```
+
+之后命令按仓库状态自动走到正确步骤：缺少精确 CHANGELOG 章节时只生成模板并停止；补全、提交并合入 `main` 后，再运行同一条命令就会安全快进本地 `main` 并执行完整预检。若同名 remote 后续被改指向其他仓库会直接拒绝。官方仓库不再接受本地 `--publish`，命令会直接提示上述 Actions 页面；本地入口不能绕过 beta/stable 的统一授权。
+
+Release workflow 不再监听新建的 `v*` tag；直接推 tag 不会发布 GitHub Release、npm 或镜像渠道。所有新 beta/stable 都必须从云端页面进入统一授权和审计链路；历史失败版本仍可通过受保护的 recovery 兼容处理。
+
+## 发布模型
+
+```text
+main 上的候选代码 + beta CHANGELOG
+  → vX.Y.Z-beta.N（预发验证）
+  → 补正式 CHANGELOG；允许继续通过 PR 合入新 commit
+  → vX.Y.Z（正式发布，封板提交必须包含该 beta 提交）
+```
+
+云端入口自动选择本次最新、已交付且未撤回的 beta；本地入口必须显式指定。流水线要求该 beta 已成功交付、未撤回，且 beta 提交必须位于正式发布封板提交的历史中——不能跳过 beta 直接发正式版，但允许在 beta 之后把经过 review 合入 `main` 的 commit 一起发布。
+
+## 预发发布
+
+运行统一入口：
+
+```bash
+dws-release v1.2.3-beta.1
+```
+
+如果 CHANGELOG 尚不存在，该命令只生成模板并停止。补全内容、删除所有 `TODO`，提交后通过 PR 合入 `main`；然后重新运行完全相同的命令，它会执行完整预检：
+
+```bash
+dws-release v1.2.3-beta.1
+```
+
+预检包含测试、策略检查、旧正式版命令树兼容检查、全平台打包、npm 安装验证，以及 macOS 环境下的 Homebrew 安装验证。它还会从默认分支触发一次无发布权限的 `Release governance preflight`，用正式流水线相同的身份检查该精确 commit 的九个 Code Admission context 和 immutable releases。通过后回到上述 Actions 页面选择 beta 和 `release_operation=publish`；云端会重新绑定当前 `main`，然后直接进入 beta 自动发布，不需要人工审批或输入确认短语。
+
+## 正式发布
+
+beta 验证通过后，运行正式版入口：
+
+```bash
+dws-release v1.2.3 --from-beta v1.2.3-beta.1
+```
+
+首次运行只生成正式版 CHANGELOG 并停止。补全内容、删除 `TODO`，提交后通过 PR 合入 `main`；重新运行同一条命令做完整预检：
+
+```bash
+dws-release v1.2.3 --from-beta v1.2.3-beta.1
+```
+
+预检通过后，在 Actions 页面选择 stable 和 `release_operation=publish`。云端入口会按上述规则唯一选择 beta，并把它写入 stable annotated tag 的 `From-Beta` 元数据；在创建 tag 前必须由另一名仓库管理员签收。
+
+## CHANGELOG 契约
+
+每个 tag 必须有唯一、非空且不含 `TODO/TBD` 的精确章节：
+
+```markdown
+## [1.2.3-beta.1] - 2026-07-11
+
+### Changed
+
+- 本次 beta 验证的用户可见变化。
+```
+
+正式版使用 `## [1.2.3] - YYYY-MM-DD`。该章节会直接成为 GitHub Release Notes。
+
+## CI/CD 保证
+
+- 只接受 `vX.Y.Z-beta.N` 和 `vX.Y.Z`，且新版本必须高于上一正式版。这里的“上一正式版”必须同时具备公开非草稿 GitHub Release 和同 tag/commit 的成功 Release workflow；只有 tag、没有交付成功的孤儿版本会阻断后续发布，要求走机器核验恢复补齐。云端 tag 会固定 `Release-Run`、requester、commit 和版本分配指纹，交付验证按该精确 run/attempt 及完整 job graph 取证，不接受任意 `workflow_dispatch`。历史版本若曾通过专用 recovery workflow 完成交付，只能使用仓库内 `delivered-stable-recoveries.json` 中精确到 tag、commit、run、workflow SHA 与 attempt 的 reviewed 证据。
+- tag 必须由云端 seal job 创建为 annotated tag；封板提交必须已通过 PR 合入并包含在远端 `main` 历史中。流水线允许其后 `main` 继续前进，但始终要求封板提交位于 `main` 历史中。
+- 日常 CI 和发布前都会对比“最新已交付正式版”的完整命令树；若长时间预检期间该 baseline 发生变化，会针对新的 baseline 重新比较。
+- GoReleaser 只构建；Darwin 重签、checksums 重算和 npm 安装验证通过后，才统一上传 GitHub Release 的最终产物。
+- 六个平台归档会逐个解包并核验二进制内嵌版本；公开资产集合、checksums 集合和 npm tarball integrity 都必须精确一致。npm tarball 固定由 npm `10.9.2` 打包，避免重跑时因 runner 自带 npm 漂移产生不同字节。
+- stable 发布到 npm `latest`；prerelease 发布到 npm `beta`。启用 `ENABLE_OSS_MIRROR=true` 后，stable 同步 OSS `latest.txt` 和共享安装脚本，prerelease 只同步 OSS `beta.txt`，不会覆盖稳定入口。
+- Release workflow 使用一个最多容纳 100 个 pending run 的串行 publication queue；版本规划、云端封板、发布、恢复、修复和撤回共享同一发布锁。
+- 云端 seal 创建远端 tag 后，后续发布归同一 run 所有；发布中途失败时先重跑同一 run 的失败 jobs，必须跨 run 时走机器核验恢复，禁止改 tag 指向或复用版本号。只有已经公开版本经过受保护的全渠道撤回并留下永久 `withdrawn/...` 墓碑后，撤回 workflow 才会在最后一步删除原 tag。
+
+npm 补发只允许从默认分支触发 Release workflow 的 `repair_npm_version`。它只支持启用 immutable releases 后、由本流水线成功产出的公开 immutable release：目标必须是 `main` 历史中的 annotated tag，并且同 commit 的 `Build immutable GitHub Release` job 已成功。即使后续 npm 分发失败，这个独立的产物封存边界仍可作为补发依据。补发会用目标 commit 的 npm 模板重组包，逐平台核验资产和二进制版本，再发布到隔离的 `backfill` dist-tag，不会回滚 `latest` / `beta`。历史 mutable release 不进入自动补发路径，避免把可被替换的资产带入 npm。
+
+已启用的 OSS 或 Gitee 分发失败且 GitHub immutable Release、npm 已交付时，从受保护的默认分支触发
+Release workflow，并且只填写 `repair_oss_version` 或 `repair_gitee_version` 之一。channel
+repair 会精确绑定失败 tag run 的最新 attempt，且 OSS repair 要求 tag 的 sealed policy 为 `enabled`；contract、构建、Developer ID 签名、
+immutable GitHub 发布和 npm delivery 必须全部成功，且只能有一个 OSS/Gitee 下游失败，
+随后才会下载并重新校验原始资产、修复所选镜像。OSS repair 必须匹配失败的 OSS step；
+Gitee repair 还允许其 job 因该 OSS 失败而 skipped，此时只代表 Gitee backfill 成功，
+不会把仍未修复的 OSS 标成成功。该证据不能用于 beta → stable 或
+stable baseline，后两者仍要求整条 Release 成功或受保护 recovery 成功。不要重跑旧
+attempt 的单个 failed job，以免在 attempts 之间拼接交付证据。独立 Gitee release
+workflow 和本地直发脚本已停用，避免绕开 publication queue 或用重新构建的不同字节覆盖镜像。
+
+## 既有 tag 的紧急恢复
+
+云端封板或本地 tag push 已成功、但 Release workflow 失败且 GitHub Release 尚未公开时，不要新建临时 workflow、移动 tag 或跳过门禁。在最新且干净的 `main` worktree 运行：
+
+```bash
+dws-release recover v1.2.3-beta.1
+```
+
+命令会自动解析 annotated tag object、peeled commit，以及 tag 绑定的失败云端 run 或最近一次匹配的失败 tag-push run；也可以用 `--failed-run <run-id>` 精确指定。确认完整版本号后，它从默认分支触发受保护的恢复模式并等待完成。恢复模式必须满足：
+
+- 输入精确绑定原 annotated tag object、commit 和失败的 sealed `Release` run；云端 run 还必须与 tag 内的 run ID、attempt、requester 完全一致，commit 必须仍在 `main` 历史中。
+- 目标只允许不存在 GitHub Release 或仍为 Draft；已经公开的版本不能全量重建：单个下游故障走对应的 channel repair，版本本身有问题则走受保护的全平台 withdrawal。
+- 恢复不再进入人工审批 environment。workflow 会机器核验 tag object、commit、原失败 run/attempt、请求人、完整 seal metadata、`main` 祖先关系以及 Release 状态；任一事实不一致都会在构建前 fail closed。
+- 恢复复用正常的 contract、构建、Developer ID 签名、资产校验、immutable 发布、Homebrew、npm，以及已启用的 OSS jobs，不存在 recovery 专用 publisher 或门禁跳过。
+- 如果 GitHub Release 已在 recovery 中封存、后续 Homebrew/npm 校验发生瞬时失败，只重跑该 run 的 failed jobs；流水线仅在隐藏 run marker、tag object、commit 和 finalized artifact 字节全部精确一致时复用公开 Release。
+
+成功的默认分支恢复 run 会成为后续 beta → stable 和 stable baseline 验证的可审计交付证据；历史临时分支恢复仍只接受 reviewed manifest 中的固定证据。
+
+seal job 写入 tag 后如果只因 GitHub API 瞬时 404/429/5xx 或后续 job 失败，可直接使用 GitHub 的 “Re-run failed jobs”。同一 run 会精确复用原 release-plan；seal 只在 version、tag object、commit、channel、beta 来源、OSS policy、请求人、run ID 和完整 message 全部匹配且原 attempt 不大于当前 attempt 时认领已有 tag。不同 run 或任一字段不匹配时不会认领。GitHub Release 尚未公开且必须跨 run 重建时走上述机器核验 recovery；已经公开且仅 npm/OSS/Gitee 某一渠道失败时走对应 repair；版本内容本身有问题时走 withdrawal。
+
+OSS 的 `latest.txt` / `beta.txt` 是镜像频道元数据；当前仓库安装器仍主要从 GitHub/Gitee 解析版本。启用 OSS 后，发布和撤回把它作为受控分发渠道处理，保证一旦外部消费者接入该 pointer，也不会继续解析到已撤回版本；未启用时两条流程都明确跳过不存在的 OSS 渠道。
+
+Release workflow 会生成 Darwin/Linux 双架构 Formula，并在不可变资产逐个校验后，由 `HOMEBREW_PR_TOKEN` 所属的受控发布身份只提交对应 stable 或 beta Formula 文件到 `main`，不再创建二次发布 PR；并发 `main` 更新会以全新 clone 最多重试三次，绝不 force push。该身份的提交不会依赖另一轮 CI 来补齐证明：workflow 只在确认该 commit 单父、唯一改动为目标 Formula、内容与本次已验证产物逐字节一致，且父 commit 九项 Code Admission 全绿后，直接为 Formula-only commit 封存同名九项成功 checks，避免下一次发布因缺失 contexts 被卡住。撤回 workflow 暂时仍使用相同模板和回退版本 checksums 打开反向 PR；问题 GitHub Release 会先被移除以阻止新安装，永久墓碑和 workflow 日志承担审计/续跑依据。
+
+## 平台治理前置
+
+仓库管理员还需要在 GitHub 平台配置以下不可由脚本替代的规则：
+
+- `main` 必须精确要求 `Lint`、`Test`、`Coverage`、`Policy`、`Edition`、`Interface Integrity`、`AI Behavior`、`CLI Smoke`、`Mock MCP` 九个 Code Admission context；Release workflow 也会通过 Checks API 再确认该封板 SHA 上九项全部成功。
+- 必须启用 immutable releases；它只保护启用后发布的 release，因此应在第一次使用新流水线前配置。为 `v*` 增加 tag ruleset，限制创建权限，并在 release 发布前保护 tag 的短暂窗口。
+- tag ruleset 还必须覆盖 `withdrawn/v*`：只允许受保护的撤回 workflow 创建墓碑，禁止更新或删除墓碑；同时应允许 Release workflow 创建新的 `v*`，允许撤回 workflow 在全部渠道回退后删除精确的问题 `v*`。若组织级规则阻止这两个 workflow 的预期动作，发布或撤回会 fail closed，不能靠手工移动 tag 绕过。
+- 配置 `RELEASE_GOVERNANCE_TOKEN` Actions secret，只授予目标仓库 `Administration: read`；内置 `GITHUB_TOKEN` 不具备 immutable-releases API 所需的仓库治理权限。每次本地预检和云端发布都使用这一个身份进行 fail-closed 验证。
+- 配置 `APPLE_CERTIFICATE_P12_BASE64`、`APPLE_CERTIFICATE_PASSWORD` 和具备发布权限的 `NPM_TOKEN`；撤回还要求该 npm 身份能够执行 `deprecate` 和修改 dist-tag。
+- 启用 OSS 镜像时，先创建有效 Bucket，再设置仓库变量 `ENABLE_OSS_MIRROR=true`，并配置 `OSS_ACCESS_KEY_ID`、`OSS_ACCESS_KEY_SECRET`、`OSS_ENDPOINT`、`OSS_BUCKET`，按需配置 `OSS_PREFIX`。启用后发布保持 fail-closed；撤回身份必须能够补齐安全版本资产、写 `latest.txt` / `beta.txt` 并删除问题版本前缀。尚未 provision Bucket 时保持该变量未设置或不等于 `true`，新 tag 会封存 `OSS-Mirror: deferred` 并跳过 OSS；该版本不能通过现有 repair 流程事后改成启用。
+- 若启用 Gitee fallback，设置 `ENABLE_GITEE_UPLOAD_FALLBACK=true`，并配置 `GITEE_TOKEN`、`GITEE_USER`、`GITEE_REPO`；该身份必须能够创建和删除目标仓库的 Release 与 tag。
+- 正常 Homebrew 发布使用现有的 `HOMEBREW_PR_TOKEN` 直接提交 Formula-only commit，不再创建 Homebrew PR，也不跑权限 canary。GitHub 不允许内置 Actions App 作为当前仓库 ruleset 的 bypass actor，因此两个默认分支 ruleset 都只给该 token 所属的指定发布管理员用户 `always` bypass；仓库脚本仍会限制提交路径、校验 Ruby、禁止 force push，并在并发更新时重新基于最新 `main`。
+- `HOMEBREW_PR_TOKEN` 应保持仓库范围的 `Contents: write` 与 `Pull requests: write` 权限；后者仅供撤回流程创建回退 PR。不要与 `RELEASE_GOVERNANCE_TOKEN` 复用，并定期审计 token owner 与 ruleset bypass actor 一致。
+- 创建 `release-beta` environment，只允许受保护分支且不配置 required reviewer；仓库内部 `write`、`maintain`、`admin` 成员的 beta 发布会直接通过该边界。
+- 创建 `release-stable` environment，只允许受保护分支，以仓库管理员为 required reviewer，禁止申请人自审并关闭管理员绕过。内部成员可以发起 stable，但必须由另一名管理员签收后才能封 tag 和写入任何发布渠道。
+- Release workflow 会在封 tag 前回读并验证上述两套 Environment 规则；规则缺失、stable reviewer 不再是仓库管理员、或 beta 被误加人工审批时都会 fail closed。
+- 创建 `release-withdrawal` environment，只允许受保护分支，设置至少一名 required reviewer、禁止申请人自审并关闭管理员绕过。撤回 workflow 会通过 API 复核这些规则；任何一项缺失都会在触碰 npm、OSS、Gitee、Homebrew 或 GitHub Release 前失败。
+- 仓库或组织的 Actions 策略必须允许 `Release` 与 `Withdraw release` workflow 的 `GITHUB_TOKEN` 获得各 job 声明的权限；正常发布由 `HOMEBREW_PR_TOKEN` 更新两个受控 Formula 路径，内置 token 只承担 workflow 自身声明的封板与校验写入。若撤回凭证采用 environment secret，确认 `release-withdrawal` 审批完成后能够读取撤回所需的 npm、OSS、Gitee 和 Homebrew 凭证。
+
+immutable releases，或任一 Code Admission context 缺失、未成功时，发布脚本会自动拒绝封 tag。tag ruleset 可能来自组织层，脚本不自动推断其最终作用范围；管理员确认不能省略，脚本约定也不能替代平台强制。

docs/robot-quickstart.md

@@ -6,7 +6,7 @@
 
 ## 第一步：安装 dws
 
-一键脚本会自动下载最新版二进制 + `dingtalk-dev` skill，只需要 curl（无需 go / git）。
+一键脚本会自动下载最新版二进制 + `dingtalk-misc` skill（开放平台应用文档落在 misc），只需要 curl（无需 go / git）。
 
 ### macOS / Linux
 

docs/schema-dynamic-endpoint-design.md

@@ -0,0 +1,322 @@
+# DWS Agent Schema 统一方案
+
+## 1. 核心定义
+
+DWS Schema 是当前二进制公开 CLI 的版本化 Agent 执行契约。它描述真实 Cobra 命令，并补充 Agent 选择、参数映射、组合约束、安全确认和接口事实。
+
+设计遵循三条硬规则：
+
+1. **Schema 描述 CLI，不制造 CLI。** `CommandRegistry`、ProductDecl / leaf `Contract`、metadata 和 Catalog 都不能凭空创建 Cobra 命令或 flag；registry 中的每个路径都必须精确绑定真实 runnable Cobra leaf。interface 事实由 leaf `Contract` / `ParamDecl` 声明（`schema_mcp_metadata` 已退役），**不得**从 MCP meta 生成 CLI flag（见 §4.1 同源决策）。
+2. **所有来源只解析一次。** 来源经过统一 resolver 进入 typed `SchemaRegistry`，所有查询、导出和门禁都消费同一个 `SchemaRegistry/SchemaIndex`。
+3. **Collector-first，Catalog 只出不进。** identity collector（`CollectIdentitySpecs`，遍历携带 `ContractFinal.Identity` 的 live Cobra leaves）是稳定 command identity/navigation 的唯一事实源；reviewed `schema_command_registry/` 已退役，identity 由声明（Contract）即代码提供，不再有独立的 reviewed identity 文件。production 通过 `RegisterSchemaSourceRoot` → `ResolveSchemaBuild` 组装 `SchemaRegistry`，并从它投影 ToolSpec wire 与 `ResolveMeta`。`cmd_schema_catalog` 只能生成 CI/local dump，`internal/cli/schema_catalog/`、`schema_meta_index.gob` 和 `schema_meta_index.json` 不得提交或成为运行时来源。`schema_agent_metadata/` / `schema_hints/` / `schema_command_registry/` 已退役；若存在则 policy 失败。生产 Agent selection / safety / interface 权威为 leaf `ContractFinal` 与 `ProductDecl`；`agent_metadata_inject.go` / `InstallBuildTimeAgentMetadataJSON` 仅作 `cmd_schema_catalog` CI/local dump 辅助，不得作为生产权威。
+
+Schema 不调用 MCP `tools/list`，不访问网络，也不读取用户本地 discovery cache。
+
+**flag / help / schema 参数面同源**（已决策）：Contract / LeafSpec 为 CLI 表面权威；分层字段归属与门禁 ID 见 [`flag-help-schema-homology.md`](flag-help-schema-homology.md)。
+
+## 2. 单向数据流
+
+```text
+identity collector (CollectIdentitySpecs)
+  walks live Cobra leaves carrying ContractFinal.Identity;
+  reviewed exclusions applied; single identity source
+  (reviewed schema_command_registry/ retired)
+                         |
+                         v
+             EffectiveCommandRegistry
+                         |
+                         v
+          exact binder to live Cobra tree
+       + native identity consistency assertions
+                         |
+                         v
+               BoundCommandRegistry
+                         |
+                         v
+live Cobra flag facts / typed parameter metadata
+  + leaf Contract (Safety / ContractDecl / ParamDecl → contract_final)
+  + ProductDecl (product routing prose; production Agent authority)
+  + schema_parameter_mapping_ledger.go (reviewed mapping exclusions / removals)
+  + leaf Contract.Interface / ParamDecl (declared interface facts)
+  + skills/mono Markdown              (evidence only; not concatenated)
+                         |
+                         v
+              source adapters + resolvers
+                         |
+                         v
+             one typed SchemaRegistry
+              (one ToolSpec per command)
+                         +
+                    typed SchemaIndex
+             +-----------+-----------+
+             |                       |
+             v                       v
+     build-time typed gates   RegisterSchemaSourceRoot
+                              -> ResolveSchemaBuild
+                              (runtime assembly; lazy Once)
+                                     |
+                                     v
+                         SchemaRegistry + SchemaIndex
+                         + ResolveMeta projection cache
+                           (in-process map; not gob/json fixture)
+                                     |
+               +---------------------+------------------+
+               |                     |                  |
+      overview/product/group        leaf              --all
+            projections          projection      full projection
+               |                     |                  |
+               +---------------------+------------------+
+                                     |
+                                     v
+                         runtime query + delivery gates
+                         (cmd_schema_catalog = CI/local dump only;
+                          InstallBuildTimeAgentMetadataJSON = dump helper,
+                          not production Agent authority)
+```
+
+`--help` 是 Cobra 自身的人类可读投影，不从 Catalog 生成。Schema projections 和 `--help` 共享同一真实 Cobra 命令面，但承担不同职责。Binder 之后不得再从 annotation、已退役 hint overlay 或生成 JSON 重新解析 command identity。
+
+## 3. 与 Lark 的关系
+
+DWS 与 Lark 保持**架构同构**，而不是强行复制字段：
+
+| Lark 分层 | DWS 对应层 |
+|---|---|
+| typed command/metadata registry | `EffectiveCommandRegistry`、`BoundCommandRegistry` 与最终 `SchemaRegistry` |
+| navigation catalog/index | 从同一 `ToolSpec` 派生的 `SchemaIndex` |
+| schema renderer/envelope | overview、product/group、leaf、`--all` projections |
+| API Commands ← 平台 OAPI meta | **不**作为 DWS 主路径；可选 1:1 MCP 透传子集见同源文档 §5 |
+| Shortcuts ← 手写声明 + Execute | LeafSpec / Shortcut + Contract 表面（路径 A） |
+
+共同点是：强类型 registry 持有已审核、已绑定、已解析的事实，index 只负责确定性导航，renderer 只投影，不重新读取来源或做 precedence。DWS 的 identity collector 从携带 `ContractFinal.Identity` 的 live Cobra leaves 收集 identity，绑定前生成唯一的 `EffectiveCommandRegistry`（reviewed `schema_command_registry/` 已退役），因此不存在 “native-first”、“legacy registry fallback” 或 Catalog fallback。飞书也是**分层单权威**（API 用平台 meta，Shortcut 用手写契约），不是全家只有 meta——DWS 对齐的是这一分层，而不是「用 MCP meta 生成全部 CLI」。
+
+DWS 内部 resolved model 为：
+
+```text
+SchemaRegistry
+  -> []ProductSpec
+       -> []ToolSpec
+            -> ToolIdentitySpec
+            -> []ParameterSpec
+            -> RuntimeSchemaConstraints + []RuntimeSchemaPositional
+            -> SafetySpec
+            -> InterfaceSpec
+            -> SelectionSpec
+            -> map[field]FieldProvenance
+```
+
+字段合并和 precedence 在进入该模型前完成。`map[string]any`/flat JSON 只允许存在于 renderer 和 snapshot/wire boundary，不能作为内部 resolver、navigation 或 gate 的第二套数据模型。
+
+DWS 当前对外仍保留兼容 wire：leaf 使用 flat `parameters`，安全和选择字段也保持现有键名。架构对齐不等于未版本化地切换到 Lark `inputSchema/outputSchema/_meta` envelope；若未来提供该格式，应作为明确版本的新投影，并保留现有兼容输出。
+
+## 4. 来源职责
+
+| 来源 | 负责内容 | 明确不负责 |
+|---|---|---|
+| Contract / LeafSpec / `corecmd.Spec` | **CLI 表面权威**：flags、defaults、required、enum、关系约束、运行时 Risk；编译为 cobra 与 help | canonical identity、selection 文案、虚构 RPC |
+| identity collector（`CollectIdentitySpecs`） | 从 live Cobra leaves 的 `ContractFinal.Identity` 声明收集稳定 canonical identity、primary CLI path、alias、exposure 和导航（reviewed `schema_command_registry/` 已退役） | 创建 Cobra 命令/flag、参数、安全、endpoint/token |
+| reviewed exclusions（`ReviewedRuntimeSchemaExclusions`） | exact、reviewed、带 reason 地将指定 public runnable leaf 排除出 effective 表面 | 运行时 fallback、prefix/wildcard 排除、创建命令 |
+| Go/Cobra | Contract 编译后的可执行投影：路径是否真实可执行、Cobra 接受的 flag、DefValue、help 文本 | 稳定 canonical identity、Agent 场景选择、虚构 RPC；**不得**成为与 Contract 平行的第二套 flag 权威 |
+| native Schema identity annotations | implementation-side consistency evidence；存在时必须与 `EffectiveCommandRegistry` 精确一致 | 提供、补全、推断或覆盖 identity |
+| typed parameter metadata / constraints | 由 Contract 约束投影而来的 `require_one_of` / 互斥等；以及仍需 reviewed 的 `required_when` 等 | 命令 identity |
+| `schema_parameter_mapping_ledger.go` | CLI flag 无直接 RPC property 的 exclusions / removals | 命令发现、risk 推断、创建 CLI flag；property 交付归 ParamDecl.Property |
+| leaf `Contract.Interface` / `ParamDecl` | 声明的 RPC identity 与 interface_* 事实（`schema_mcp_metadata.json` 已退役） | CLI identity、运行时路由、**创建 CLI flag**、risk 推断 |
+| ProductDecl + leaf `Contract.Selection` | reviewed selection / product routing prose（`contract_final`） | 创建 Cobra 命令或参数、改写 safety；`schema_hints/` 已退役 |
+| Skills/Markdown | 产品路由、工作流和使用建议 | 命令存在性和 flag 事实 |
+| `cmd_schema_catalog` CI/local dump（可选 `schema_catalog/` / meta-index） | resolved registry 的兼容序列化快照，仅供 jq/determinism；不得提交为 runtime 来源 | production delivery、`ResolveMeta` 权威、identity fallback、手工修复源 |
+
+identity collector 从 live Cobra leaves 的 `ContractFinal.Identity` 声明生成 `CommandSpec`，应用 reviewed exclusions 后按确定性规则索引为 effective registry；从 binder 开始，下游只看到一个稳定 identity/navigation 模型。reviewed `schema_command_registry/` 已退役，其历史角色（稳定 canonical identity/navigation 事实源）由 collector 承接。旧 wire 中的 `surface_hash` / `surface_tools` 字段仅为兼容名称，语义已经是 effective Registry hash/coverage，不构成第二事实源。
+
+### 4.1 flag / help / schema 同源（路径 A + 嵌入）
+
+完整决策、字段归属表、`HOM-*` 门禁规划与可选 MCP 透传准入条件见 [`flag-help-schema-homology.md`](flag-help-schema-homology.md)。
+
+摘要：
+
+- **同源面**：Contract → cobra flags ≡ `--help` Flags ≡ **嵌入注解后的** schema `parameters` / 关系约束；显式 `Risk` 经 `dws.schema.risk` overlay 进 Schema Safety。
+- **嵌入点**：`command.embedContractIntoSchema` 写入 `dws.schema.contract` / property / type / required；`AnnotateConstraints` 写入 constraints；Schema 组装（`runtimeToolSpecFromMetadata` / `ResolveSchemaBuild`）消费这些注解进入 typed `SchemaRegistry`（runtime assembly；非 `go:embed` catalog 交付）。
+- **硬规则**：CLI 表面事实 = **声明（Contract 数据字段）OR 人工标注**；禁止纯推断。非 CLI 表面字段（identity / selection / interface）必须有**评审源**。声明写法见同源文档 §1.2；标注见 §1.3；**`ToolSpec` 全字段权威见 RFC §5.0.4 / 同源 §1.4**。
+- **非同源面（有意）**：identity（collector）、selection 文案（ProductDecl / leaf `Contract.Selection`）、RPC 形状（MCP meta 仅 `interface_*`）、dry-run 正能力 registry。
+- **禁止**：以 MCP meta 为主通道生成 Leaf/Shortcut 的 flag；已退役的 hint overlay 改写 type/required/default；Schema 字段无权威归属。
+
+## 5. 统一解析与 precedence
+
+### 5.1 Identity
+
+- identity collector（`CollectIdentitySpecs`）是 stable canonical identity、primary path、alias 和 navigation 的唯一基础事实源：每个携带 `ContractFinal.Identity` 的 runnable Cobra leaf（含 Hidden deprecated/migration shims）贡献一条 identity，reviewed exclusions 精确应用；reviewed `schema_command_registry/` 已退役，其历史角色由 collector 承接。
+- Collector 输出的 `CommandSpec` 在索引时 fail-closed 校验 canonical/product/path/alias/visibility 的合法性与唯一性；重复 identity、alias 复用 primary path 或 alias collision 全部失败，不能按 precedence 静默覆盖。
+- Binder 必须把 effective entry 的 primary path 和每个 alias 精确解析到同一个真实 executable leaf；stale path、phantom path、重复 identity 或 alias collision 全部失败。
+- Native identity annotation 是可选的一致性证据：存在时必须与 effective entry 精确一致；缺失不触发补写、推断或 fallback。
+- Public runnable Cobra leaf 未进入 effective registry 时，必须存在 exact、reviewed、带 reason 的 exclusion；不得用 prefix/wildcard 排除。
+- Identity 不做名称推断，不从 Catalog/generated metadata fallback，也没有多来源 winner。
+
+删除 native materialization 前已做写入审计：旧
+`ApplyNativeRuntimeSchemaContracts` 的唯一写操作是对已存在命令调用
+`AttachRuntimeSchema`，只写 command identity 的 product/tool/source annotation；
+它不写 flag property/type/required、constraints、positionals、title/description
+或 interface mapping。这些字段原本已分别由 parameter binding/metadata、
+constraint、Cobra help 和 interface resolver 提供，因此删除该过渡层没有数据迁移缺口。
+CI 同时禁止重新加入 generated native contracts 或 materialization 入口。
+
+#### Identity 输入审计（registry 已退役）
+
+历史上 identity 来自 reviewed `schema_command_registry/`（`registry.json` +
+`products/*.json`，由随附 JSON Schema 校验）。该 reviewed registry 已退役，
+改由 identity collector 承接：identity 现在由 `CollectIdentitySpecs` 从 live
+Cobra 树的 `ContractFinal.Identity` 声明收集，输入契约即声明本身。严格 Go
+索引（`indexCommandSpecs`）继续 fail-closed 校验：
+
+- canonical identity、`source_product_id` 和精确 CLI path 的格式；
+- `aliases` 唯一且不能复用 primary path；
+- `visibility` 只允许 `public | compat | internal`，省略时明确归一化为
+  `public`；
+- primary path、alias、canonical 和 product 之间的交叉唯一性约束。
+
+Registry semantic hash 仍覆盖 canonical、primary CLI path、alias 集合、
+`source_product_id` 和 normalized visibility。格式、顺序以及省略的等价默认值
+不改变 hash；上述任一稳定契约字段变化都必须改变 hash。测试逐字段验证这一点，
+不使用当前命令数量作为常量。
+
+普通 `go generate ./internal/cli` 只生成
+`param_aliases_generated.go`；production Catalog / `ResolveMeta` 由 runtime
+`ResolveSchemaBuild` 装配（`deliverySchemaCatalog` Once 后缓存 Meta 投影）。
+`cmd_schema_catalog` 仅按需打 CI/local dump；其 `InstallBuildTimeAgentMetadataJSON`
+inject 仅服务 dump，生产 Agent 权威仍是 leaf `ContractFinal` / `ProductDecl`。
+不写也不 embed `schema_agent_metadata/`。drift policy 禁止已退役的
+`schema_command_registry/` 在生成前后重新出现；原独立脚本
+`check-schema-command-registry.sh` 的 registry-agnostic side guards（禁用旧
+native materialization 符号、go:generate 单轨、lazy loader 纪律）已迁入
+`check-schema-catalog.sh`。
+
+### 5.2 Parameter
+
+每个字段按明确的来源 precedence 选择一次，并把 winner、候选值和来源写入 provenance。precedence **与值无关**：不能因为 `required=true` 看起来更严格就让它越级获胜。更高优先级的 reviewed manual override 可以把 `required`、映射、interface type 或描述调高，也可以调低。
+
+实现中的参数字段顺序固定为：
+
+```text
+versioned binding > command constraint > typed metadata
+                > native/Cobra contract (ParamDecl / ContractFinal)
+                > MCP metadata > inference/default
+```
+
+命令 `title` / `description` 使用独立但同样确定的文本顺序：
+
+```text
+description: Cobra Long > ContractDecl description (contract_final) > MCP metadata > inference
+             (Long 胜出时 provenance = cobra_help / cobra_help_preferred)
+title:       ContractDecl / ContractFinal > Cobra Short > MCP metadata > inference
+```
+
+因此多个 CLI leaf 复用同一个 RPC 时，通用 RPC 文案只能作为未选中的
+provenance candidate 保留；参数级 RPC 文案可进入 `interface_description`，
+但不得覆盖 leaf 自己的标题和执行语义。
+
+Cobra hard-required 是独立的 executable fact，并通过 `cli_required`/provenance 保留；它不应在 renderer 中再次静默改写已经解析的 Agent projection。
+
+### 5.3 Safety、selection 与 interface
+
+`effect`、`risk`、`confirmation`、`idempotency`、selection 和 interface disposition 同样按 source precedence 解析，而不是按值的“严格程度”合并。更高优先级的 reviewed explicit/manual source 可以升高或降低最终值；同 precedence 的不同值必须报冲突。
+
+最终 interface disposition 还必须满足 conflict matrix：
+
+- `mode` 与 `availability` 正交：`mode` 只允许 `mcp | local | composite`，`availability` 只允许 `available | unavailable`；`unavailable` 不是第四种 mode。
+- `mcp + available`：只表示命令可由一个 pinned、参数可映射且语义等价的 `interface_ref` 完整表达；本地 wrapper 只是固定默认值或投影返回值时，也必须先证明参数和执行语义没有漂移。
+- `local + available`：仅用于纯本地进程、静态数据或策略操作，不得携带 direct `interface_ref`；“远端 RPC 尚未进入 pinned metadata”不能归类为 local。
+- `composite + available`：用于多 RPC、条件路由、本地投影，或 reviewed unpinned remote adapter；不得用单个 `interface_ref` 冒充完整实现，且必须提供 reviewed reason。未来需要表达多个 RPC 时使用单独的复合接口模型。
+- 任意合法 mode + `unavailable`：不得携带 `interface_ref`，必须提供明确 reason，并且 Agent 不得把它当作可用接口。
+
+## 6. Schema、Help 与业务数据边界
+
+| 问题 | 事实源 |
+|---|---|
+| 当前二进制是否暴露命令、Cobra 接受哪些 flags | `dws <path> --help` |
+| Agent 选哪个命令、CLI 参数/required/约束、risk/confirmation | Agent leaf `dws schema "<path>" --compact` |
+| CLI↔RPC 参数映射、接口绑定、provenance | full leaf 配合 `--jq` / `--fields` 精确投影 |
+| 钉钉中的文档、文件、日程、消息等实际数据 | 真正执行 `dws doc read`、`dws drive search` 等 read/search/list 命令 |
+
+Schema 和 Help 冲突是契约漂移，不能静默猜测：
+
+- 执行参数以 Cobra 实际接受的 flags 为准；不要发送 Help 中不存在的 flag。
+- 安全语义冲突时不要采用更宽松值。先按更保守的解释确认；如果无法确定安全执行方式，停止并报告漂移。
+- Schema/Help 只完成命令发现和契约读取。需要业务结果时，必须继续执行真实 read/search/list 命令。
+
+上述运行时漂移策略不改变构建期的 value-neutral precedence；前者是在契约已经互相矛盾时保护用户，后者是在确定性生成同一契约。
+
+## 7. 查询投影
+
+```bash
+dws schema                                      # 产品紧凑概览
+dws schema calendar --compact                   # Agent 产品摘要
+dws schema "calendar event" --compact          # Agent 分组摘要
+dws schema "calendar event create" --compact   # Agent leaf
+dws schema "calendar event create"             # full leaf，仅用于映射/provenance 审计
+dws schema --all                               # 所有工具的完整 leaf 导出
+```
+
+`schema list` 是根概览的兼容入口。
+
+`schema --all` 必须包含最终 `SchemaIndex` 中每个 tool 的完整 leaf 参数、约束和安全语义；无业务参数的命令也要包含空 `parameters` 对象。它用于审计、CI 和参数防丢 baseline，但输出很大，普通 Agent 命令发现不得使用，应按 overview -> product/group -> leaf 渐进查询。
+
+`--compact` 是普通 Agent 查询的规范视图：通过正向字段白名单保留选参、约束与安全语义，full 新增字段不会自动进入 Agent 上下文。省略它的 leaf 包含参数 property、接口绑定和 provenance，只用于定向审计；`schema --all --compact` 也可执行，但不能作为完整兼容性 baseline。
+
+兼容旧二进制时，如果 Schema 查询返回 `unknown_flag: --compact`，只去掉 `--compact` 重试同一个查询。这是展示能力降级，不代表 leaf 缺失，也不能改用 Schema 查询业务数据。
+
+## 8. 生成与发布
+
+当 Cobra、flag、identity、binding、leaf `Contract` / ProductDecl 或 Skill 发生变化时：
+
+1. 审核真实 Cobra 变化，确认命令和 flag 已实际存在。新增或修改稳定 command identity、primary CLI path 或 alias 现在通过 leaf Contract 的 `ContractFinal.Identity` 声明完成（identity collector 据此收集；reviewed `schema_command_registry/` 已退役）。参数、Skill 或 metadata 单独变化时不要机械改写 identity 声明，也不要从旧 Catalog 反向生成它。
+2. 不应进入稳定 Agent 契约的 public runnable leaf 使用 reviewed exclusions（exact path、带 reason）。Native identity annotation 若存在，应作为与 identity 声明一致的实现断言维护，而不是用来 materialize identity。
+3. 生成参数别名并验证运行时 Schema 组装。`go generate ./internal/cli` 只运行 `cmd_param_aliases`；`cmd_schema_catalog` 仅按需生成 CI/local dump。生产权威为 leaf `ContractFinal` / `ProductDecl`；CI dump 可经 `agent_metadata_inject.go` / `InstallBuildTimeAgentMetadataJSON` 在内存中注入 Agent metadata，不写 `schema_agent_metadata/`：
+
+   ```bash
+   make generate-schema
+   go generate ./internal/cli
+   # 可选：生成 artifacts/ 下的 CI/local dump
+   make generate-schema-catalog
+   ```
+
+`cmd_schema_agent_metadata` 可保留为非交付工具/测试，但不是 `go:generate` 入口，也不应再作为发布步骤。
+
+生成文件只有 `internal/cli/param_aliases_generated.go`（参数别名生成物）。`cmd_schema_catalog` 的 Catalog 和 meta-index 是可选 CI/local dump，不是交付物，且不得写入或提交到 `internal/cli/`。
+
+`schema_agent_metadata/`、`schema_agent_metadata_audit.json` 与 `schema_hints/` 已退役；若存在则 policy 失败。只编辑来源；不要手工编辑或提交 Catalog / meta-index dump。
+
+## 9. Completeness 与 final-delivery invariant
+
+门禁必须验证最终交付对象，而不是某个中间层或数量：
+
+- 每个 public runnable Cobra leaf 要么能通过最终 embedded `SchemaIndex` 查询，要么有 exact、reviewed、带 reason 的 exclusion。
+- 每个最终 canonical path、primary CLI path 和 alias 都必须解析到同一个可执行 leaf；不得有 phantom path 或 collision。
+- `EffectiveCommandRegistry`、`SchemaRegistry/SchemaIndex` 与 Catalog canonical sets 必须精确一致（含组装时内存 inject 的 Agent metadata 语义），不能只比较 count。
+- Leaf payload、`--all` 中对应 tool 和 Catalog full tool 必须是同一个 resolved `ToolSpec` 的内容级等价投影，并通过 production loader round-trip。
+- overview/product/group summary 与 Catalog summary 必须等于同一个 `ToolSpec.ToSummaryPayload()`；alias 查询只允许 `cli_path` 和 `is_alias` 这两个视图字段变化。
+- 每个最终字段及 parameter field 的 provenance winner value 必须与 delivered value 精确一致；不能只验证 provenance source、count 或字段是否存在。
+- 每个 MCP `interface_ref` 必须在 pinned interface registry 精确存在；local/composite/unavailable 必须满足同一 conflict matrix。
+- `--all` 的 tool set 必须与最终 index 一对一，且每个工具包含完整参数契约。
+- 连续两次生成必须字节稳定，提交的生成物不得漂移。
+- **同源门禁（规划，见同源文档 §4）**：受管命令逐步满足 `HOM-P1`–`P3`（parameters ≡ cobra/Contract）、`HOM-S1`–`S3`（Safety/Risk 对齐）、`HOM-I1`（interface 不创建 flag）、`HOM-D1`（help ≡ schema parameters）。hints 不得作为 type/required/default 的 winner。
+
+推荐本地验证：
+
+```bash
+make generate-schema
+./scripts/policy/check-generated-drift.sh
+./scripts/policy/check-schema-catalog.sh
+go test ./internal/cli ./internal/app ./internal/generator/... -count=1
+```
+
+## 10. 明确禁止
+
+- 运行时调用 MCP `tools/list` 或访问网络生成 Schema。
+- 从 `schema_catalog/` 等生成 JSON 反向创建/补齐 Cobra leaf、flag、CommandRegistry 或下一轮 Catalog。
+- 重新引入 `schema_agent_metadata/`（或 audit JSON）作为交付物、`go:embed` 目标，或把它写回 `go:generate` 入口。
+- 从 MCP meta / 已退役的 `schema_mcp_metadata` **生成或补齐** LeafSpec/Shortcut 主路径的 CLI flag（interface overlay 除外）；未满足同源文档 §5 准入条件时启用「MCP 透传生成通道」。
+- 把 native annotation、legacy registry 或 Catalog 当作 identity fallback；或在 `EffectiveCommandRegistry` 之后再次选择 identity winner。
+- renderer、query 或 gate 在 `SchemaRegistry` 之后重新读取 source 并做第二次 merge。
+- 用 prefix/wildcard exclusion 隐藏未来命令。
+- 让 ProductDecl / leaf `Contract`、CommandRegistry 或 interface metadata 宣称一个不存在的命令、flag 或 RPC 可用。
+- 重新引入 `schema_hints/`（含 selection/metadata/imported/audit JSON）或任何 HintFile overlay，并在与 Contract/cobra 冲突时赢得 type/required/default。
+- 把 `schema --all` 当作普通业务数据查询，或把其完整结果无条件注入 Agent 上下文。
+- 将 LeafSpec、`+shortcut` 或 write-guard/cursor/多步命令注册为 `mcp_passthrough` 表面。

docs/shortcut-backend-mcp-issues.md

@@ -0,0 +1,174 @@
+# Shortcut 真实测试：后端 / MCP 问题整理
+
+这份报告只汇总 `failure_category = backend-or-mcp-error` 的 case，已尽量排除权限、缺真实资源、当前账号无数据等噪音。
+
+## 总览
+
+- Backend/MCP case 总数：33
+- 聚合问题数：8
+- 复现口径：真实 dws CLI；无 mock；无 dry-run；命令输入和 trace_id 均来自真实测试结果。
+
+## 建议优先看
+
+1. [P1] Chat/IM 会话 ID 字段在 MCP/后端映射中疑似丢失（15 case）
+2. [P1] Chat card 发送 receiverUid 疑似未从 receiver 透传（1 case）
+3. [P1] Chat 入群审批 applicantUid/inviterUid 疑似未透传（1 case）
+4. [P1] AI 表格 MCP 错误 envelope 语义不一致：success=true 但 error 非空/status=error（5 case）
+5. [P1] AI 表格 Workflow 查询在真实 Base 下返回系统级错误（2 case）
+6. [P1] AI 表格 roleId 参数疑似未被 MCP 正确读取（3 case）
+7. [P2] AI 表格记录主文档查询在真实 record 下返回 no record/SYSTEM_ERROR（2 case）
+8. [P2] AI 表格无效 Base/Table/Field/Record 被包装成 SYSTEM_ERROR（4 case）
+
+## Chat/IM 会话 ID 字段在 MCP/后端映射中疑似丢失
+
+- 优先级：P1
+- 建议 owner：IM MCP / IM 后端字段映射
+- 现象：CLI 已传 group/conversation-id/open-conversation-id（部分 case 使用真实 cid），后端仍报 openCid/openConversationId/cid required。
+- 期望：MCP schema/网关应接受并透传 openConversationId/openCid/cid 中的兼容字段；如果资源无效，应返回“无效会话”，而不是 required。
+- 涉及 case：15
+
+| 套件 | shortcut | operation | trace_id | 证据 |
+|---|---|---|---|---|
+| `read` | `chat +chat-members-get` | `tools/call` | `2127d89817840997754345760e07bd` | [UNCLASSIFIED] openCid or cid is required (operation: im/list_group_member_by_ids) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +chat-members-get --id DWSREALREADNOSUCHID0000000000000 --users '冬翔' --yes --format json` |
+| `read` | `chat +chat-messages` | `tools/call` | `2104a64c17840997767792656e085e` | [UNCLASSIFIED] openCid or cid is required hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +chat-messages --group cid3Jijzhe2aqs9ysOXjhi05g== --time '2026-07-15 10:00:00' --limit 10 --direction older --yes --format json` |
+| `read` | `chat +messages-list` | `tools/call` | `2127d89817840997797873841e0757` | [UNCLASSIFIED] openCid or cid is required hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +messages-list --group cid3Jijzhe2aqs9ysOXjhi05g== --time '2026-07-15 10:00:00' --forward --limit 10 --yes --format json` |
+| `write` | `chat +chat-mute-member` | `tools/call` | `2104a64c17840999166036583e08a3` | [UNCLASSIFIED] openConversationId or cid is required (operation: im/set_group_member_mute_list) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +chat-mute-member --group cidDWSREALTESTNOSUCHCONV --users __DWS_SHORTCUT_REAL_TEST_NO_SUCH_USER__ --mute-time 1 --off --yes --format json` |
+| `write` | `chat +chat-transfer-owner` | `tools/call` | `0b5deb3217840999222318863e087a` | [UNCLASSIFIED] openConversationId is required (operation: im/transfer_group_owner) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +chat-transfer-owner --group cidDWSREALTESTNOSUCHCONV --new-owner __DWS_SHORTCUT_REAL_TEST_NO_SUCH_USER__ --yes --format json` |
+| `write` | `chat +conversation-clear-messages` | `tools/call` | `2127d89817840999254816721e079b` | [UNCLASSIFIED] openConversationId or cid is required (operation: im/clear_conversation_messages) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +conversation-clear-messages --conversation-id cidDWSREALTESTNOSUCHCONV --yes --format json` |
+| `write` | `chat +conversation-clear-red-point` | `tools/call` | `2104a64c17840999265511295e085f` | [UNCLASSIFIED] openConversationId or cid is required (operation: im/clear_conversation_red_point) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +conversation-clear-red-point --conversation-id cidDWSREALTESTNOSUCHCONV --yes --format json` |
+| `write` | `chat +conversation-hide` | `tools/call` | `2127d89817840999276117140e079b` | [UNCLASSIFIED] openConversationId or cid is required (operation: im/hide_conversation) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +conversation-hide --conversation-id cidDWSREALTESTNOSUCHCONV --yes --format json` |
+| `write` | `chat +conversation-mark-unread` | `tools/call` | `0bb7c36217840999298744910e0758` | [UNCLASSIFIED] openConversationId or cid is required (operation: im/mark_conversation_unread) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +conversation-mark-unread --conversation-id cidDWSREALTESTNOSUCHCONV --yes --format json` |
+| `write` | `chat +conversation-mute` | `tools/call` | `2104a64c17840999309241865e085f` | [UNCLASSIFIED] openConversationId is required (operation: im/update_notification_off) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +conversation-mute --conversation-id cidDWSREALTESTNOSUCHCONV --off --yes --format json` |
+| `write` | `chat +conversation-mute-at-all` | `tools/call` | `2127d89817840999320028068e07dd` | [UNCLASSIFIED] openConversationId is required (operation: im/update_at_all_notification_off) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +conversation-mute-at-all --conversation-id cidDWSREALTESTNOSUCHCONV --off --yes --format json` |
+| `write` | `chat +conversation-mute-red-envelope` | `tools/call` | `0bb7c36217840999330228283e07fe` | [UNCLASSIFIED] openConversationId is required (operation: im/update_red_env_notification_off) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +conversation-mute-red-envelope --conversation-id cidDWSREALTESTNOSUCHCONV --off --yes --format json` |
+| `write` | `chat +conversation-set-top` | `tools/call` | `2127d89817840999341302961e07fe` | [UNCLASSIFIED] openConversationId or cid is required (operation: im/set_top_conversation) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +conversation-set-top --conversation-id cidDWSREALTESTNOSUCHCONV --off --yes --format json` |
+| `write` | `chat +messages-set-pin` | `tools/call` | `0bb7c36217840999521117991e0758` | [UNCLASSIFIED] openConversationId or cid is required (operation: im/set_pin_message) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +messages-set-pin --open-conversation-id cidDWSREALTESTNOSUCHCONV --msg-id DWSREALTESTNOSUCHID0000000000000 --yes --format json` |
+| `write` | `chat +messages-unset-pin` | `tools/call` | `2104a64c17840999544428103e08ee` | [UNCLASSIFIED] openConversationId or cid is required (operation: im/unset_pin_message) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +messages-unset-pin --open-conversation-id cidDWSREALTESTNOSUCHCONV --msg-id DWSREALTESTNOSUCHID0000000000000 --yes --format json` |
+
+## Chat card 发送 receiverUid 疑似未从 receiver 透传
+
+- 优先级：P1
+- 建议 owner：IM MCP / card 发送参数映射
+- 现象：CLI 传入 receiver=103262，后端仍报 receiverUid 和 openConversationId 不能同时为空。
+- 期望：receiver 应映射为 receiverUid，或 schema 明确要求 receiverUid；真实入参不应在 MCP 层丢失。
+- 涉及 case：1
+
+| 套件 | shortcut | operation | trace_id | 证据 |
+|---|---|---|---|---|
+| `write` | `chat +messages-send-card` | `tools/call` | `2104a64c17840999509255753e081a` | [UNCLASSIFIED] receiverUid和openConversationId不能同时为空 (operation: im/create_and_send_card) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +messages-send-card --receiver 103262 --yes --format json` |
+
+## Chat 入群审批 applicantUid/inviterUid 疑似未透传
+
+- 优先级：P1
+- 建议 owner：IM MCP / 入群审批参数映射
+- 现象：CLI 传入 applicant=103262、inviter=519019，后端仍报 applicantUid required。
+- 期望：applicant/inviter 应映射为 applicantUid/inviterUid；如果 recordId/group 无效，应返回对应资源错误而不是 applicantUid 缺失。
+- 涉及 case：1
+
+| 套件 | shortcut | operation | trace_id | 证据 |
+|---|---|---|---|---|
+| `write` | `chat +chat-audit-join` | `tools/call` | `0bb7c36217840999154832733e0758` | [UNCLASSIFIED] applicantUid is required (operation: im/audit_join_group) hint: Use --verbose for detailed error logs |
+| | input | | | `/private/tmp/dws-real-test chat +chat-audit-join --group cidDWSREALTESTNOSUCHCONV --record-id 999999999999 --applicant 103262 --inviter 519019 --status AuditApprove --description 'DWS shortcut 真实测试描述，可删除' --yes --format json` |
+
+## AI 表格 MCP 错误 envelope 语义不一致：success=true 但 error 非空/status=error
+
+- 优先级：P1
+- 建议 owner：AI 表格 MCP wrapper
+- 现象：多条 AI 表格命令返回 MCP_TOOL_ERROR，内部 JSON 同时出现 success=true、status=error、error 非空。
+- 期望：只要 error 非空或 status=error，success 应为 false，外层也应按业务错误返回稳定错误码/trace。
+- 涉及 case：5
+
+| 套件 | shortcut | operation | trace_id | 证据 |
+|---|---|---|---|---|
+| `read` | `aitable +export-data` | `-` | `2104a64c17840997514714448e0817` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"INVALID_PARAMS","message":"taskId cannot be combined with scope, format, tableId or viewId","retryable":false,"type":"INPUT_ERROR"},"m… |
+| | input | | | `/private/tmp/dws-real-test aitable +export-data --base-id gpG2NdyVXQyZ0OmoSbd1vbA6JMwvDqPk --task-id DWSREALREADNOSUCHID0000000000000 --scope all --format excel --table-id hERWDMS --view-id qvGDAH2 --timeout-ms 1 --yes` |
+| `write` | `aitable +chart-update` | `-` | `2106d98117840998553244877e08df` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"INVALID_PARAMS","message":"config is required","retryable":false,"type":"INPUT_ERROR"},"meta":{},"status":"error","success":true,"summ… |
+| | input | | | `/private/tmp/dws-real-test aitable +chart-update --base-id DWSREALTESTNOSUCHID0000000000000 --dashboard-id DWSREALTESTNOSUCHID0000000000000 --chart-id DWSREALTESTNOSUCHID0000000000000 --config '{}' --layout '{}' --yes --format json` |
+| `write` | `aitable +record-update` | `-` | `0bab027317840998747383236e090b` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"INVALID_RECORDS","message":"records must contain at least one writable record","retryable":false,"type":"INPUT_ERROR"},"meta":{},"stat… |
+| | input | | | `/private/tmp/dws-real-test aitable +record-update --base-id DWSREALTESTNOSUCHID0000000000000 --table-id DWSREALTESTNOSUCHID0000000000000 --records '[{"recordId":"recDWSREALTEST","cells":{}}]' --yes --format json` |
+| `write` | `aitable +record-upsert` | `-` | `2106d98117840998759832182e087b` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"INVALID_PARAMETER","message":"records is required and must not be empty","retryable":false,"type":"INPUT_ERROR"},"meta":{},"status":"e… |
+| | input | | | `/private/tmp/dws-real-test aitable +record-upsert --base-id DWSREALTESTNOSUCHID0000000000000 --table-id DWSREALTESTNOSUCHID0000000000000 --records '[]' --yes --format json` |
+| `write` | `aitable +view-set-fill-color-rule` | `-` | `2106d98117840998924181709e08df` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"INVALID_PARAMS","message":"conditionalFormats is required","retryable":false,"type":"INPUT_ERROR"},"meta":{},"status":"error","success… |
+| | input | | | `/private/tmp/dws-real-test aitable +view-set-fill-color-rule --base-id DWSREALTESTNOSUCHID0000000000000 --table-id DWSREALTESTNOSUCHID0000000000000 --view-id DWSREALTESTNOSUCHID0000000000000 --json '{}' --yes --format json` |
+
+## AI 表格 Workflow 查询在真实 Base 下返回系统级错误
+
+- 优先级：P1
+- 建议 owner：AI 表格 Workflow MCP / 后端
+- 现象：使用真实可访问 Base 查询 workflow list/get，返回 LIST_WORKFLOWS_ERROR/GET_WORKFLOW_ERROR。
+- 期望：无 workflow 时应返回空列表或 WORKFLOW_NOT_FOUND；有后端异常时需提供稳定错误码和可排查 trace。
+- 涉及 case：2
+
+| 套件 | shortcut | operation | trace_id | 证据 |
+|---|---|---|---|---|
+| `read` | `aitable +workflow-get` | `-` | `2104a64c17840997556363676e08ee` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"GET_WORKFLOW_ERROR","message":"调用远程服务业务异常","retryable":true,"type":"SYSTEM_ERROR"},"meta":{},"status":"error","success":true,"summary"… |
+| | input | | | `/private/tmp/dws-real-test aitable +workflow-get --base-id gpG2NdyVXQyZ0OmoSbd1vbA6JMwvDqPk --workflow-id DWSREALREADNOSUCHID0000000000000 --yes --format json` |
+| `read` | `aitable +workflow-list` | `-` | `2127d89817840997572427879e075d` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"LIST_WORKFLOWS_ERROR","message":"biz error","retryable":true,"type":"SYSTEM_ERROR"},"meta":{},"status":"error","success":true,"summary… |
+| | input | | | `/private/tmp/dws-real-test aitable +workflow-list --base-id gpG2NdyVXQyZ0OmoSbd1vbA6JMwvDqPk --limit 10 --offset 1 --yes --format json` |
+
+## AI 表格 roleId 参数疑似未被 MCP 正确读取
+
+- 优先级：P1
+- 建议 owner：AI 表格 MCP role 接口
+- 现象：CLI 已传 --role-id，但 MCP 返回 roleId is required。
+- 期望：role-id/roleId 字段应被正确映射；如果 role 不存在，返回 ROLE_NOT_FOUND，而不是 required。
+- 涉及 case：3
+
+| 套件 | shortcut | operation | trace_id | 证据 |
+|---|---|---|---|---|
+| `read` | `aitable +role-get` | `-` | `2104a64c17840997542904838e0817` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"INVALID_PARAMS","message":"roleId is required","retryable":false,"type":"INPUT_ERROR"},"meta":{},"status":"error","success":true,"summ… |
+| | input | | | `/private/tmp/dws-real-test aitable +role-get --base-id gpG2NdyVXQyZ0OmoSbd1vbA6JMwvDqPk --role-id x --yes --format json` |
+| `write` | `aitable +role-delete` | `-` | `2106d98117840998782482701e089c` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"INVALID_PARAMS","message":"roleId is required","retryable":false,"type":"INPUT_ERROR"},"meta":{},"status":"error","success":true,"summ… |
+| | input | | | `/private/tmp/dws-real-test aitable +role-delete --base-id DWSREALTESTNOSUCHID0000000000000 --role-id DWSREALTESTNOSUCHID0000000000000 --yes --format json` |
+| `write` | `aitable +role-update` | `-` | `2132f5ca17840998794483634e08d8` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"INVALID_PARAMS","message":"roleId is required","retryable":false,"type":"INPUT_ERROR"},"meta":{},"status":"error","success":true,"summ… |
+| | input | | | `/private/tmp/dws-real-test aitable +role-update --base-id DWSREALTESTNOSUCHID0000000000000 --role-id DWSREALTESTNOSUCHID0000000000000 --name 'DWS shortcut 真实测试 20260715-151724' --role-type x --flow-type x --sub-roles '[]' --yes --format json` |
+
+## AI 表格记录主文档查询在真实 record 下返回 no record/SYSTEM_ERROR
+
+- 优先级：P2
+- 建议 owner：AI 表格 primary doc MCP / 后端
+- 现象：record-query 已能查到真实 recordId，但 primary-doc 查询返回 no record、type=SYSTEM_ERROR。
+- 期望：若该记录无主文档，应返回空/未创建；若 recordId 语义不匹配，应返回明确参数错误，不应是系统错误。
+- 涉及 case：2
+
+| 套件 | shortcut | operation | trace_id | 证据 |
+|---|---|---|---|---|
+| `read` | `aitable +base-get-primary-doc-id` | `-` | `0b5deb3217840997466255627e08ee` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"-1","message":"no record","retryable":true,"type":"SYSTEM_ERROR"},"meta":{},"status":"error","success":true,"summary":"Failed to query… |
+| | input | | | `/private/tmp/dws-real-test aitable +base-get-primary-doc-id --base-id gpG2NdyVXQyZ0OmoSbd1vbA6JMwvDqPk --table-id hERWDMS --record-id 1015oH3OXy --yes --format json` |
+| `read` | `aitable +record-primary-doc-get` | `-` | `2127d89817840997528393730e079c` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"-1","message":"no record","retryable":true,"type":"SYSTEM_ERROR"},"meta":{},"status":"error","success":true,"summary":"Failed to query… |
+| | input | | | `/private/tmp/dws-real-test aitable +record-primary-doc-get --base-id gpG2NdyVXQyZ0OmoSbd1vbA6JMwvDqPk --table-id hERWDMS --record-id 1015oH3OXy --yes --format json` |
+
+## AI 表格无效 Base/Table/Field/Record 被包装成 SYSTEM_ERROR
+
+- 优先级：P2
+- 建议 owner：AI 表格 MCP wrapper / 后端错误码
+- 现象：安全负向 ID 下，部分写接口返回 getDentryDTO returns null、type=SYSTEM_ERROR、retryable=true。
+- 期望：资源不存在应返回 INPUT_ERROR/NOT_FOUND 且 retryable=false，避免误导调用方重试。
+- 涉及 case：4
+
+| 套件 | shortcut | operation | trace_id | 证据 |
+|---|---|---|---|---|
+| `write` | `aitable +field-delete` | `-` | `0bab027317840998611338768e08c8` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"404","message":"getDentryDTO returns null","retryable":true,"type":"SYSTEM_ERROR"},"meta":{},"status":"error","success":true,"summary"… |
+| | input | | | `/private/tmp/dws-real-test aitable +field-delete --base-id DWSREALTESTNOSUCHID0000000000000 --table-id DWSREALTESTNOSUCHID0000000000000 --field-id DWSREALTESTNOSUCHID0000000000000 --yes --format json` |
+| `write` | `aitable +field-update` | `-` | `213ee25c17840998623207342e08e2` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"404","message":"getDentryDTO returns null","retryable":true,"type":"SYSTEM_ERROR"},"meta":{},"status":"error","success":true,"summary"… |
+| | input | | | `/private/tmp/dws-real-test aitable +field-update --base-id DWSREALTESTNOSUCHID0000000000000 --table-id DWSREALTESTNOSUCHID0000000000000 --field-id DWSREALTESTNOSUCHID0000000000000 --name 'DWS shortcut 真实测试 20260715-151724' --config '{}' --ai-config '{}' --yes --format json` |
+| `write` | `aitable +record-delete` | `-` | `2132f5ca17840998724753149e0853` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"404","message":"getDentryDTO returns null","retryable":true,"type":"SYSTEM_ERROR"},"meta":{},"status":"error","success":true,"summary"… |
+| | input | | | `/private/tmp/dws-real-test aitable +record-delete --base-id DWSREALTESTNOSUCHID0000000000000 --table-id DWSREALTESTNOSUCHID0000000000000 --record-ids DWSREALTESTNOSUCHID0000000000000 --yes --format json` |
+| `write` | `aitable +table-update` | `-` | `213ee25c17840998877246229e087f` | [MCP_TOOL_ERROR] {"data":{},"error":{"code":"404","message":"getDentryDTO returns null","retryable":true,"type":"SYSTEM_ERROR"},"meta":{},"status":"error","success":true,"summary"… |
+| | input | | | `/private/tmp/dws-real-test aitable +table-update --base-id DWSREALTESTNOSUCHID0000000000000 --table-id DWSREALTESTNOSUCHID0000000000000 --name 'DWS shortcut 真实测试 20260715-151724' --description 'DWS shortcut 真实测试描述，可删除' --record-name-key task --yes --format json` |

docs/shortcut-error-review.html

@@ -0,0 +1,279 @@
+<!doctype html>
+<html lang="zh-CN">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>Shortcut 真实测试失败逐项 review</title>
+<style>
+:root{--bg:#0f1420;--card:#151d2b;--line:#263246;--text:#dce7f7;--muted:#91a0b5;--blue:#8fd3ff;--green:#66d38a;--yellow:#e2b23c;--red:#f27272;--purple:#d3a7ff}
+*{box-sizing:border-box}
+body{margin:0;background:var(--bg);color:var(--text);font:13px/1.55 -apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,"Helvetica Neue",Arial,"PingFang SC","Microsoft YaHei",sans-serif}
+header{padding:28px 32px 14px;border-bottom:1px solid var(--line);background:linear-gradient(180deg,#172033,#0f1420)}
+h1{margin:0 0 8px;font-size:26px}
+h2{margin:28px 0 10px;font-size:18px}
+.sub,.note,.count{color:var(--muted)}
+.wrap{padding:18px 32px 40px;max-width:1800px;margin:0 auto}
+.note{background:var(--card);border:1px solid var(--line);border-radius:10px;padding:12px 14px;margin:10px 0 18px}
+.stats{display:grid;grid-template-columns:repeat(auto-fit,minmax(150px,1fr));gap:12px;margin:18px 0}
+.stat{background:var(--card);border:1px solid var(--line);border-radius:12px;padding:14px}
+.stat .n{font-size:24px;color:var(--blue);font-weight:700}
+.stat .l{color:var(--muted);font-size:12px}
+.summary-grid{display:grid;grid-template-columns:repeat(auto-fit,minmax(320px,1fr));gap:18px;margin:16px 0 22px}
+table{width:100%;border-collapse:collapse;background:var(--card);border:1px solid var(--line);border-radius:10px;overflow:hidden}
+th,td{padding:8px 10px;border-bottom:1px solid var(--line);vertical-align:top;text-align:left}
+th{background:#1b2536;color:var(--muted);font-size:12px;font-weight:600;position:sticky;top:0;z-index:1}
+tr:last-child td{border-bottom:none}
+code{font-family:"SF Mono",Menlo,Consolas,monospace;color:#c7cfdb;font-size:12px}
+.review{table-layout:fixed}
+.review th:nth-child(1),.review td:nth-child(1){width:44px}
+.review th:nth-child(2),.review td:nth-child(2){width:165px}
+.review th:nth-child(3),.review td:nth-child(3){width:70px}
+.review th:nth-child(4),.review td:nth-child(4){width:120px}
+.review th:nth-child(5),.review td:nth-child(5){width:130px}
+.review th:nth-child(8),.review td:nth-child(8){width:130px}
+.review th:nth-child(9),.review td:nth-child(9){width:180px}
+.review td{word-break:break-word}
+.num{color:var(--muted);text-align:right}
+.risk{color:var(--green)}
+.cat{color:var(--yellow)}
+.owner{color:var(--purple)}
+.evidence{color:#c7cfdb;font-size:12px}
+a{color:var(--blue)}
+</style>
+</head>
+<body>
+<header>
+<h1>Shortcut 真实测试失败逐项 review</h1>
+<div class="sub">由 <code>scripts/gen_shortcut_error_review.py</code> 从真实测试结果生成；目标是把每个失败项落到“应该改哪里”。</div>
+</header>
+<main class="wrap">
+<div class="note">
+Read：204 条，成功 162，失败 41，超时 0。
+Write：162 条，成功 48，失败 114，超时 0。
+判定口径：如果 fake MCP 已看到字段但真实后端仍报 required，按后端/MCP schema 映射处理；如果真实后端报资源无效/不存在，按 fixture 处理；权限类不在 CLI 中绕过。
+</div>
+
+<div class="stats"><div class="stat"><div class="n">41</div><div class="l">Read 失败</div></div>
+<div class="stat"><div class="n">114</div><div class="l">Write 失败</div></div>
+<div class="stat"><div class="n">156</div><div class="l">逐项 review</div></div>
+<div class="stat"><div class="n">72</div><div class="l">测试数据/fixture</div></div>
+<div class="stat"><div class="n">22</div><div class="l">权限/应用配置</div></div>
+<div class="stat"><div class="n">33</div><div class="l">后端/MCP schema</div></div></div>
+<div class="summary-grid">
+<section><h2>按错误类型</h2><table><thead><tr><th>类型</th><th>数量</th></tr></thead><tbody><tr><td>输入/业务校验</td><td>36</td></tr>
+<tr><td>后端/MCP</td><td>33</td></tr>
+<tr><td>缺 AI 表格 fixture</td><td>31</td></tr>
+<tr><td>缺真实资源</td><td>30</td></tr>
+<tr><td>鉴权/权限</td><td>22</td></tr>
+<tr><td>缺妙记 fixture</td><td>3</td></tr>
+<tr><td>敏感/高风险暂缓</td><td>1</td></tr></tbody></table></section>
+<section><h2>按要改哪里</h2><table><thead><tr><th>要改哪里</th><th>数量</th></tr></thead><tbody><tr><td>测试数据</td><td>72</td></tr>
+<tr><td>后端/MCP schema</td><td>33</td></tr>
+<tr><td>权限/应用配置</td><td>22</td></tr>
+<tr><td>测试输入/业务校验</td><td>13</td></tr>
+<tr><td>后端业务/测试 fixture</td><td>11</td></tr>
+<tr><td>测试输入</td><td>2</td></tr>
+<tr><td>人工安全确认</td><td>1</td></tr>
+<tr><td>测试输入/shortcut 枚举</td><td>1</td></tr>
+<tr><td>测试输入/业务规则</td><td>1</td></tr></tbody></table></section>
+</div>
+
+
+<h2>缺真实资源复盘 <span class="count">· 可自造/可查资源处理结果</span></h2>
+<table>
+<thead><tr><th>命令/范围</th><th>处理状态</th><th>本次实际排查/造数结果</th><th>后续建议</th></tr></thead>
+<tbody><tr><td><code>chat +group-members</code></td><td><span class="cat">已补齐</span></td><td>查到真实群名 `浅曦-kida,Dennis,秋画`，runner 已改为用群名而不是 openConversationId；真实回归成功。</td><td>无需后续动作。</td></tr>
+<tr><td><code>chat +messages-mget</code></td><td><span class="cat">已补齐</span></td><td>复用真实单聊消息 `msgEuOor1PmFBNlx9M06N9z1Q==`；真实回归成功。</td><td>无需后续动作。</td></tr>
+<tr><td><code>chat +messages-read-status</code></td><td><span class="cat">已补齐</span></td><td>复用真实单聊会话 `cidie1367hAfBxqipzE59k5sknHLrHmvYkw98NADhfnjPI=` 与同一 openMessageId；真实回归成功。</td><td>无需后续动作。</td></tr>
+<tr><td><code>chat +messages-query-send-status</code></td><td><span class="cat">已补齐</span></td><td>复用真实发送返回的 openTaskId；真实回归成功。</td><td>无需后续动作。</td></tr>
+<tr><td><code>ding +receiver-status</code></td><td><span class="cat">已补齐</span></td><td>先只读 `ding +list` 找到已有 openDingId，再查询 receiver status；没有新发 DING，真实回归成功。</td><td>无需后续动作。</td></tr>
+<tr><td><code>sheet +list-sheets</code></td><td><span class="cat">已自造</span></td><td>创建临时在线表格 `DWS shortcut 真实测试表格 20260715`，nodeId=`mweZ92PV6O36dZbnsMZx70ylJxEKBD6p`；真实回归成功。</td><td>后续可保留为稳定 fixture，或测试结束后人工清理。</td></tr>
+<tr><td><code>todo +todo-done</code></td><td><span class="cat">已自造并修复 CLI</span></td><td>runner 会先创建当前账号自己的临时待办，再执行 `todo +todo-done`；同时修复了 todo 列表 pageSize=50 返回空、响应多层 result unwrap 不稳的问题；真实回归成功。</td><td>无需后续动作；代码已有单测覆盖 nested result。</td></tr>
+<tr><td><code>contact +by-mobile</code></td><td><span class="cat">已按用户授权补齐</span></td><td>使用用户指定手机号 `13161187007` 作为真实 fixture；runner 只在该命令上替换 mobile，不扩散到其它服务。</td><td>真实回归成功后该项将从失败列表移除；若后续要脱敏公开报告，可再加展示层脱敏。</td></tr>
+<tr><td><code>attendance +get-class / +get-group / +get-group-filtered</code></td><td><span class="cat">不建议自造</span></td><td>`attendance +search-class` 与 `+search-group --type FIXED` 均返回空；创建班次/考勤组会改组织考勤配置，属于高影响业务数据。</td><td>需要考勤后端/业务同学提供可读测试班次与考勤组 ID。</td></tr>
+<tr><td><code>chat +chat-get-by-id</code></td><td><span class="cat">暂未找到</span></td><td>该 shortcut 只接受数字 groupId；真实群列表只返回 openConversationId，没有数字群号字段。</td><td>需要 IM 后端提供可用数字 groupId，或评估是否新增 openConversationId 形态的 shortcut。</td></tr>
+<tr><td><code>chat +messages-resource-url</code></td><td><span class="cat">暂未自造</span></td><td>需要真实含 mediaId 的图片/文件/视频消息；当前文本消息无法产生 resource-id。</td><td>可在测试群发一条图片/文件消息并提取 mediaId 后补 fixture；注意会产生群消息。</td></tr>
+<tr><td><code>chat +thread-replies</code></td><td><span class="cat">暂未自造</span></td><td>需要真实话题消息 topicId；普通群消息不能替代。</td><td>需要话题群 fixture，或由 IM 同学提供当前账号可访问 topicId。</td></tr>
+<tr><td><code>aitable +dashboard-share-get</code></td><td><span class="cat">真实资源仍失败</span></td><td>已有真实 base/dashboard，但 share-get 返回 404 `Failed to get dashboard share config`；更像分享配置未开启或后端接口行为问题。</td><td>需要 AI 表格/后端确认如何创建/开启 dashboard share fixture，或修正 404 语义。</td></tr>
+<tr><td><code>write 类 delete/recall/approve/wiki move/copy 等</code></td><td><span class="cat">不自动自造</span></td><td>这些命令即便能造资源，也会涉及删除、撤回、审批通过、知识库移动/复制等高影响动作。</td><td>需要逐项授权和专门测试空间/机器人/审批单据，不建议混在批量回归里自动跑。</td></tr></tbody>
+</table>
+
+
+<h2>Read 失败逐项 <span class="count">· 42 条</span></h2>
+<table class="review">
+<thead><tr>
+<th>#</th><th>命令</th><th>风险</th><th>类型</th><th>要改哪里</th><th>具体改法</th><th>验证方式</th><th>operation</th><th>trace_id</th><th>证据</th>
+</tr></thead>
+<tbody>
+<tr><td class="num">1</td><td><code>aitable +base-get-primary-doc-id</code></td><td><span class="risk">read</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;-1&quot;,&quot;message&quot;:&quot;no record&quot;,&quot;retryable&quot;:true,&quot;type&quot;:&quot;SYSTEM_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to query cell doc for record 1015oH3OXy in table…</td></tr>
+<tr><td class="num">2</td><td><code>aitable +chart-share-get</code></td><td><span class="risk">read</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `aitable +chart-share-get`；若仍是 permission，再看 trace_id。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;403&quot;,&quot;message&quot;:&quot;Forbidden&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;AUTH_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to get chart share config for chart widget-dlxFo…</td></tr>
+<tr><td class="num">3</td><td><code>aitable +dashboard-share-get</code></td><td><span class="risk">read</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `aitable +dashboard-share-get`。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;404&quot;,&quot;message&quot;:&quot;Not Found&quot;,&quot;retryable&quot;:true,&quot;type&quot;:&quot;SYSTEM_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to get dashboard share config for dashboard KY9…</td></tr>
+<tr><td class="num">4</td><td><code>aitable +export-data</code></td><td><span class="risk">read</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_PARAMS&quot;,&quot;message&quot;:&quot;taskId cannot be combined with scope, format, tableId or viewId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,…</td></tr>
+<tr><td class="num">5</td><td><code>aitable +record-primary-doc-get</code></td><td><span class="risk">read</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;-1&quot;,&quot;message&quot;:&quot;no record&quot;,&quot;retryable&quot;:true,&quot;type&quot;:&quot;SYSTEM_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to query cell doc for record 1015oH3OXy in table…</td></tr>
+<tr><td class="num">6</td><td><code>aitable +role-get</code></td><td><span class="risk">read</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_PARAMS&quot;,&quot;message&quot;:&quot;roleId is required&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to get role because roleId …</td></tr>
+<tr><td class="num">7</td><td><code>aitable +workflow-get</code></td><td><span class="risk">read</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;GET_WORKFLOW_ERROR&quot;,&quot;message&quot;:&quot;调用远程服务业务异常&quot;,&quot;retryable&quot;:true,&quot;type&quot;:&quot;SYSTEM_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to get workflow in base &#x27;gpG2Nd…</td></tr>
+<tr><td class="num">8</td><td><code>aitable +workflow-list</code></td><td><span class="risk">read</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;LIST_WORKFLOWS_ERROR&quot;,&quot;message&quot;:&quot;biz error&quot;,&quot;retryable&quot;:true,&quot;type&quot;:&quot;SYSTEM_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to list workflows in base &#x27;gpG…</td></tr>
+<tr><td class="num">9</td><td><code>attendance +get-class</code></td><td><span class="risk">read</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实考勤班次/考勤组/员工/假期等资源 ID；当前 no-such ID 只能验证负向路径。</td><td>先用考勤列表/管理后台拿真实 ID，再重跑该 attendance 命令。</td><td><code>tools/call</code></td><td><code>2127d89817840997588238831e0757</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/get_class_detail) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">10</td><td><code>attendance +get-global-setting</code></td><td><span class="risk">read</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `attendance +get-global-setting`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2104a64c17840997604718062e085e</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/query_global_setting) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">11</td><td><code>attendance +get-group</code></td><td><span class="risk">read</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实考勤班次/考勤组/员工/假期等资源 ID；当前 no-such ID 只能验证负向路径。</td><td>先用考勤列表/管理后台拿真实 ID，再重跑该 attendance 命令。</td><td><code>tools/call</code></td><td><code>0b5deb3217840997620512305e08ef</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/get_group_detail) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">12</td><td><code>attendance +get-group-filtered</code></td><td><span class="risk">read</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实考勤班次/考勤组/员工/假期等资源 ID；当前 no-such ID 只能验证负向路径。</td><td>先用考勤列表/管理后台拿真实 ID，再重跑该 attendance 命令。</td><td><code>tools/call</code></td><td><code>2104a64c17840997638062468e08c7</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/get_group_filtered_detail) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">13</td><td><code>attendance +get-leave-balance</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `attendance +get-leave-balance` 并比较 stdout/stderr。</td><td><code>tools/call</code></td><td><code>2104a64c17840997652901307e081a</code></td><td class="evidence">[UNCLASSIFIED] 亲，假期类型没有余额 (operation: attendance-wukong/get_leave_balance_quota) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">14</td><td><code>attendance +list-report-columns</code></td><td><span class="risk">read</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `attendance +list-report-columns`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2127d89817840997666188472e0756</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/get_report_columns) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">15</td><td><code>attendance +query-report-leave</code></td><td><span class="risk">read</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `attendance +query-report-leave`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2104a64c17840997679973065e085f</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/get_leave_time_by_leave_names) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">16</td><td><code>calendar +find-room</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入</span></td><td>会议室类命令需要真实 roomId 或更小会议室分组；修改 runner 先定位会议室/分组，再喂给查询命令。</td><td>用真实 roomId/分组重跑 calendar room/freebusy 命令。</td><td><code>tools/call</code></td><td><code>0bb7c36217840997694975303e0758</code></td><td class="evidence">[UNCLASSIFIED] 查询范围内的会议室数量，超过上限100，请选择更小范围的分组进行查询。 hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">17</td><td><code>calendar +room-find</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入</span></td><td>会议室类命令需要真实 roomId 或更小会议室分组；修改 runner 先定位会议室/分组，再喂给查询命令。</td><td>用真实 roomId/分组重跑 calendar room/freebusy 命令。</td><td><code>tools/call</code></td><td><code>2104a64c17840997709913005e0819</code></td><td class="evidence">[UNCLASSIFIED] 查询范围内的会议室数量，超过上限100，请选择更小范围的分组进行查询。 (operation: calendar/query_available_meeting_room) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">18</td><td><code>chat +category-list-conversations</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `chat +category-list-conversations` 并比较 stdout/stderr。</td><td><code>tools/call</code></td><td><code>0bb7c36217840997724375834e0758</code></td><td class="evidence">[UNCLASSIFIED] listConversationsByCategoryV2 error hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">19</td><td><code>chat +chat-get-by-id</code></td><td><span class="risk">read</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `chat +chat-get-by-id`。</td><td><code>tools/call</code></td><td><code>2127d89817840997739165535e07bd</code></td><td class="evidence">[UNCLASSIFIED] verifyGroupId error: The group id does not exit (operation: im/get_conv_info_by_group_id) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">20</td><td><code>chat +chat-members-get</code></td><td><span class="risk">read</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>2127d89817840997754345760e07bd</code></td><td class="evidence">[UNCLASSIFIED] openCid or cid is required (operation: im/list_group_member_by_ids) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">21</td><td><code>chat +chat-messages</code></td><td><span class="risk">read</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>2104a64c17840997767792656e085e</code></td><td class="evidence">[UNCLASSIFIED] openCid or cid is required hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">22</td><td><code>chat +messages-list</code></td><td><span class="risk">read</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>2127d89817840997797873841e0757</code></td><td class="evidence">[UNCLASSIFIED] openCid or cid is required hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">23</td><td><code>chat +messages-resource-url</code></td><td><span class="risk">read</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `chat +messages-resource-url`。</td><td><code>tools/call</code></td><td><code>2127d89817840997855423145e07dd</code></td><td class="evidence">[UNCLASSIFIED] failed to get download url for resourceId: x (operation: im/get_resource_download_url) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">24</td><td><code>chat +search-msg</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试数据</span></td><td>准备能命中的真实数据，或把 runner 查询词改成当前账号一定存在的对象；shortcut 本身不需要改。</td><td>造数后重跑，预期从 validation empty result 变为成功。</td><td><code>tools/call</code></td><td><code>0b5deb3217840997866824643e0853</code></td><td class="evidence">[UNCLASSIFIED] 当前用户暂无消息搜索权益，无法执行本次搜索。请提示用户开通消息搜索权益后重试。 hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">25</td><td><code>chat +thread-replies</code></td><td><span class="risk">read</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `chat +thread-replies`。</td><td><code>tools/call</code></td><td><code>0b5deb3217840997883504915e0853</code></td><td class="evidence">[UNCLASSIFIED] failed to decrypt openConvThreadId hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">26</td><td><code>contact +get-roster</code></td><td><span class="risk">read</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `contact +get-roster`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2106d98117840997920166915e087b</code></td><td class="evidence">[UNCLASSIFIED] 操作人无花名册管理权限 (operation: hrmregister/get_authorized_emp_rosterInfo) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">27</td><td><code>contact +list-roster-fields</code></td><td><span class="risk">read</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `contact +list-roster-fields`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>213ee25c17840997934295673e08e2</code></td><td class="evidence">[UNCLASSIFIED] 操作人无花名册管理权限 (operation: hrmregister/list_authorized_roster_fields) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">28</td><td><code>devapp +credentials-get</code></td><td><span class="risk">read</span></td><td><span class="cat">敏感/高风险暂缓</span></td><td><span class="owner">人工安全确认</span></td><td>该项涉及敏感读取或无安全负向目标，不适合自动用真实资源跑；需要在安全环境逐项人工确认。</td><td>人工确认后单独重跑 `devapp +credentials-get`，并避免在报告中泄露密钥/凭证。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">该命令会读取真实应用凭证/密钥；不能用真实 app 自动执行。当前仅用占位 ID 验证负向路径，真实成功需人工在安全环境单独确认。</td></tr>
+<tr><td class="num">29</td><td><code>drive +download</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `drive +download` 并比较 stdout/stderr。</td><td><code>tools/call</code></td><td><code>0bab027317840997956923965e08c9</code></td><td class="evidence">[UNCLASSIFIED] 该文件类型不支持通过 download_file 下载。download_file 仅支持普通文件（如 PDF、Word、Excel 等），不支持钉钉在线文档/表格/脑图等在线编辑类型。如需导出在线文档内容，请使用钉钉文档导出相关接口。 (operation: drive/download_file) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">30</td><td><code>drive +list</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `drive +list` 并比较 stdout/stderr。</td><td><code>tools/call</code></td><td><code>2106d98117840997968627612e087b</code></td><td class="evidence">[UNCLASSIFIED] parentId 不属于指定的 spaceId，请确认 parentId 和 spaceId 属于同一个钉盘空间。 hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">31</td><td><code>minutes +action-items</code></td><td><span class="risk">read</span></td><td><span class="cat">缺妙记 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备当前账号可见的真实妙记/听记/录制会话，或把 runner 的搜索关键词改成必然能命中的会议产物。</td><td>用真实 taskUuid/note/minutes 资源重跑 minutes 命令。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">暂无妙记</td></tr>
+<tr><td class="num">32</td><td><code>minutes +latest-minutes</code></td><td><span class="risk">read</span></td><td><span class="cat">缺妙记 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备当前账号可见的真实妙记/听记/录制会话，或把 runner 的搜索关键词改成必然能命中的会议产物。</td><td>用真实 taskUuid/note/minutes 资源重跑 minutes 命令。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">暂无妙记</td></tr>
+<tr><td class="num">33</td><td><code>minutes +minutes-search</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试数据</span></td><td>准备能命中的真实数据，或把 runner 查询词改成当前账号一定存在的对象；shortcut 本身不需要改。</td><td>造数后重跑，预期从 validation empty result 变为成功。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">没搜到妙记</td></tr>
+<tr><td class="num">34</td><td><code>minutes +transcript</code></td><td><span class="risk">read</span></td><td><span class="cat">缺妙记 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备当前账号可见的真实妙记/听记/录制会话，或把 runner 的搜索关键词改成必然能命中的会议产物。</td><td>用真实 taskUuid/note/minutes 资源重跑 minutes 命令。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">暂无妙记</td></tr>
+<tr><td class="num">35</td><td><code>oa +done-approvals</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试数据</span></td><td>准备能命中的真实数据，或把 runner 查询词改成当前账号一定存在的对象；shortcut 本身不需要改。</td><td>造数后重跑，预期从 validation empty result 变为成功。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">没有已处理的审批记录</td></tr>
+<tr><td class="num">36</td><td><code>oa +pending</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试数据</span></td><td>准备能命中的真实数据，或把 runner 查询词改成当前账号一定存在的对象；shortcut 本身不需要改。</td><td>造数后重跑，预期从 validation empty result 变为成功。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">当前没有待我审批的任务</td></tr>
+<tr><td class="num">37</td><td><code>report +report-latest</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试数据</span></td><td>准备能命中的真实数据，或把 runner 查询词改成当前账号一定存在的对象；shortcut 本身不需要改。</td><td>造数后重跑，预期从 validation empty result 变为成功。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">暂无日志</td></tr>
+<tr><td class="num">38</td><td><code>todo +due-today</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试数据</span></td><td>准备能命中的真实数据，或把 runner 查询词改成当前账号一定存在的对象；shortcut 本身不需要改。</td><td>造数后重跑，预期从 validation empty result 变为成功。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">今天没有到期的待办</td></tr>
+<tr><td class="num">39</td><td><code>todo +related-tasks</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试数据</span></td><td>准备能命中的真实数据，或把 runner 查询词改成当前账号一定存在的对象；shortcut 本身不需要改。</td><td>造数后重跑，预期从 validation empty result 变为成功。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">没有与你相关的待办（creator/executor/participant 三种角色下均为空）</td></tr>
+<tr><td class="num">40</td><td><code>wiki +node-list</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">后端业务/测试 fixture</span></td><td>后端只返回 success=false，信息不足；先准备真实合法 fixture，若仍无细节，需要后端补充错误码/错误信息。</td><td>用真实资源重跑；若仍 success=false，把 operation+trace_id 给后端。</td><td><code>tools/call</code></td><td><code>2132f5ca17840998189126304e08d9</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">41</td><td><code>wiki +resolve-space</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试数据</span></td><td>准备能命中的真实数据，或把 runner 查询词改成当前账号一定存在的对象；shortcut 本身不需要改。</td><td>造数后重跑，预期从 validation empty result 变为成功。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">没有找到名称包含 DWS shortcut 真实测试 的知识空间</td></tr>
+<tr><td class="num">42</td><td><code>wiki +space-list</code></td><td><span class="risk">read</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `wiki +space-list` 并比较 stdout/stderr。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">参数 --type 取值 &quot;ALL&quot; 不合法，允许值：orgWikiSpace, myWikiSpace</td></tr>
+</tbody>
+</table>
+
+
+<h2>Write 失败逐项 <span class="count">· 114 条</span></h2>
+<table class="review">
+<thead><tr>
+<th>#</th><th>命令</th><th>风险</th><th>类型</th><th>要改哪里</th><th>具体改法</th><th>验证方式</th><th>operation</th><th>trace_id</th><th>证据</th>
+</tr></thead>
+<tbody>
+<tr><td class="num">1</td><td><code>aitable +advperm-disable</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `aitable +advperm-disable`；若仍是 permission，再看 trace_id。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to set adv…</td></tr>
+<tr><td class="num">2</td><td><code>aitable +advperm-enable</code></td><td><span class="risk">write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `aitable +advperm-enable`；若仍是 permission，再看 trace_id。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to set adv…</td></tr>
+<tr><td class="num">3</td><td><code>aitable +attachment-upload</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;BASE_NOT_FOUND&quot;,&quot;message&quot;:&quot;Specified base does not exist, has been deleted, or is inaccessible&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:t…</td></tr>
+<tr><td class="num">4</td><td><code>aitable +base-copy</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:null,&quot;message&quot;:&quot;Invalid source baseId: DWSREALTESTNOSUCHID0000000000000&quot;,&quot;retryable&quot;:null,&quot;type&quot;:&quot;USER_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Invalid sou…</td></tr>
+<tr><td class="num">5</td><td><code>aitable +base-delete</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `aitable +base-delete`。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;52600003&quot;,&quot;message&quot;:&quot;Data not found&quot;,&quot;retryable&quot;:true,&quot;type&quot;:&quot;SYSTEM_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to delete base DWSREALTESTNOSUCHID000…</td></tr>
+<tr><td class="num">6</td><td><code>aitable +base-update</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;BASE_NOT_FOUND&quot;,&quot;message&quot;:&quot;Specified base does not exist, has been deleted, or is inaccessible&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:t…</td></tr>
+<tr><td class="num">7</td><td><code>aitable +chart-delete</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to delete …</td></tr>
+<tr><td class="num">8</td><td><code>aitable +chart-share-update</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to update …</td></tr>
+<tr><td class="num">9</td><td><code>aitable +chart-update</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_PARAMS&quot;,&quot;message&quot;:&quot;config is required&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to update chart because con…</td></tr>
+<tr><td class="num">10</td><td><code>aitable +dashboard-arrange</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to align d…</td></tr>
+<tr><td class="num">11</td><td><code>aitable +dashboard-delete</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to delete …</td></tr>
+<tr><td class="num">12</td><td><code>aitable +dashboard-share-update</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to update …</td></tr>
+<tr><td class="num">13</td><td><code>aitable +dashboard-update</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to update …</td></tr>
+<tr><td class="num">14</td><td><code>aitable +field-delete</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;404&quot;,&quot;message&quot;:&quot;getDentryDTO returns null&quot;,&quot;retryable&quot;:true,&quot;type&quot;:&quot;SYSTEM_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to get current field info befor…</td></tr>
+<tr><td class="num">15</td><td><code>aitable +field-update</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;404&quot;,&quot;message&quot;:&quot;getDentryDTO returns null&quot;,&quot;retryable&quot;:true,&quot;type&quot;:&quot;SYSTEM_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to get current field info for u…</td></tr>
+<tr><td class="num">16</td><td><code>aitable +form-delete</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to delete …</td></tr>
+<tr><td class="num">17</td><td><code>aitable +form-field-hide</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to update …</td></tr>
+<tr><td class="num">18</td><td><code>aitable +form-field-update</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to update …</td></tr>
+<tr><td class="num">19</td><td><code>aitable +form-share-update</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to update …</td></tr>
+<tr><td class="num">20</td><td><code>aitable +form-update</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to update …</td></tr>
+<tr><td class="num">21</td><td><code>aitable +import-data</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `aitable +import-data`。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_IMPORT_ID&quot;,&quot;message&quot;:&quot;importId not found: either invalid or expired&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;impo…</td></tr>
+<tr><td class="num">22</td><td><code>aitable +import-upload</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;无法解析 baseId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;无效的 baseId&quot;,&quot;trace_id&quot;:&quot;0bab027317840998…</td></tr>
+<tr><td class="num">23</td><td><code>aitable +record-delete</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;404&quot;,&quot;message&quot;:&quot;getDentryDTO returns null&quot;,&quot;retryable&quot;:true,&quot;type&quot;:&quot;SYSTEM_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to delete records&quot;,&quot;trace_id&quot;:&quot;…</td></tr>
+<tr><td class="num">24</td><td><code>aitable +record-primary-doc-create</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;RESOLVE_DOC_ID_ERROR&quot;,&quot;message&quot;:&quot;Failed to resolve docId from baseId&quot;,&quot;retryable&quot;:true,&quot;type&quot;:&quot;SYSTEM_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to c…</td></tr>
+<tr><td class="num">25</td><td><code>aitable +record-update</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_RECORDS&quot;,&quot;message&quot;:&quot;records must contain at least one writable record&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Fa…</td></tr>
+<tr><td class="num">26</td><td><code>aitable +record-upsert</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_PARAMETER&quot;,&quot;message&quot;:&quot;records is required and must not be empty&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;records …</td></tr>
+<tr><td class="num">27</td><td><code>aitable +role-create</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to create …</td></tr>
+<tr><td class="num">28</td><td><code>aitable +role-delete</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_PARAMS&quot;,&quot;message&quot;:&quot;roleId is required&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to delete role because role…</td></tr>
+<tr><td class="num">29</td><td><code>aitable +role-update</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_PARAMS&quot;,&quot;message&quot;:&quot;roleId is required&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to patch role because roleI…</td></tr>
+<tr><td class="num">30</td><td><code>aitable +section-create</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to create …</td></tr>
+<tr><td class="num">31</td><td><code>aitable +section-delete</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to delete …</td></tr>
+<tr><td class="num">32</td><td><code>aitable +section-move-node</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to move no…</td></tr>
+<tr><td class="num">33</td><td><code>aitable +section-rename</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to rename …</td></tr>
+<tr><td class="num">34</td><td><code>aitable +section-reorder</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to reorder…</td></tr>
+<tr><td class="num">35</td><td><code>aitable +table-delete</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;BASE_NOT_FOUND&quot;,&quot;message&quot;:&quot;Specified base does not exist, has been deleted, or is inaccessible&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:t…</td></tr>
+<tr><td class="num">36</td><td><code>aitable +table-update</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;404&quot;,&quot;message&quot;:&quot;getDentryDTO returns null&quot;,&quot;retryable&quot;:true,&quot;type&quot;:&quot;SYSTEM_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to update table DWSREALTESTNOSU…</td></tr>
+<tr><td class="num">37</td><td><code>aitable +view-delete</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to delete …</td></tr>
+<tr><td class="num">38</td><td><code>aitable +view-duplicate</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to duplica…</td></tr>
+<tr><td class="num">39</td><td><code>aitable +view-lock</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to lock_or…</td></tr>
+<tr><td class="num">40</td><td><code>aitable +view-set-fill-color-rule</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>修 aitable MCP wrapper 的参数校验和错误语义：不要返回 success=true+error；对 required 字段给出 CLI 可识别的参数名，系统错误要带 retryable/trace。</td><td>MCP 修完后重跑该 aitable 命令，并确认 stdout JSON 不再出现 success=true 但 error 非空。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_PARAMS&quot;,&quot;message&quot;:&quot;conditionalFormats is required&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to set fill col…</td></tr>
+<tr><td class="num">41</td><td><code>aitable +view-set-frozen-cols</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to set fro…</td></tr>
+<tr><td class="num">42</td><td><code>aitable +view-set-row-height</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to set cel…</td></tr>
+<tr><td class="num">43</td><td><code>aitable +view-update</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;INVALID_BASE_ID&quot;,&quot;message&quot;:&quot;baseId cannot be resolved to docId&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;status&quot;:&quot;error&quot;,&quot;success&quot;:true,&quot;summary&quot;:&quot;Failed to update …</td></tr>
+<tr><td class="num">44</td><td><code>aitable +workflow-disable</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;BASE_NOT_FOUND&quot;,&quot;message&quot;:&quot;Cannot resolve base &#x27;DWSREALTESTNOSUCHID0000000000000&#x27;, please check if the baseId is valid&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;sta…</td></tr>
+<tr><td class="num">45</td><td><code>aitable +workflow-enable</code></td><td><span class="risk">write</span></td><td><span class="cat">缺 AI 表格 fixture</span></td><td><span class="owner">测试数据</span></td><td>准备真实 Base/Table/View/Field/Record/Role/Chart/Dashboard 等 fixture，并把真实 ID 写入真实测试 runner；当前安全负向 ID 只能证明调用链，不可能成功。</td><td>fixture 准备好后重跑对应 aitable 命令；预期从 not_found/invalid_base_id 变为成功或更具体业务错误。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">[MCP_TOOL_ERROR] {&quot;data&quot;:{},&quot;error&quot;:{&quot;code&quot;:&quot;BASE_NOT_FOUND&quot;,&quot;message&quot;:&quot;Cannot resolve base &#x27;DWSREALTESTNOSUCHID0000000000000&#x27;, please check if the baseId is valid&quot;,&quot;retryable&quot;:false,&quot;type&quot;:&quot;INPUT_ERROR&quot;},&quot;meta&quot;:{},&quot;sta…</td></tr>
+<tr><td class="num">46</td><td><code>attendance +boss-check</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">后端业务/测试 fixture</span></td><td>后端只返回 success=false，信息不足；先准备真实合法 fixture，若仍无细节，需要后端补充错误码/错误信息。</td><td>用真实资源重跑；若仍 success=false，把 operation+trace_id 给后端。</td><td><code>tools/call</code></td><td><code>213ee25c17840998998942184e087d</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/boss_check) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">47</td><td><code>attendance +create-class</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">后端业务/测试 fixture</span></td><td>后端只返回 success=false，信息不足；先准备真实合法 fixture，若仍无细节，需要后端补充错误码/错误信息。</td><td>用真实资源重跑；若仍 success=false，把 operation+trace_id 给后端。</td><td><code>tools/call</code></td><td><code>0bab027317840999009926838e090b</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/create_class_setting) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">48</td><td><code>attendance +create-group</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">后端业务/测试 fixture</span></td><td>后端只返回 success=false，信息不足；先准备真实合法 fixture，若仍无细节，需要后端补充错误码/错误信息。</td><td>用真实资源重跑；若仍 success=false，把 operation+trace_id 给后端。</td><td><code>tools/call</code></td><td><code>2106d98117840999021121258e08b8</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/create_group_setting) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">49</td><td><code>attendance +import-schedule</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">后端业务/测试 fixture</span></td><td>后端只返回 success=false，信息不足；先准备真实合法 fixture，若仍无细节，需要后端补充错误码/错误信息。</td><td>用真实资源重跑；若仍 success=false，把 operation+trace_id 给后端。</td><td><code>tools/call</code></td><td><code>2104a64c17840999034845189e085e</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/generateTurnSchedule) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">50</td><td><code>attendance +save-leave-balance</code></td><td><span class="risk">write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `attendance +save-leave-balance`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2127d89817840999045751134e079c</code></td><td class="evidence">[UNCLASSIFIED] 无权更新指定员工的假期余额 (operation: attendance-wukong/update_leave_balance) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">51</td><td><code>attendance +update-class</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">后端业务/测试 fixture</span></td><td>后端只返回 success=false，信息不足；先准备真实合法 fixture，若仍无细节，需要后端补充错误码/错误信息。</td><td>用真实资源重跑；若仍 success=false，把 operation+trace_id 给后端。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999058236471e08b5</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/update_class_setting) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">52</td><td><code>attendance +update-group</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">后端业务/测试 fixture</span></td><td>后端只返回 success=false，信息不足；先准备真实合法 fixture，若仍无细节，需要后端补充错误码/错误信息。</td><td>用真实资源重跑；若仍 success=false，把 operation+trace_id 给后端。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999068826691e087a</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/update_group_setting) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">53</td><td><code>attendance +update-group-members</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">后端业务/测试 fixture</span></td><td>后端只返回 success=false，信息不足；先准备真实合法 fixture，若仍无细节，需要后端补充错误码/错误信息。</td><td>用真实资源重跑；若仍 success=false，把 operation+trace_id 给后端。</td><td><code>tools/call</code></td><td><code>2127d89817840999081094644e07dd</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: attendance-wukong/update_group_member) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">54</td><td><code>attendance +update-leave-type</code></td><td><span class="risk">write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `attendance +update-leave-type`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999093924897e07fe</code></td><td class="evidence">[RESOURCE_NOT_FOUND] Requested resource not found (operation: attendance-wukong/save_leave_type) hint: Check if the resource exists or if your account has permission</td></tr>
+<tr><td class="num">55</td><td><code>calendar +respond-event</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `calendar +respond-event`。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999105062121e07db</code></td><td class="evidence">[UNCLASSIFIED] code: 300000, developerMessage: Event does not exist. (operation: calendar/respond) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">56</td><td><code>chat +category-add-conversation</code></td><td><span class="risk">write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `chat +category-add-conversation`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999117776203e08f9</code></td><td class="evidence">[RESOURCE_NOT_FOUND] Requested resource not found (operation: im/add_conv_to_categories) hint: Check if the resource exists or if your account has permission</td></tr>
+<tr><td class="num">57</td><td><code>chat +category-remove-conversation</code></td><td><span class="risk">write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `chat +category-remove-conversation`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2104a64c17840999130466520e085e</code></td><td class="evidence">[RESOURCE_NOT_FOUND] Requested resource not found (operation: im/remove_conv_from_categories) hint: Check if the resource exists or if your account has permission</td></tr>
+<tr><td class="num">58</td><td><code>chat +chat-add-bot</code></td><td><span class="risk">write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `chat +chat-add-bot`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2104a64c17840999144193460e08c7</code></td><td class="evidence">[RESOURCE_NOT_FOUND] Requested resource not found (operation: bot/add_robot_to_group) hint: Check if the resource exists or if your account has permission</td></tr>
+<tr><td class="num">59</td><td><code>chat +chat-audit-join</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999154832733e0758</code></td><td class="evidence">[UNCLASSIFIED] applicantUid is required (operation: im/audit_join_group) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">60</td><td><code>chat +chat-mute-member</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>2104a64c17840999166036583e08a3</code></td><td class="evidence">[UNCLASSIFIED] openConversationId or cid is required (operation: im/set_group_member_mute_list) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">61</td><td><code>chat +chat-quit</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `chat +chat-quit` 并比较 stdout/stderr。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999176728426e0779</code></td><td class="evidence">[UNCLASSIFIED] listBaseConversationByIds error (operation: im/quit_group) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">62</td><td><code>chat +chat-remove-bot</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `chat +chat-remove-bot`。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999189888305e08b5</code></td><td class="evidence">[UNCLASSIFIED] 无效的会话 (operation: bot/remove_robot_in_group) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">63</td><td><code>chat +chat-role-remove</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `chat +chat-role-remove` 并比较 stdout/stderr。</td><td><code>tools/call</code></td><td><code>2127d89817840999200483204e079c</code></td><td class="evidence">[UNCLASSIFIED] listBaseConversationByIds error (operation: im/remove_custom_group_role) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">64</td><td><code>chat +chat-role-remove-user</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `chat +chat-role-remove-user` 并比较 stdout/stderr。</td><td><code>tools/call</code></td><td><code>2127d89817840999211317952e0757</code></td><td class="evidence">[UNCLASSIFIED] listBaseConversationByIds error (operation: im/remove_custom_user_roles) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">65</td><td><code>chat +chat-transfer-owner</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999222318863e087a</code></td><td class="evidence">[UNCLASSIFIED] openConversationId is required (operation: im/transfer_group_owner) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">66</td><td><code>chat +chat-update-icon</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `chat +chat-update-icon` 并比较 stdout/stderr。</td><td><code>tools/call</code></td><td><code>2104a64c17840999232478654e0819</code></td><td class="evidence">[UNCLASSIFIED] listBaseConversationByIds error (operation: im/update_group_icon) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">67</td><td><code>chat +chat-update-settings</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/shortcut 枚举</span></td><td>把测试输入的 setting-key 从 x 改为后端支持的 key；同时可在 shortcut flag 上补 enum，避免用户传非法 key。</td><td>改 runner 后重跑；如果补 enum，跑 shortcut 单测确认校验文案。</td><td><code>tools/call</code></td><td><code>2127d89817840999244326971e07dd</code></td><td class="evidence">[UNCLASSIFIED] unsupported setting key: x (operation: im/update_group_settings) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">68</td><td><code>chat +conversation-clear-messages</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>2127d89817840999254816721e079b</code></td><td class="evidence">[UNCLASSIFIED] openConversationId or cid is required (operation: im/clear_conversation_messages) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">69</td><td><code>chat +conversation-clear-red-point</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>2104a64c17840999265511295e085f</code></td><td class="evidence">[UNCLASSIFIED] openConversationId or cid is required (operation: im/clear_conversation_red_point) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">70</td><td><code>chat +conversation-hide</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>2127d89817840999276117140e079b</code></td><td class="evidence">[UNCLASSIFIED] openConversationId or cid is required (operation: im/hide_conversation) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">71</td><td><code>chat +conversation-mark-read</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为当前账号真实可访问的群/会话 openConversationId；如果用真实群仍报“无效”，再查 IM 后端解析。</td><td>先用 `chat +my-groups` 或群搜索拿真实会话 ID，再重跑。</td><td><code>tools/call</code></td><td><code>2127d89817840999287654280e07bd</code></td><td class="evidence">[UNCLASSIFIED] openConversationId无效，无法解析为cid (operation: im/mark_message_read) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">72</td><td><code>chat +conversation-mark-unread</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999298744910e0758</code></td><td class="evidence">[UNCLASSIFIED] openConversationId or cid is required (operation: im/mark_conversation_unread) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">73</td><td><code>chat +conversation-mute</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>2104a64c17840999309241865e085f</code></td><td class="evidence">[UNCLASSIFIED] openConversationId is required (operation: im/update_notification_off) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">74</td><td><code>chat +conversation-mute-at-all</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>2127d89817840999320028068e07dd</code></td><td class="evidence">[UNCLASSIFIED] openConversationId is required (operation: im/update_at_all_notification_off) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">75</td><td><code>chat +conversation-mute-red-envelope</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999330228283e07fe</code></td><td class="evidence">[UNCLASSIFIED] openConversationId is required (operation: im/update_red_env_notification_off) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">76</td><td><code>chat +conversation-set-top</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>2127d89817840999341302961e07fe</code></td><td class="evidence">[UNCLASSIFIED] openConversationId or cid is required (operation: im/set_top_conversation) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">77</td><td><code>chat +messages-add-emoji</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实 openMessageId，并保证消息属于当前账号可访问会话；当前 no-such ID 只能验证负向路径。</td><td>先用消息列表拿 messageId，再重跑消息详情/状态/撤回类命令。</td><td><code>tools/call</code></td><td><code>2127d89817840999352941910e0756</code></td><td class="evidence">[UNCLASSIFIED] invalid openMsgId: DWSREALTESTNOSUCHID0000000000000 (operation: im/add_emoji_reaction) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">78</td><td><code>chat +messages-add-text-emotion</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实 openMessageId，并保证消息属于当前账号可访问会话；当前 no-such ID 只能验证负向路径。</td><td>先用消息列表拿 messageId，再重跑消息详情/状态/撤回类命令。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999362862711e08b5</code></td><td class="evidence">[UNCLASSIFIED] invalid openMsgId: DWSREALTESTNOSUCHID0000000000000 (operation: im/add_text_emotion) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">79</td><td><code>chat +messages-batch-recall-by-bot</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `chat +messages-batch-recall-by-bot`。</td><td><code>tools/call</code></td><td><code>2104a64c17840999373456305e0817</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: bot/batch_recall_robot_users_msg) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">80</td><td><code>chat +messages-batch-send-by-bot</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `chat +messages-batch-send-by-bot`。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999385748776e0757</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: bot/batch_send_robot_msg_to_users) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">81</td><td><code>chat +messages-combine-forward</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为当前账号真实可访问的群/会话 openConversationId；如果用真实群仍报“无效”，再查 IM 后端解析。</td><td>先用 `chat +my-groups` 或群搜索拿真实会话 ID，再重跑。</td><td><code>tools/call</code></td><td><code>2104a64c17840999396596163e08ee</code></td><td class="evidence">[UNCLASSIFIED] srcOpenCid无效，无法解析为cid (operation: im/combine_forward_messages) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">82</td><td><code>chat +messages-create-text-emotion</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务规则</span></td><td>换成后端支持的文字表情组合，或把该命令保留为业务负向；CLI 不应绕过后端限制。</td><td>用一个真实可保存的表情模板重跑。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999407422891e08cd</code></td><td class="evidence">[UNCLASSIFIED] 暂不支持保存该文字表情 (operation: im/create_text_emotion) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">83</td><td><code>chat +messages-forward</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实 openMessageId，并保证消息属于当前账号可访问会话；当前 no-such ID 只能验证负向路径。</td><td>先用消息列表拿 messageId，再重跑消息详情/状态/撤回类命令。</td><td><code>tools/call</code></td><td><code>2104a64c17840999417966407e08ee</code></td><td class="evidence">[UNCLASSIFIED] openMessageId解密失败 (operation: im/forward_message) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">84</td><td><code>chat +messages-forward-topic</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实 openMessageId，并保证消息属于当前账号可访问会话；当前 no-such ID 只能验证负向路径。</td><td>先用消息列表拿 messageId，再重跑消息详情/状态/撤回类命令。</td><td><code>tools/call</code></td><td><code>2127d89817840999428541474e07dd</code></td><td class="evidence">[UNCLASSIFIED] openMessageId解密失败 (operation: im/forward_topic) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">85</td><td><code>chat +messages-recall</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为当前账号真实可访问的群/会话 openConversationId；如果用真实群仍报“无效”，再查 IM 后端解析。</td><td>先用 `chat +my-groups` 或群搜索拿真实会话 ID，再重跑。</td><td><code>tools/call</code></td><td><code>2104a64c17840999441138103e08c7</code></td><td class="evidence">[UNCLASSIFIED] openConversationId无效，无法解析为cid (operation: im/recall_message) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">86</td><td><code>chat +messages-recall-by-bot</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">后端业务/测试 fixture</span></td><td>后端只返回 success=false，信息不足；先准备真实合法 fixture，若仍无细节，需要后端补充错误码/错误信息。</td><td>用真实资源重跑；若仍 success=false，把 operation+trace_id 给后端。</td><td><code>tools/call</code></td><td><code>2104a64c17840999454382709e08a3</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: bot/recall_robot_group_message) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">87</td><td><code>chat +messages-remove-emoji</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实 openMessageId，并保证消息属于当前账号可访问会话；当前 no-such ID 只能验证负向路径。</td><td>先用消息列表拿 messageId，再重跑消息详情/状态/撤回类命令。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999467733059e08f9</code></td><td class="evidence">[UNCLASSIFIED] invalid openMsgId: DWSREALTESTNOSUCHID0000000000000 (operation: im/remove_emoji_reaction) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">88</td><td><code>chat +messages-remove-text-emotion</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实 openMessageId，并保证消息属于当前账号可访问会话；当前 no-such ID 只能验证负向路径。</td><td>先用消息列表拿 messageId，再重跑消息详情/状态/撤回类命令。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999478923406e087f</code></td><td class="evidence">[UNCLASSIFIED] invalid openMsgId: DWSREALTESTNOSUCHID0000000000000 (operation: im/remove_text_emotion) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">89</td><td><code>chat +messages-send-by-bot</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `chat +messages-send-by-bot`。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999494005921e08ef</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: bot/send_robot_group_message) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">90</td><td><code>chat +messages-send-card</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>2104a64c17840999509255753e081a</code></td><td class="evidence">[UNCLASSIFIED] receiverUid和openConversationId不能同时为空 (operation: im/create_and_send_card) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">91</td><td><code>chat +messages-set-pin</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999521117991e0758</code></td><td class="evidence">[UNCLASSIFIED] openConversationId or cid is required (operation: im/set_pin_message) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">92</td><td><code>chat +messages-set-top</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实 openMessageId，并保证消息属于当前账号可访问会话；当前 no-such ID 只能验证负向路径。</td><td>先用消息列表拿 messageId，再重跑消息详情/状态/撤回类命令。</td><td><code>tools/call</code></td><td><code>2104a64c17840999532438476e0817</code></td><td class="evidence">[UNCLASSIFIED] openMessageId解密失败 (operation: im/set_top_message) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">93</td><td><code>chat +messages-unset-pin</code></td><td><span class="risk">write</span></td><td><span class="cat">后端/MCP</span></td><td><span class="owner">后端/MCP schema</span></td><td>CLI fake MCP 已证明字段已装配；需要修 MCP tool schema 或网关字段映射，确认 openConversationId/openCid/cid、receiverUid、applicantUid 等字段没有在 schema 校验/转发时被丢弃。</td><td>修 MCP 后不改 shortcut，直接重跑真实命令；预期错误从 required 变为资源无效或成功。</td><td><code>tools/call</code></td><td><code>2104a64c17840999544428103e08ee</code></td><td class="evidence">[UNCLASSIFIED] openConversationId or cid is required (operation: im/unset_pin_message) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">94</td><td><code>chat +messages-unset-top</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实 openMessageId，并保证消息属于当前账号可访问会话；当前 no-such ID 只能验证负向路径。</td><td>先用消息列表拿 messageId，再重跑消息详情/状态/撤回类命令。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999554154221e08f9</code></td><td class="evidence">[UNCLASSIFIED] openMessageId解密失败 (operation: im/unset_top_message) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">95</td><td><code>devapp +event-subscribe</code></td><td><span class="risk">write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `devapp +event-subscribe`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2104a64c17840999564022020e08c7</code></td><td class="evidence">[UNCLASSIFIED] 当前用户没有应用事件订阅权限 (operation: devapp/subscribe_dev_app_events) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">96</td><td><code>devapp +event-unsubscribe</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `devapp +event-unsubscribe`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999575698500e07db</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: devapp/unsubscribe_dev_app_events) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">97</td><td><code>devapp +permission-add</code></td><td><span class="risk">write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `devapp +permission-add`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2127d89817840999587758457e079c</code></td><td class="evidence">[UNCLASSIFIED] 当前用户没有开发者身份 (operation: devapp/apply_dev_app_permissions) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">98</td><td><code>devapp +permission-remove</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `devapp +permission-remove`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2104a64c17840999598445247e085e</code></td><td class="evidence">[UNCLASSIFIED] 当前用户没有开发者身份 (operation: devapp/remove_dev_app_permissions) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">99</td><td><code>devapp +robot-config</code></td><td><span class="risk">write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `devapp +robot-config`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999609828961e07db</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: devapp/set_extension_robot_config) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">100</td><td><code>devapp +robot-disable</code></td><td><span class="risk">high-risk-write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `devapp +robot-disable`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2104a64c17840999620461113e08ee</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: devapp/disable_dev_app_robot) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">101</td><td><code>devapp +robot-enable</code></td><td><span class="risk">write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `devapp +robot-enable`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999632641525e0758</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: devapp/enable_dev_app_robot) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">102</td><td><code>devapp +security-config</code></td><td><span class="risk">write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `devapp +security-config`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2104a64c17840999645522046e0817</code></td><td class="evidence">[UNCLASSIFIED] 当前用户没有开发者身份 (operation: devapp/update_dev_app_security_config) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">103</td><td><code>devapp +version-create</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">后端业务/测试 fixture</span></td><td>后端只返回 success=false，信息不足；先准备真实合法 fixture，若仍无细节，需要后端补充错误码/错误信息。</td><td>用真实资源重跑；若仍 success=false，把 operation+trace_id 给后端。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999656705920e0853</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: devapp/create_dev_app_version) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">104</td><td><code>devapp +version-publish</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">后端业务/测试 fixture</span></td><td>后端只返回 success=false，信息不足；先准备真实合法 fixture，若仍无细节，需要后端补充错误码/错误信息。</td><td>用真实资源重跑；若仍 success=false，把 operation+trace_id 给后端。</td><td><code>tools/call</code></td><td><code>2127d89817840999667906017e075d</code></td><td class="evidence">[UNCLASSIFIED] business error: success=false (operation: devapp/publish_dev_app_version) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">105</td><td><code>ding +send-by-message</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `ding +send-by-message` 并比较 stdout/stderr。</td><td><code>tools/call</code></td><td><code>0b5deb3217840999678466213e0853</code></td><td class="evidence">[UNCLASSIFIED] remindType非法，合法值：APP/SMS/PHONE (operation: im/send_ding_by_message) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">106</td><td><code>doc +comment-create-inline</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实文档/节点 ID；文档评论、分享、版本等命令需要资源存在且账号可访问。</td><td>用真实 doc/node 重跑；若仍失败再看 doc/doc-comment 工具字段。</td><td><code>tools/call</code></td><td><code>2127d89817840999689931893e079c</code></td><td class="evidence">[TABLE_NOT_FOUND] Requested resource not found (operation: doc-comment/create_inline_comment) hint: Document may have been deleted or moved</td></tr>
+<tr><td class="num">107</td><td><code>doc +template-apply</code></td><td><span class="risk">write</span></td><td><span class="cat">鉴权/权限</span></td><td><span class="owner">权限/应用配置</span></td><td>给当前登录账号、DWS 应用或对应资源补齐权限/scope；本仓库 shortcut 不应绕过权限。拿 trace_id 给服务端/开放平台排查具体 scope。</td><td>补权限后重跑 `doc +template-apply`；若仍是 permission，再看 trace_id。</td><td><code>tools/call</code></td><td><code>2127d89817840999701186954e0757</code></td><td class="evidence">[RESOURCE_NOT_FOUND] Requested resource not found (operation: doc/apply_doc_template) hint: Check if the resource exists or if your account has permission</td></tr>
+<tr><td class="num">108</td><td><code>minutes +record-pause</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `minutes +record-pause` 并比较 stdout/stderr。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999713124987e0757</code></td><td class="evidence">[UNCLASSIFIED] aiAgentTestRunCmdUnknownError (operation: minutes/执行听记指令-发起AI听记录音) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">109</td><td><code>minutes +record-resume</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `minutes +record-resume` 并比较 stdout/stderr。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999724452848e0758</code></td><td class="evidence">[UNCLASSIFIED] aiAgentTestRunCmdUnknownError (operation: minutes/执行听记指令-发起AI听记录音) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">110</td><td><code>minutes +record-stop</code></td><td><span class="risk">write</span></td><td><span class="cat">输入/业务校验</span></td><td><span class="owner">测试输入/业务校验</span></td><td>当前命令已进入后端业务校验；先把测试输入换成真实合法 fixture，再判断是否需要改 shortcut。</td><td>重跑 `minutes +record-stop` 并比较 stdout/stderr。</td><td><code>tools/call</code></td><td><code>2127d89817840999735397508e0757</code></td><td class="evidence">[UNCLASSIFIED] aiAgentTestRunCmdUnknownError (operation: minutes/执行听记指令-发起AI听记录音) hint: Use --verbose for detailed error logs</td></tr>
+<tr><td class="num">111</td><td><code>oa +approve-by</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `oa +approve-by`。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">没找到待审批单据：待我处理的审批里没有标题/单号包含 &quot;__DWS_SHORTCUT_REAL_TEST_NO_SUCH_APPROVAL_20260715-151724__&quot; 的单据。</td></tr>
+<tr><td class="num">112</td><td><code>wiki +node-copy</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实文档/节点 ID；文档评论、分享、版本等命令需要资源存在且账号可访问。</td><td>用真实 doc/node 重跑；若仍失败再看 doc/doc-comment 工具字段。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999769818495e0779</code></td><td class="evidence">[TABLE_NOT_FOUND] Requested resource not found (operation: doc/copy_document) hint: Document may have been deleted or moved</td></tr>
+<tr><td class="num">113</td><td><code>wiki +node-move</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>替换为真实文档/节点 ID；文档评论、分享、版本等命令需要资源存在且账号可访问。</td><td>用真实 doc/node 重跑；若仍失败再看 doc/doc-comment 工具字段。</td><td><code>tools/call</code></td><td><code>0bb7c36217840999780286509e07fe</code></td><td class="evidence">[TABLE_NOT_FOUND] Requested resource not found (operation: doc/move_document) hint: Document may have been deleted or moved</td></tr>
+<tr><td class="num">114</td><td><code>wiki +wiki-new-doc</code></td><td><span class="risk">write</span></td><td><span class="cat">缺真实资源</span></td><td><span class="owner">测试数据</span></td><td>把 runner 里的安全负向 ID 换成真实资源 ID；当前错误说明调用已进后端，但资源不存在。</td><td>准备 fixture 后重跑 `wiki +wiki-new-doc`。</td><td><code>-</code></td><td><code>-</code></td><td class="evidence">没找到名为 &quot;__DWS_SHORTCUT_REAL_TEST_NO_SUCH_SPACE__&quot; 的知识库；换个更完整/精确的空间名再试。</td></tr>
+</tbody>
+</table>
+
+</main>
+</body>
+</html>

docs/shortcut-handoff.md

@@ -0,0 +1,223 @@
+# DWS Shortcut — 方法论与进展交接文档（换会话续跑用）
+
+> 目的：换新会话直接照此续跑。记录**方法论、已完成进展、如何继续、关键坑位、验证命令**。
+> 分支：`feature/shortcut`（**改动全部未提交**，commit 由用户主动决定）。
+
+---
+
+## 0. 一句话现状
+
+> 2026-07-09 **去冗余（重大修订）**：复盘发现 `internal/helpers/` 早有 ~697 个 `dws <svc> <verb>` 产品命令封装了 281 个 tool；1:1 shortcut 层有 235 个 tool 与之重复。按「tool 已被 helper 封装 且 shortcut 用 CallMCP 无投影」精确删除 **213 条纯重复 shortcut**，1:1 层 511→298、总数 579→366、服务 19→16（aisearch/live/devdoc 整包移除，其 dws 命令仍由 helper 提供）。全绿+真机复验保留命令可用。**教训：建封装层前先审已有封装。**
+
+
+在 `internal/shortcut/` 下建成一套声明式 shortcut 体系：**366 条命令 = 298 条 1:1 封装 + 68 条真·智能编排**（1:1 层原 511，已删 213 条纯重复——helper 层早已封装同一批 tool），另有 **~60 条封装升级到 lark 输出投影保真度**、P2 高频自动沉淀闭环、深度对齐 lark 矩阵。**全绿**（build/gofmt/vet/shortcut 全量测试 `shortcuts=366 assembled=322 validated=44 failed=0`/app 全量回归 72s/真机抽验）。
+
+> 2026-07-09 批33（净新增 1 条·+conflicts 的互补品）：`calendar +free-slots`（找某天工作时段内的空闲时段，list_calendar_events + 合并忙碌区间 + 工作窗口求补集，默认今天 09:00-18:00/--from/--to/--in-days）。真机验证：今日 4 段空档（09:00-09:15/15min、12:00-13:30/90min、14:00-14:30/30min、16:00-17:15/75min），正是忙碌事件的精确补集。conflicts+free-slots 构成真实排期智能。总数 578→579、smart 67→68。全绿。
+
+> 2026-07-09 批32b（净新增 1 条·dws 原生编排，lark 也没有）：`calendar +conflicts`（检测某天日程时间冲突/双重预订，list_calendar_events + 本地两两 [start,end) 重叠检测，默认今天/--in-days）。真机验证：抓到今日 9 个日程里 2 处真实冲突（技术标评审 10:00-11:00 × AIX 共创 10:30-12:00；尖角班 14:30-15:30 × 中控项目 15:00-16:00）。证明复杂写死胡同之外，纯 MCP-tool 的本地编排仍有净新增价值。总数 577→578、smart 66→67。全绿。
+
+> 2026-07-09 批32（review lark 复杂写类 + 架构边界结论，用户指定方向）：fresh review lark 复杂写 shortcut（mail +send/+reply/+forward、drive +import、doc +media-insert/download、base +record-upload-attachment），交叉 dws helpers。**关键结论——架构性死胡同，非没使劲**：① `mail +send` dws 已有 1:1 `+send`(send_email)+`+draft-*`，且 **contact 无 email 字段**→无法按名解析收件人（矩阵早 skip），也无 signature/template/lint 工具；② `drive +import`/`doc +media-*`/`base +record-upload-attachment` 核心是**本地文件字节 PUT/下载落盘**——dws 里由 helper 内部 `httpPutFile`/`http PUT/GET`（drive.go/doc.go）实现、已在 1:1 层覆盖（如 `drive upload --convert`），但 **shortcut 框架 `rt.CallMCP/CallMCPData` 只编排 MCP tool、结构上做不了原始文件 I/O**，故无法在 smart 层组合。**建议**：剩余复杂写要么卡此边界、要么已被 1:1 覆盖；真要补文件类能力应在 helper/1:1 层加命令，而非 shortcut 层。本批 review、无代码改动，总数仍 577。注：本轮触及 session 限额（8:20pm 重置）+ 分类器一度不可用，Bash 受限。
+
+> 2026-07-09 批31（净新增 1 条）：`calendar +my-free`（我自己的忙闲，自动解析当前 userId、默认今天，复用 +free 的 freebusySlots 投影；无需像 +free 传别人姓名）。真机正向验证：返回今日/明日真实忙碌时段 {busy:[{start,end}],userId,free}。总数 576→577、smart 65→66。文档全量同步。全绿。
+
+> 2026-07-09 批30（净新增 1 条）：`contact +me`（当前用户 `get_current_user_profile` + 投影 `{name,userId,mobile,dept,org,email}`，agent 的「我是谁」；区别于 1:1 `+get-self` 吐冗长 `result[].orgEmployeeModel` raw）。真机正向验证：董鑫阳/202397/模型算法/钉钉。总数 575→576、smart 64→65。文档全量同步。全绿。
+
+> 2026-07-09 批29（净新增便利读 3 条 + 真机抓修 1 bug）：`oa +done-approvals`（审批历史 get_done_tasks）、`mail +recent-mail`（近期收件 list_mailbox_threads + 解析绑定邮箱/收件箱 folder）、`attendance +this-month`（本月打卡 query_check_record on attendance-wukong，复用 +my-attendance）。**真机抓到并修复 bug**：`+done-approvals` 原来 --limit 不传时 pageSize=0 → 后端 business error（1:1 list-executed 有默认所以正常）；改为默认 pageSize=20，真机复验走空路径「没有已处理的审批记录」。+this-month 真机有效空、+recent-mail 正确报未绑定邮箱。总数 572→575、smart 61→64。文档全量同步。全绿。
+
+> 2026-07-09 批28（净新增便利读 smart，多 agent 并行 + 手工）：再建 3 条只读 smart——`oa +pending`（`list_pending_approvals` 只读列待我审批，区别于会审批的 +approve-by）、`todo +due-today`（`get_user_todos_in_current_org` + `planFinishDateStart/End` 服务端过滤今天到期，区别于 +overdue 已过期）、`calendar +tomorrow`（明天日程，复用 +today/+week 投影）。真机：+tomorrow 返回真实明日日程；+pending/+due-today 空路径正确且复用已验证 helper。3 条为 dws 原生便利读、不对应 lark gap，矩阵 42/48 不变，总数 569→572、smart 58→61。文档全量同步。全绿。
+
+> 2026-07-09 批27（写类输出扫荡收尾，确认无更多 bug）：扫 smart 里丢弃 CallMCPData 结果的 3 处——`broadcast`（per-recipient 循环、最终结构化输出，OK）、`book`（弃 add-participant 结果但最终 `get_calendar_detail` 确认 + 失败回滚，OK）、`reschedule`（弃存在性 check detail 是有意的，随后打 update 结果，OK）。**无更多 silent-success bug**。结论：**输出质量扫荡完成**，只读投影 clean、写类确认结果、honor --format。剩余仅复杂 net-new gap-buildable（mail +send/drive +import 等多步写、难安全真机验）或 commit。真机+一致性改进累计 13 条，总数 569，全绿。
+
+> 2026-07-09 批26（写类 smart 输出一致性批量修）：扫 smart 里用 `fmt.Print*` / 无标准输出的。修 2 条：`chat +broadcast`（`fmt.Printf` 群发摘要忽略 --format → `rt.Output({sentCount,failedCount,sent,failed})`）；`wiki +wiki-new-doc`（**原创建文档后丢弃 create_file 结果、静默 return nil**，真 UX bug 拿不到新文档 id/url → 捕获并 `rt.Output({created,space,title,result})`）。写类无法真机验（会真建/发），assemble 测试确认组装正确、低风险。总数仍 569。
+
+> 2026-07-09 批25（+next-event 输出一致性修复 + sweep 确认多数已 clean）：sweep 探 chat +conversation-list/+category-list、aitable +base-list 等——**多数 1:1 只读命令输出已 clean**（属先前 ~60 升级覆盖），保真度工作基本到位。**修复 1 条一致性**：`calendar +next-event` 原用 `fmt.Println` 打固定文本行、**忽略 --format/--jq/--fields**，改为 `rt.Output(map{event:项目投影})`（复用 +today/+week 同款投影），真机复验 `--format json` 出结构化 `{event:{title,start,end,location,eventId}}`、`--jq '.event.title'` 可用。这是 Agent 友好性修复。同时删除死代码 `shortcutNextEventSummary` + 无用 fmt import。总数仍 569。
+
+> 2026-07-09 批24（1:1 层保真度升级续）：`contact +list-followings` 原 `rt.CallMCP` 吐 `{arguments,result:{models:[…]}}` 信封噪音，改为 `CallMCPData`+`listFollowingsProject`，真机复验干净 `{count:13, followings:[{openDingTalkId}]}`。总数仍 569。**判断**：真机验证 + 保真度升级已到深度边际收益区（本批仅拍平 ID 列表）；1:1 层多数只读命令要么需特定参数、要么后端权限受限、要么输出已可接受。**建议优先 commit 留存 24 批成果**（10 个真机改进 + 58 smart + 保真度升级 + P2 + 全套文档），再按需推进剩余 1:1 微升级。
+
+> 2026-07-09 批23（验证驱动的 1:1 层保真度升级起步）：真机探 1:1 只读命令，`drive +recent` 原 `rt.CallMCP` 吐冗长 raw（logId/nextCursor 噪音 + 每项巨型 docUrl + hasMore），改为 `CallMCPData`+`recentListProject`：投影 `{count, hasMore, items:[{name,nodeType,contentType,accessTime,docUrl,nodeId}], nextCursor}`，去 logId 噪音、保留分页与链接，真机复验干净。`drive +list-spaces` 真机空（有 errorCode 包裹噪音但 result.items 为空，暂不动）。总数仍 569。**注**：1:1 层仍有数十个 list 命令可类似升级，但属边际收益、量大，建议按需/被动推进，优先 commit 留存已有成果。
+
+> 2026-07-09 批22（报告综合更新，反映真机验证战役）：给 `shortcut-report.md` 新增 §2.4「真机验证战役」：记录 9 批真机验证（正向验证 20+ 条、抓修 8 个真实 bug 的表格、后端受限项、resolveUser 非 bug 澄清）；`shortcut-report.html` §③ 测试表补 2 行 + 一段说明。诚实反映「assemble 合成测试盲区 → 真机验证补齐」的价值。纯文档，无代码改动，总数仍 569。
+
+> 2026-07-09 批21（find-record 验证 + +suggest-time 保真度升级）：`aitable +find-record` 真机正常（返回真实记录；cells 按字段 ID 键值、内含附件对象，天然复杂，clean 投影需 field-id→name 解析属更大改造，暂留）。**升级 1 条**：`calendar +suggest-time` 原 `rt.CallMCP` 吐 `result.recommendEventTimes[]` 且 `timeConflictAttendees:[null]` 噪音，改为 `CallMCPData`+`suggestTimeSlots`+`Output`：拍平 result、丢弃 null 冲突项，真机复验干净 `{suggestions:[{start,end}]}`（有真实冲突时才带 conflicts）。总数仍 569。**说明**：真机验证扫荡已进入边际收益递减区（明显 raw-verbose 的 wart 基本清完），剩余多为 minutes org-gated、写类、或输出已可接受。列出 12 条只读仍用 raw `rt.CallMCP` 的 smart（多为 minutes org-gated 或已验证 clean）。**升级 1 条**：`calendar +today` 原直吐 17 字段冗长事件（含完整 attendees 数组），改为 `CallMCPData`+复用 `+week` 的 `shortcutNextEventList/Start` 投影，真机复验干净输出 `{events:[{title,start,end,location,eventId}]}`（与 +week 一致 + location）。总数仍 569。剩余 raw-CallMCP 只读 smart：action-items/latest-minutes/transcript（minutes org-gated 无法真机验）、org/report-latest（已验 clean）、find-record/suggest-time/by-mobile/lookup/team（待验或权限受限）。
+
+> 2026-07-09 批19（日历只读 smart 验证 + +free 保真度升级）：`calendar +next-event` 真机正常（可读摘要「下一个日程：致拓 AI FDE 经验分享…」，但**忽略 --format json 只吐文本**，已知小瑕疵未改）。**升级 1 条**：`calendar +free` 终结步原 `rt.CallMCP` 直吐冗长 `result[].scheduleItems[].{start,end}.dateTime` 嵌套，改为 `CallMCPData`+`freebusySlots`+`Output`，真机复验干净输出 `{who,userId,free,busy:[{start,end}]}`（董鑫阳 2026-07-10 忙 4 段）。总数仍 569。
+
+> 2026-07-09 批18（真机验证续 + +group-members 保真度升级）：**验证正常**：`contact +org`（董鑫阳→模型算法/17人，3步链）、`drive +find-file`（干净投影 {dentryId,fileSize,name,type}）。**后端受限（非 bug）**：`contact +team`（列部门成员 `PAT_MEDIUM_RISK_NO_PERMISSION`）、`chat +search-msg`（org 未开 CLI 数据访问 `TOKEN_VERIFIED_FAILED`）。**升级 1 条**：`chat +group-members` 终结步原用 `rt.CallMCP`（直吐原始冗长 `result.list[]` + memberAvatarMediaId + arguments/errorCode 噪音），改为 `CallMCPData`+`groupMemberProject`+`Output`，真机复验干净输出 `{count, members:[{name,nick,role,openDingtalkId}]}`（刘力/怒龙/群主…）。总数仍 569。
+
+> 2026-07-09 批17（修复批16 发现的 +at-me 投影）：`chat +at-me` 原来因 `atMeMessageItems` 不认识真实两层嵌套 `result.conversationMessagesList[].messages[]` → 命中 fallback、直接吐原始结构。真机 dump 出真实结构（group 有 title/openConversationId/messages；message 有 sender/content/createTime/openConversationId），新增 `atMeFlattenGroups` 把各会话组拍平成单一消息列表、并把组的会话 title 下沉到每条消息。真机复验：43 条消息干净投影为 `{conversation,sender,text,time}`（如 conversation:"AI全栈"、sender:"龙衔"）。总数仍 569。
+
+> 2026-07-09 批16（只读 smart 真机验证扫荡 + 质量修复）：真机跑一批时间/自身类只读 smart。**验证正常**：`calendar +today`（真实日程+参会人）、`calendar +week`（干净投影）、`todo +overdue`（空）、`attendance +my-attendance`（空）、`report +report-latest`（"暂无日志"空路径）、`oa +my-initiated`（真实审批数据）。**修复 1 个输出 wart**：`chat +unread-chats` 每行都吐 `unread: null`——因 `unread_message_conversation_list` 根本不返回每会话未读数（在列表里即代表未读），改为「仅当 gateway 真返回未读数时才带 unread 字段」，真机复验输出已干净 `{conversationId,name}`。**已知待优化（未改）**：`chat +at-me` 返回 `result.conversationMessagesList[].messages[]` 冗长嵌套原始结构、未拍平成干净消息列表（功能正常，投影可再优化）。总数仍 569。
+
+> 2026-07-09 批15（质量修复 + resolveUser 排查，真机）：**修复** `contact +dept-members` 消歧消息 `<red>` 标记泄漏——复用 `stripHighlightTags`（resolve_dept.go）在 name 提取处剥离，真机复验消息已干净（"开放平台(666202009)、技术平台-开放平台研发(1085781688)…"）。注：`dept_members.go` 本身容器解析(含 deptList)+数值 deptId 早已健壮，仅 name markup 未剥。**排查澄清（非 bug）**：`resolveUser` 对 `董鑫阳` 真机端到端正常（userId 202397、部门 模型算法）；但对 `秋画` 这类联系人 `search_contact_by_key_word` 返回 name/userId 全 null（仅 openDingTalkId），resolveUser 正确报「没找到」而非瞎猜——这是钉钉数据模型现实（外部/受限联系人无 userId），非代码 bug。**已知限制**：按名解析仅对「搜索能返回 userId 的组织内成员」有效。总数仍 569。
+
+> 2026-07-09 批14（真机验证续，需具体 ID 的只读 shortcut）：**正向验证过**：`chat +my-groups`（98 真实群+投影）、`aitable +base-list`/`+list-tables`（真实 base/table）、`aitable +resolve-table`（单命中 通用→99dV75A、多候选消歧，容器 key `tables` 正确，无 deptList-class bug）。**后端权限受限、无法正向验证（非代码 bug）**：`chat +chat-messages`（`PAT_MEDIUM_RISK_NO_PERMISSION`，读会话消息需更高权限）、`minutes +*`（该 org 未开启 CLI 数据访问 `TOKEN_VERIFIED_FAILED`）。结论：可验证的 read/resolve shortcut 全部投影正确，仅批13 的 resolve-dept 有真 bug 已修。
+
+> 2026-07-09 批13（真机验证 + bug 修复，登录态 corp「钉钉」）：用登录态把批9-12 只读 shortcut 打真实后端。**正向验证过**：`doc +find-doc`（10 真实文档、投影干净）、`aitable +resolve-base`（多候选真实 baseId）、`mail +find-mail-user`（命中真实用户+邮箱）、`contact +resolve-dept`（修复后返回真实候选）。**真机抓到并修复 1 个真 bug**：`contact +resolve-dept` 原来对任何真实部门名都返回「未找到」——真实 `search_dept_by_keyword` 响应容器 key 是 **`deptList`**（agent 的探测清单漏了），且 `deptName` 带 `<red>…</red>` 高亮标记、`deptId` 是数值。已修：容器加 `deptList`、`stripHighlightTags` 去标记、deptId 数值 coerce 成串（`resolve_dept.go`），真机复验通过（开放平台→666202009、财务→846624121，名称干净）。**已知遗留（未改）**：`contact +dept-members` 的消歧提示消息里 `<red>` 标记未剥离（仅 cosmetic，功能正常）。总数仍 569，本批未加新命令。
+
+> 2026-07-09 批12（多 agent 并行，dws 原生 resolver 层）：再建 3 条「按名解析 ID」智能 shortcut——`wiki +resolve-space`（search_wikiSpaces 名→spaceId）、`aitable +resolve-table`（get_tables 在 Base 内本地名→tableId）、`contact +resolve-dept`（search_dept_by_keyword 名→deptId，**已修数值 ID 兼容**：deptId 为 JSON number 时 coerce 成串，非 string-only）。均 0/1/多候选消歧，对标 resolveUser 各资源版。这 3 条不对应具体 lark gap（是 dws 原生便利层），故 gap-buildable/covered-smart 矩阵计数不变(42/48)，仅总数 566→569、smart 55→58。文档全量同步。全绿。
+
+> 2026-07-09 批11（多 agent 并行）：再建 3 条智能 shortcut——`aitable +resolve-base`（search_bases 按名解析 baseId + 0/1/多候选消歧）、`chat +chat-messages`（群/单聊会话消息 list_conversation_message_v2 / list_individual_chat_message，ExactlyOne 互斥 + 投影）、`mail +find-mail-user`（search_mail_users 按名搜企业邮箱联系人 + 投影）。文档全量同步 566/505/55（gap-buildable 49→42、covered-smart→48）。全绿。注：本批 app 回归首跑因并发负载 flaky FAIL 一次（80s），连跑 2 次稳定 PASS（71s）——非本次改动导致。
+
+> 2026-07-09 批10（多 agent 并行）：再建 3 条智能 shortcut——`chat +thread-replies`（list_topic_replies 拉话题回复 + sender/text/time 投影）、`todo +related-tasks`（get_user_todos_in_current_org 三角色 creator+executor+participant 并集 + taskId 去重 + 投影）、`doc +find-doc`（search_documents 关键词搜文档 + title/url/type/token 投影）。均以 helper 为 ground truth、0 编造。文档全量同步到 563/503/52（report.md/html、lark-alignment.md gap-buildable 49→44、covered-smart→46、comparison.html 重生成）。全量测试 + app 回归全绿。
+
+> 2026-07-09 续跑增量（批9·手工）：新建 3 条智能 shortcut——`minutes +detail`（单命令聚合一条听记 basic/summary/keywords/transcript/todos、partial-failure 容错）、`minutes +replace-batch`（多组 `原文=>替换` 批量替换、去重校验+逐组聚合）、`aitable +record-share-links`（>20 条记录分享链接：去重+分片≤20/批+跨 `aitable-helper` server fanout+合并）。均以 helper 为 ground truth。**同步刷新全部文档到 560/501/49**：`shortcut-report.md`、`shortcut-report.html`、`shortcut-lark-alignment.md`（gap-buildable 49→46、covered-smart 41→44）、重生成 `shortcut-comparison.html`。全量测试 + app 回归全绿。
+
+---
+
+## 1. 背景与目标
+
+- 对齐基准：`/Users/dennis/Projects/larksuite/cli`（lark-cli 的 `shortcuts/` 框架，飞书 REST API）。
+- dws 执行底座：**钉钉 MCP**（粗粒度：一个 tool = 一个完整操作）。
+- 目标：把 lark 的 shortcut 能力**深度对齐每一个**到 dws，并补钉钉侧系统性能力。
+- 关键认知：lark 的"组合性"多源于飞书 API 细粒度（先查 token→id→再操作）；钉钉 MCP 粗粒度，**lark 的多步在钉钉大量塌缩成 1:1（已被封装层覆盖）**。真正需要"编排"的是「按名解析 ID + 多工具串联 + 跨服务」——这些做成了 smart 层。
+
+---
+
+## 2. 架构与关键文件
+
+```
+internal/shortcut/
+  types.go        # Shortcut / Flag / Risk 声明结构
+  runner.go       # RuntimeContext + mount(编译成cobra) + CallMCP/CallMCPData/Output + 校验/dry-run/风险确认
+  validate.go     # 跨字段校验 helper：MutuallyExclusive/AtLeastOne/ExactlyOne/RangeInt/RequireAll
+  register.go     # Register() / Commands() / All()
+  shortcut_test.go# 框架单测
+  builtin/
+    builtin.go       # blank-import 所有服务包 + smart 包；Commands() 汇总
+    coverage_test.go # ★全量测试：TestAllShortcutsAssemble / TestAllToolLiteralsAreReal / TestAllHaveIntent / TestNoDuplicateCommands
+  <service>/         # 19 个服务包：contact/chat/calendar/todo/doc/drive/mail/wiki/minutes/oa/report/attendance/aitable/sheet/devapp/ding/aisearch/live/devdoc
+    <service>.go     # 该服务的 1:1 封装 shortcut（var + init(){shortcut.Register(...)}）
+  smart/             # ★真·智能层（多步/编排/按名解析/跨服务）
+    resolve.go       # resolveUser(rt,name) 名→userId+消歧；contactUser{userID,name}；extractUsers/userLabels
+    dm.go lookup.go assign.go book.go free.go ... # 每条一个文件
+  usage/          # P2 埋点：recorder.go(记形状不记值) stats.go command.go(dws shortcut list/stats/suggest/add)
+  userdef/        # P2 自定义 shortcut YAML 运行时加载 loader.go
+
+internal/app/legacy.go   # 接线点：newLegacyPublicCommands 里 append builtin.Commands() + userdef.Load()
+internal/app/root.go     # 装配 recordingToolCaller(埋点) + dws shortcut 命令
+internal/helpers/*.go     # ★Ground truth：钉钉真实 MCP tool 名 + 参数（callMCPTool("tool",{...})）
+
+docs/
+  shortcut-plan.md            # 总规划
+  shortcut-p2-design.md       # P2 自动沉淀设计
+  shortcut-report.md / .html  # 综合报告 + GSB
+  shortcut-comparison.html    # 逐条三方对照(dws vs lark vs 原生MCP)
+  shortcut-lark-alignment.md  # ★深度对齐矩阵(lark 361条逐条分析, 49 gap-buildable)
+  shortcut-handoff.md         # 本文件
+scripts/gen_shortcut_comparison.py  # 生成三方对照 HTML
+```
+
+---
+
+## 3. 框架契约（写新 shortcut 必读）
+
+一个 shortcut = 包级 `var X = shortcut.Shortcut{...}` + `func init(){ shortcut.Register(X) }`。
+
+```go
+var SearchUser = shortcut.Shortcut{
+    Service: "contact",        // 顶层命令
+    Command: "+search-user",   // + 前缀，kebab-case
+    Product: "contact",        // MCP server id（默认=Service；注意跨 server，见坑位）
+    Description: "...",         // 一行
+    Intent: "自然语言：做什么/何时用/副作用",  // 每条必填(TestAllHaveIntent 强制)
+    Risk: shortcut.RiskRead,   // Read / Write / HighWrite(删除等，框架二次确认)
+    Flags: []shortcut.Flag{{Name:"query", Type:shortcut.FlagString, Required:true, Desc:"...", Enum:[]string{...}}},
+    Validate: func(rt *shortcut.RuntimeContext) error { return rt.RequireAll("query") }, // 可选
+    Execute: func(rt *shortcut.RuntimeContext) error { ... },
+}
+```
+
+RuntimeContext 方法（`internal/shortcut/runner.go`/`validate.go`）：
+- 读参数：`rt.Str/Bool/Int/StrSlice(name)`、`rt.Changed(name)`
+- **调 MCP 并打印**（终结步，1:1 封装用）：`rt.CallMCP(tool, params) error`（用自身 Product）
+- **调 MCP 拿数据**（多步/投影用，不打印，可跨 server）：`rt.CallMCPData(product, tool, params) (map[string]any, error)`
+- **投影输出**：`rt.Output(payload) error`（吃 --format/--jq/--fields）
+- 校验：`rt.MutuallyExclusive/AtLeastOne/ExactlyOne(flags...)`、`rt.RangeInt(flag,min,max)`、`rt.RequireAll(flags...)`
+- smart 复用：`resolveUser(rt, name) (contactUser, error)`（名→userId+消歧，在 smart/resolve.go）
+
+对标 lark：`CallMCPData`≈`CallAPITyped`；`resolveUser`≈`ResolveOpenIDsTyped`；`rt.Output`≈`OutFormat`；`Validate helper`≈lark 的 MutuallyExclusive/AtLeastOne。
+
+---
+
+## 4. 方法论（怎么高效批量建，屡试不爽）
+
+**核心：多 agent workflow 并行 + helper 为 ground truth + 严格 skip + build/test 门禁。**
+
+1. **每个 shortcut/服务一个 agent**，并行（`parallel(...)`）。
+2. **Ground truth 铁律**：tool 名和参数 key **只能逐字取自 `internal/helpers/<svc>.go` 的真实 `callMCPTool("tool",{params})` 调用点**，严禁编造。agent 必须先 Read+grep helper。
+3. **宁缺勿错**：拿不准的 tool/参数/结构 → **skip 并说明**，不瞎写（已多次证明 agent 会正确 skip，如 mail 无 email 字段）。
+4. **响应字段防御式解析**：返回结构无契约保证 → 多候选 key 探测（result/data/list/items + 字段别名），不硬编码。
+5. **不同 agent 写不同文件**（服务包 vs smart 包，或不同 service 文件）→ 无写冲突；**禁止 agent 改 builtin.go**（我事后统一维护 blank import）。
+6. **落地后统一**：`gofmt -w` → `go build ./...` → shortcut 全量测试 → 命名冲突用 rename 修（如 smart 的 `+approve` 撞 1:1 层 → 改 `+approve-by`）。
+7. **周期性 app 全量回归**（改多个服务文件后）：`go test ./internal/app/...`（~73s）验证接线。
+
+workflow 脚本模板见任意 `~/.claude/.../workflows/scripts/build-smart-*.js` 或 `upgrade-fidelity-*.js`（每次 Workflow 调用都存了盘，可 `{scriptPath}` 复用/改）。
+
+---
+
+## 5. 已完成进展
+
+### 5.1 覆盖层（511 条 1:1 封装 / 19 服务）
+chat89 aitable86 mail43 attendance36 doc33 minutes31 devapp30 sheet29 calendar24 drive24 oa20 todo18 wiki15 contact14 report7 ding7 aisearch3 live1 devdoc1。每条带自然语言 Intent。
+
+### 5.2 智能层（~46 条 `internal/shortcut/smart/`）
+按名操作人/群、多步编排+失败回滚、时间/自身智能、跨服务、钉钉原生编排。已建（举例）：
+`chat +dm/+send-to-group/+broadcast/+group-members/+at-me/+search-msg/+unread-chats`、`contact +lookup/+org/+team/+by-mobile/+dept-members`、`calendar +book(回滚)/+free/+today/+week/+next-event/+invite/+suggest-time/+reschedule/+cancel-event/+respond-event/+find-room`、`todo +assign/+assign-multi/+overdue/+todo-done/+remind/+created-todos`、`minutes +latest-minutes/+action-items/+transcript/+minutes-search/+detail/+replace-batch`、`oa +approve-by/+my-initiated`、`attendance +my-attendance`、`report +report-latest`、`aitable +find-record/+list-tables`、`doc +share-doc/+doc-append`、`wiki +wiki-new-doc`、`drive +find-file`、`mail +search-mail/+unread-mail`。
+
+### 5.3 框架系统性能力（对齐 lark）
+`resolveUser`、`CallMCPData`、`rt.Output`、`Validate×5`。
+
+### 5.4 保真度升级（~64 条封装：CallMCP→CallMCPData+投影+Output，对齐 lark 96% 输出投影）
+覆盖 contact/chat/calendar/todo/doc/drive/mail/wiki/aitable/oa/devapp/attendance/minutes/report/sheet 等服务的列表类命令。
+
+### 5.5 P2 高频自动沉淀（差异化，lark 无）
+埋点(记形状不记值,默认关/opt-in DWS_USAGE_TRACKING=1) → `dws shortcut stats/suggest` → `dws shortcut add` 写 `~/.dws/shortcuts/*.yaml` → 运行时 `userdef.Load()` 编译注册。**闭环端到端跑通。**
+
+### 5.6 深度对齐矩阵
+`docs/shortcut-lark-alignment.md`：逐条分析 lark 361 条 → covered-1to1 144 / no-dingtalk-tool 127 / **gap-buildable 49** / covered-smart 41。
+
+---
+
+## 6. 如何继续（下一步 backlog）
+
+1. **保真度升级剩余列表命令**（还有部分服务的 list 命令仍是裸 CallMCP）：起 `upgrade-fidelity-N` workflow，每服务 agent 挑 1-2 个未升级(`grep 'rt.CallMCP('`)的列表读命令，改成 CallMCPData+投影+Output。范式见 `contact.go` 的 `searchUserProject`/`listRolesProject`。
+2. **补剩余 gap-buildable smart shortcut**（矩阵里 49 个，已建 ~20+）：起 `build-smart-N` workflow（smart 包，不碰服务包）。剩余偏复杂（sheets/base 操作、消息富化、分片下载），谨慎、允许 skip。
+3. **每批**：gofmt→build→shortcut 测试→（改多文件后）app 回归→更新 `docs/shortcut-report.md`。
+4. **收尾**：把 `docs/shortcut-report.md/html` 的计数刷新到最终（部分 §2 测试数字可能还停在旧值 511/456，实际 ~557/~500），重跑 `python3 scripts/gen_shortcut_comparison.py`。
+
+---
+
+## 7. 关键坑位（务必注意）
+
+- **跨 server 路由**：有些 tool 不在本服务 server。已知：contact 花名册 tool 走 `hrmregister`；chat 部分 tool 走 `im`/`bot`；`query_check_record` 走 `attendance-wukong`（不是 attendance）；wiki `create_file` 走 `doc` server。→ 这类必须用 `rt.CallMCPData("<真实server>", ...)`，不能用 `rt.CallMCP`（它按 shortcut.Product 路由会打错 server）。判断依据：helper 里是 `callMCPToolOnServer("<server>", ...)`。
+- **命名冲突**：smart 命令别撞 1:1 层（如 `+approve`→用 `+approve-by`，`+freebusy`→用 `+free`）。`TestNoDuplicateCommands` 会抓。
+- **参数别名**：aitable 查询关键词是 `keyword`（不是 query）；todo 建待办用嵌套 `PersonalTodoCreateVO`；日程 create/update 时间是 ISO 字符串，而 list/busy 是毫秒——一切以 helper 调用点为准。
+- **中文 const tool 名**：minutes 录音 tool 是中文 `"执行听记指令-发起AI听记录音"`（const `listeningNoteCmdTool`）——真实，别当编造。
+- **测试真机验证**：token 有时效（`dws auth status` 看 expires），过期会报错非代码问题。真机跑命令时第一次有 catalog 发现 banner（stderr），结果 JSON 在后面，别用 `head` 截断。
+- **工具标签格式**（给 AI：本会话我多次误用错标签导致工具调用失败——务必用正确的 function-call 格式）。
+
+---
+
+## 8. 验证命令速查
+
+```bash
+cd /Users/dennis/Projects/dingtalk-workspace/dingtalk-workspace-cli
+go build ./...                                    # 编译
+gofmt -l internal/shortcut/                        # 格式(应空)
+go test ./internal/shortcut/...                    # shortcut 全量(含 assemble/tool-real/intent/no-dup)
+go test ./internal/app/...                         # app 回归(~73s，改服务文件后必跑)
+go test ./internal/shortcut/builtin/ -run TestAllShortcutsAssemble -v 2>&1 | grep shortcuts=  # 看总数
+
+# 真机(需登录 dws auth login)
+DWS_USAGE_TRACKING=0 dws contact +lookup --name <真实姓名>   # 智能多步示范
+DWS_USAGE_TRACKING=0 dws calendar +today                     # 时间智能
+DWS_USAGE_TRACKING=0 dws contact +search-user --query <名>   # 投影输出示范
+```
+
+测试口径：`TestAllShortcutsAssemble` = 假 Caller 拦截，给每条命令(含写/删)喂合成参数、走完解析→校验→组装 MCP 调用、断言 tool 真实无 panic（零副作用全量验证）。`TestAllToolLiteralsAreReal` = 所有 CallMCP tool 名比对 helper ground truth 防编造。
+
+---
+
+## 9. 未提交提醒
+
+所有工作在 `feature/shortcut`，**未 commit**。建议尽快分语义化 commit 留存（框架 / 511封装 / smart层 / 保真度升级 / P2 / 文档）。

docs/shortcut-p2-design.md

@@ -0,0 +1,204 @@
+# DWS Shortcut P2 详细设计 — 高频场景自动沉淀为自定义 Shortcut
+
+> 前置：P1 已交付静态声明式 shortcut 框架（`internal/shortcut/`），见 `docs/shortcut-plan.md`。
+> P2 目标：**观察用户高频使用 → 主动建议 → 一键沉淀为可复用的自定义 shortcut**，
+> 让 CLI 越用越顺手。这是 dws 相对 larksuite/cli 的差异化能力。
+
+## 0. 体验闭环（一句话）
+
+```
+用户反复敲 dws chat send_message --json '{"open_conversation_id":"cid_x","text":"..."}'
+        │  （每次执行被静默记录到 ~/.dws/usage.jsonl）
+        ▼
+第 N 次后，dws 主动提示：
+  💡 你已 12 次向「项目群」发消息，是否沉淀为 `dws chat +notify-team`？[y/N]
+        │  y
+        ▼
+写入 ~/.dws/shortcuts/chat.notify-team.yaml
+        ▼
+之后：dws chat +notify-team --text "发布完成"   ← 参数从一堆 JSON 收敛成一个 flag
+```
+
+## 1. 总体架构
+
+复用 P1 的 `Shortcut` 模型作为「编译目标」，新增四个部件：
+
+| 部件 | 位置（建议） | 职责 |
+|------|-------------|------|
+| Usage 埋点 | `internal/shortcut/usage/recorder.go` | 每次 MCP 调用后追加一条 usage 记录 |
+| 模式挖掘 | `internal/shortcut/usage/miner.go` | 聚合 usage → 高频候选 + 打分 |
+| 主动提示 | `internal/shortcut/usage/nudge.go` | 命中候选时在命令收尾处提示 |
+| YAML 加载 | `internal/shortcut/userdef/loader.go` | 扫描 `~/.dws/shortcuts/*.yaml` → 编译成 `Shortcut` → 注册 |
+| 管理命令 | `internal/shortcut/usage`（cobra） | `dws shortcut list/suggest/add/rm/stats` |
+
+数据流：
+
+```
+CallMCP ──► recorder.Append(usage)                 [写侧，热路径，必须极轻]
+                                    ┌─► miner.TopCandidates()  [读侧，suggest 时才算]
+~/.dws/usage.jsonl ─────────────────┤
+                                    └─► nudge (命令收尾抽样触发)
+~/.dws/shortcuts/*.yaml ──► loader.Compile() ──► shortcut.Register()  [启动时]
+```
+
+## 2. Usage 埋点
+
+### 2.1 采集点（choke point）
+
+**首选**：装饰 `executor.Runner` / `edition.ToolCaller`。`internal/app/tool_caller_adapter.go`
+的 `CallTool(ctx, productID, toolName, args)` 是**所有 MCP 调用的唯一必经点**，天然拿到
+`(product, tool, args)` 三元组。用装饰器包一层即可，零侵入命令层：
+
+```go
+type recordingCaller struct{ inner edition.ToolCaller }
+func (r recordingCaller) CallTool(ctx, product, tool string, args map[string]any) (*edition.ToolResult, error) {
+    res, err := r.inner.CallTool(ctx, product, tool, args)
+    usage.Append(product, tool, args, err == nil)   // 异步/带 recover，绝不影响主流程
+    return res, err
+}
+```
+
+> 注意：P1 的 shortcut 也走这条 `CallMCP → helpers → deps.Caller`，所以内建 shortcut 的
+> 使用同样会被记录，可用于「哪些内建 shortcut 最受欢迎」的洞察。
+
+### 2.2 记录内容（**隐私优先：记形状不记值**）
+
+`~/.dws/usage.jsonl`，每行一条：
+
+```json
+{
+  "ts": "2026-07-08T10:12:33+08:00",
+  "product": "chat",
+  "tool": "send_message",
+  "arg_keys": ["open_conversation_id", "text"],
+  "const_args": {"open_conversation_id": "cid_x"},
+  "ok": true
+}
+```
+
+- `arg_keys`：参数键集合（排序），用于识别「同一种调用形状」。
+- `const_args`：**仅收敛出的「疑似固定值」**（见 §3 挖掘时判定），写入时不保证脱敏，
+  因此需要一层白名单/黑名单：`text/content/body/message` 等自由文本字段**永不入库**，
+  只保留看起来像 ID/枚举的短值（长度阈值 + 无空格 + 非多行）。
+- 绝不记录：token、手机号、邮箱、文件内容、消息正文。用 `internal/logging/redact.go`
+  已有的脱敏能力复核。
+
+### 2.3 热路径约束
+
+- 追加写用 `O_APPEND`，单行 < 1KB；失败静默（`recover` + debug 日志），**绝不阻断命令**。
+- 文件滚动：超过 N 行（如 5000）或 M 天，截断/归档，避免无限增长。
+- 开关：默认关闭（opt-in），环境变量 `DWS_USAGE_TRACKING=1` 开启（本地遥测即便只记形状也不应未经用户同意默认开启）。
+  首次启用时在 `dws` 首跑给一次性告知（尊重知情）。
+
+## 3. 模式挖掘
+
+`dws shortcut suggest` 触发（也被 nudge 复用）。算法：
+
+1. 读 usage.jsonl，按 `(product, tool, arg_keys)` 分桶。
+2. 对每桶：
+   - `count` = 出现次数；低于阈值（默认 5）直接丢弃。
+   - 对每个 arg_key，统计其值的分布：某值占比 ≥ 80% → 判定为**固定值**（进 `const_args`）；
+     否则判定为**可变参数**（沉淀后成为 flag）。
+   - `recency` = 最近一次使用距今；越近权重越高。
+3. 打分 `score = count * log(distinct_days+1) * recencyDecay`，取 TopN。
+4. 生成候选 `Candidate{product, tool, fixed{...}, varFlags[...], score, samples}`。
+
+输出示例（`dws shortcut suggest --format table`）：
+
+```
+候选  | 命令建议            | 依据                     | 固定参数           | 可变flag
+#1    | chat +notify-team  | 12 次 / 近 3 天          | open_conv=cid_x    | text
+#2    | doc +new-agenda    | 7 次 / 近 5 天           | template=agenda    | title
+```
+
+## 4. 主动提示（nudge）
+
+- **时机**：命令成功收尾时（root `PersistentPostRunE`），**抽样**触发（如每 N 次调用或每次
+  命中新达标候选时），避免打扰。仅在 TTY 交互态提示；非交互（Agent/管道/`--yes`）**不提示**。
+- **频控**：同一候选提示过一次被拒后，冷却期内不再提示（记 `~/.dws/shortcuts/.declined`）。
+- **交互**：
+  ```
+  💡 检测到高频操作：你已 12 次向同一会话发消息。
+     沉淀为快捷指令  dws chat +notify-team --text "..."  ？
+     [y] 沉淀  [n] 以后再说  [d] 不再提示此项
+  ```
+- y → 走 §5 生成 YAML；命名默认 `+<tool 去下划线的动宾>`，允许用户改名。
+
+## 5. 自定义 Shortcut：YAML 格式与运行时加载
+
+### 5.1 YAML schema（与 P1 `Shortcut` 一一对应）
+
+`~/.dws/shortcuts/chat.notify-team.yaml`：
+
+```yaml
+version: 1
+service: chat
+command: "+notify-team"
+product: chat
+description: "发消息到 项目群（自动沉淀于 2026-07-08）"
+risk: write                    # 默认 read；send 类判定为 write
+source: auto                   # auto=沉淀 / manual=手写
+flags:
+  - name: text
+    type: string
+    required: true
+    desc: 消息内容
+execute:
+  tool: send_message
+  bind:                        # 参数绑定：常量 + ${flag} 模板
+    open_conversation_id: "cid_x"
+    text: "${text}"
+```
+
+### 5.2 编译与注册
+
+`userdef.Compile(yaml)` → `shortcut.Shortcut`，其 `Execute` 由 `bind` 生成：
+遍历 `bind`，`${flag}` 用 `rt.Str(flag)` 填充，常量原样，组装 params 后 `rt.CallMCP(tool, params)`。
+完全复用 P1 的 runner，不新增执行路径。
+
+加载时机：`legacy.go` 装配点，在 `builtin.Commands()` 之后追加 `userdef.Commands()`，
+一起 merge。复用 `internal/plugin/loader.go` 已验证的「扫 `~/.dws/` 目录 + 挂 cobra」模式。
+
+### 5.3 冲突与优先级
+
+- 自定义 shortcut 命令名若与内建 shortcut / helper 冲突：**内建优先**，自定义重命名或跳过并告警。
+- `+` 前缀天然与 helper leaf 区分，冲突面小。
+
+## 6. 管理命令面
+
+```
+dws shortcut list                 # 列出内建 + 自定义 shortcut
+dws shortcut suggest [--min N]    # 展示高频候选（不写入）
+dws shortcut add <candidate|--from-last>   # 交互式/从最近一次调用沉淀
+dws shortcut rm <service> <+cmd>  # 删除自定义 shortcut
+dws shortcut stats                # usage 统计概览
+```
+
+`dws shortcut` 本身作为一个新的顶层 utility 命令注册（对齐 `dws plugin`）。
+
+## 7. 安全与隐私边界（红线）
+
+1. **值不入库**：自由文本/正文/凭证一律不记；`const_args` 仅短 ID/枚举，且过 redact 复核。
+2. **可关可清**：`DWS_USAGE_TRACKING=0` 关闭；`dws shortcut stats --purge` 清空 usage。
+3. **执行白名单**：自定义 shortcut 的 `execute.tool` 必须解析到合法 MCP server（过
+   `internal/security` endpoint 白名单），禁止指向任意 endpoint。
+4. **不自动执行**：沉淀只生成命令定义，**绝不**自动发起写操作；写类 shortcut 仍受 P1 的
+   risk 确认约束。
+5. **知情**：首次开启埋点一次性告知；提示可永久关闭。
+
+## 8. 实现顺序（P2 分步，便于 loop 推进）
+
+- P2-1 usage 埋点：`recordingCaller` 装饰器 + `usage.Append` + jsonl 写 + 开关 + 脱敏白名单。
+- P2-2 `dws shortcut stats` / `list`：先让数据可见，验证埋点质量。
+- P2-3 miner + `dws shortcut suggest`：离线挖掘与打分。
+- P2-4 userdef YAML 加载 + `Compile` + 注册 + 冲突处理（打通「手写 YAML 也能用」）。
+- P2-5 `dws shortcut add`（从候选/最近调用沉淀）。
+- P2-6 nudge 主动提示（最后做，最谨慎，默认保守频控）。
+
+## 9. 待决策点（需产品确认）
+
+1. 埋点默认开还是默认关？→ **修订后：默认关（opt-in）+ 开启后首跑一次性告知**（原设计默认开，反思后改为 opt-in：自主 agent 不应单方面默认开本地遥测）
+   （`DWS_USAGE_TRACKING=0` / 配置项关闭；`dws shortcut stats --purge` 清空）。实现时以此为准。
+2. `const_args` 允许记录的字段白名单粒度？（保守起步：只记形如 `*_id/*Id/type/status` 的短值）
+3. nudge 触发频率与渠道？（建议：仅 TTY、命中新候选时、每候选一生仅一次）
+4. 自定义 shortcut 是否需要跨设备同步？（v1 先本地 `~/.dws/`，同步留待后续）

docs/shortcut-plan.md

@@ -0,0 +1,127 @@
+# DWS Shortcut 能力 — 总体规划
+
+> 目标：为 dws 引入一套 **声明式高保真命令（Shortcut）** 能力，对齐 larksuite/cli 的 `+command`
+> 体验（如 `lark-cli contact +search-user`），并在此之上做 dws 差异化：**基于用户高频使用场景，
+> 主动把常用操作沉淀为自定义 shortcut**。
+
+## 1. 背景与动机
+
+dws 当前的命令有三类来源：
+
+1. **MCP 运行时动态发现** —— `dws mcp <service> <tool> --json '{...}'`，通用但裸、参数需手拼 JSON。
+2. **`internal/helpers/` 产品命令** —— 手写 cobra 命令，体验好但每个都从零写、缺统一框架。
+3. **`internal/registry/recipes.yaml`** —— 多步工作流的静态描述。
+
+痛点：想新增一个「精选、参数友好、带 dry-run/format/身份」的单命令，只能手写 helper，
+没有统一的声明式框架，重复劳动多、一致性差。larksuite 的 shortcut 框架正好解决这一层。
+
+## 2. 与 larksuite/cli 的架构差异（关键）
+
+| 维度 | larksuite/cli | dws-cli |
+|------|---------------|---------|
+| 命令来源 | 静态硬编码 Go shortcut | MCP 运行时动态发现 + helpers |
+| 调用底座 | Lark SDK 直连 API | MCP JSON-RPC（`executor.Runner`） |
+| 精选命令层 | `shortcuts/`（200+ 声明式） | `internal/helpers/`（手写 cobra） |
+| 全局 flag | 框架注入 | root 已内建 `--format/--dry-run/--jq/--yes/--fields/--profile` |
+
+**结论**：不能直接搬代码。移植的是 shortcut 的**声明式设计**，执行底座换成 dws 的
+`executor.Runner`，全局能力复用 dws 已有的 output/safety/auth。
+
+## 3. 分期目标
+
+### P1 — 静态声明式框架（本期，正在做）
+- 新建独立模块 `internal/shortcut/`（零侵入现有 helpers）。
+- `types.go`：`Shortcut` / `Flag` / `RuntimeContext` 声明层。
+- `runner.go`：把 `Shortcut` 编译成 `*cobra.Command`，串起 flag 注册 → 校验 → dry-run →
+  `executor.Runner.Run` → `output.WriteCommandPayload`。
+- `register.go`：按 service 分组产出命令，在 `internal/app/legacy.go` 装配点 merge 进命令树。
+- 样板命令 `contact +search-user`：打通 MCP 执行 / format / dry-run / 身份，作为后续命令模板。
+- 交付判据：`dws contact +search-user --help`、`--dry-run` 正常；`go build` / `go test` 通过。
+
+### P2 — 高频场景自动沉淀（后续，先设计再实现）
+- **使用埋点**：命令执行入口记录 `~/.dws/usage.jsonl`（只记参数形状，不记敏感值）。
+- **模式挖掘**：`dws shortcut suggest` 聚合高频 `(service, tool, 固定参数组合)`。
+- **主动沉淀**：命中候选时提示用户，一键写入 `~/.dws/shortcuts/*.yaml`（声明式，与 P1 结构对应）。
+- **运行时加载**：`register.go` 额外扫描 `~/.dws/shortcuts/*.yaml` 动态注册，复用
+  `internal/plugin/loader.go` 已验证的「从 `~/.dws/` 加载并挂 cobra 命令」模式。
+
+## 4. 落地方式（P1）
+
+采用**独立模块 + 装配点 merge**，不改 helpers 内部：
+
+```
+internal/shortcut/
+  types.go            # Shortcut / Flag / RuntimeContext
+  runner.go           # 声明式→cobra 编译 + 执行管道
+  register.go         # Commands(runner) []*cobra.Command，按 service 分组
+  contact/
+    search_user.go    # 样板：var SearchUser = shortcut.Shortcut{...}
+    shortcuts.go      # Shortcuts() []shortcut.Shortcut
+```
+
+接线：`internal/app/legacy.go: newLegacyPublicCommands` 里，
+`helpers.NewPublicCommands(runner)` 之后追加 `shortcut.Commands(runner)`，
+一起走 `mergeTopLevelCommands`（同名 service 命令自动合并，`+xxx` 作为其子命令）。
+
+复用点：
+- 执行：`executor.NewHelperInvocation` + `runner.Run`（与 helper 完全一致的调用路径）。
+- 输出：`output.WriteCommandPayload(cmd, resp, output.FormatJSON)`（自动吃 root 的 `--format/--jq/--fields`）。
+- dry-run：读 root `--dry-run`，置 `Invocation.DryRun`，由 runner 返回请求预览。
+- 身份/安全：复用 `--profile`、`internal/safety`（高风险 `--yes` 确认）。
+
+## 5. Shortcut 声明模型（草案）
+
+```go
+type Shortcut struct {
+    Service     string   // "contact"       → 顶层命令
+    Command     string   // "+search-user"  → 子命令（保留 + 前缀，对齐 larksuite）
+    Description string
+    Risk        string   // read | write | high-risk-write
+    Flags       []Flag
+    Validate    func(*RuntimeContext) error
+    Execute     func(*RuntimeContext) error   // 必填；内部调 rt.CallMCP(...)
+}
+
+type Flag struct {
+    Name, Type, Default, Desc string
+    Required bool
+    Enum     []string
+}
+```
+
+`RuntimeContext` 给 Execute 提供：flag 读取（`Str/Bool/Int/StrSlice/Changed`）、
+`CallMCP(product, tool, params)`（内部 `runner.Run`）、`Output(payload)`、`DryRun()`。
+
+## 6. 风险与边界
+
+- **与 `dws mcp` 通道的边界**：shortcut 是「人工精选的薄封装」，不替代通用 MCP 通道；
+  一个 tool 可以既能 `dws mcp` 直调，也能有 shortcut。
+- **自定义 shortcut 安全（P2）**：YAML 的 `execute` 若允许任意 MCP 调用，需过
+  `internal/security` 的 endpoint 白名单，且沉淀的参数值要脱敏。
+- **命名冲突**：`+` 前缀天然与现有 leaf 命令区分，降低与 helper 命令的冲突面。
+- **edition 差异**：oss / enterprise 的可用 service 不同，注册时按 edition 过滤（后续接入）。
+
+## 7. 进度看板
+
+- [x] P1-1 types.go — `Shortcut` / `Flag` / `Risk` 声明层
+- [x] P1-2 runner.go — `RuntimeContext` + `mount` 编译 + 校验/确认/dry-run；`CallMCP` 委托 `helpers.CallMCPToolOnServer`（复用错误分类/输出/dry-run）
+- [x] P1-3 register.go + `internal/shortcut/contact/search_user.go` + `builtin` 聚合包
+- [x] P1-4 接线 `legacy.go`（append 到 `mergeTopLevelCommands`）+ build/test 全绿
+- [ ] P2 设计文档（进行中）
+
+### P1 落地实证（已验证）
+
+- `dws contact +search-user --help`：命令挂载，继承全局 `--format/--dry-run/--jq/...`。
+- 必填校验：不传 `--query` → 结构化 validation 错误。
+- `--dry-run`：走 helpers 路径输出 `[DRY-RUN]` 预览（tool + 参数）。
+- 命令树 merge：`+search-user` 与现有 `user/dept/label/relation` 共存，`contact user search` 未受影响。
+- 测试：`internal/shortcut` 单测通过；`internal/app` 全量回归通过。
+
+### P1 关键决策记录
+
+- **执行底座复用 helpers 而非裸 `runner.Run`**：`CallMCP` 委托 `helpers.CallMCPToolOnServer(product, tool, params)`，
+  一步获得错误分类（auth/PAT/业务）+ 格式化输出 + dry-run，避免重造劣质输出层。代价是
+  `internal/shortcut → internal/helpers` 的单向依赖（无环）。后续若要 shortcut 做多调用编排/输出重塑，
+  再补一个返回原始 payload 的 `CallMCPRaw`。
+- **避免 import 环**：service 包（contact）import 核心 `shortcut` 包并在 `init()` 注册；
+  `builtin` 聚合包 blank-import 各 service 包；`app` 只依赖 `builtin`。

docs/shortcut-real-test-followups.md

@@ -0,0 +1,168 @@
+# Shortcut 真实测试跟进清单
+
+生成时间：`2026-07-15T16:58:39`
+
+来源：`docs/shortcut-real-read-results.json` 与 `docs/shortcut-real-write-results.json`。
+
+口径：记录真实后端测试中需要继续定位的 case，用于 CR 和问题分派；Agent 使用入口以公开 shortcut catalog 和产品 skill 为准。
+
+总计：156 条。
+
+| # | suite | shortcut | risk | status | category | fixability | 处理依据 |
+|---:|---|---|---|---|---|---|---|
+| 1 | read | `aitable +base-get-primary-doc-id` | read | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 2 | read | `aitable +chart-share-get` | read | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 3 | read | `aitable +dashboard-share-get` | read | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 4 | read | `aitable +export-data` | read | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 5 | read | `aitable +record-primary-doc-get` | read | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 6 | read | `aitable +role-get` | read | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 7 | read | `aitable +workflow-get` | read | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 8 | read | `aitable +workflow-list` | read | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 9 | read | `attendance +get-class` | read | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 10 | read | `attendance +get-global-setting` | read | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 11 | read | `attendance +get-group` | read | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 12 | read | `attendance +get-group-filtered` | read | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 13 | read | `attendance +get-leave-balance` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 14 | read | `attendance +list-report-columns` | read | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 15 | read | `attendance +query-report-leave` | read | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 16 | read | `calendar +find-room` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 17 | read | `calendar +room-find` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 18 | read | `chat +category-list-conversations` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 19 | read | `chat +chat-get-by-id` | read | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 20 | read | `chat +chat-members-get` | read | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 21 | read | `chat +chat-messages` | read | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 22 | read | `chat +messages-list` | read | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 23 | read | `chat +messages-resource-url` | read | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 24 | read | `chat +search-msg` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 25 | read | `chat +thread-replies` | read | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 26 | read | `contact +get-roster` | read | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 27 | read | `contact +list-roster-fields` | read | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 28 | read | `devapp +credentials-get` | read | held | held | manual-approval | 高风险或无安全目标，需人工逐项授权后执行。 |
+| 29 | read | `drive +download` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 30 | read | `drive +list` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 31 | read | `minutes +action-items` | read | real-error | missing-real-minutes-fixture | not-cli-fixable-without-fixture | 当前账号没有满足条件的妙记/听记或录制会话；需准备真实会议产物后复测。 |
+| 32 | read | `minutes +latest-minutes` | read | real-error | missing-real-minutes-fixture | not-cli-fixable-without-fixture | 当前账号没有满足条件的妙记/听记或录制会话；需准备真实会议产物后复测。 |
+| 33 | read | `minutes +minutes-search` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 34 | read | `minutes +transcript` | read | real-error | missing-real-minutes-fixture | not-cli-fixable-without-fixture | 当前账号没有满足条件的妙记/听记或录制会话；需准备真实会议产物后复测。 |
+| 35 | read | `oa +done-approvals` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 36 | read | `oa +pending` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 37 | read | `report +report-latest` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 38 | read | `todo +due-today` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 39 | read | `todo +related-tasks` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 40 | read | `wiki +node-list` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 41 | read | `wiki +resolve-space` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 42 | read | `wiki +space-list` | read | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 43 | write | `aitable +advperm-disable` | high-risk-write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 44 | write | `aitable +advperm-enable` | write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 45 | write | `aitable +attachment-upload` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 46 | write | `aitable +base-copy` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 47 | write | `aitable +base-delete` | high-risk-write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 48 | write | `aitable +base-update` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 49 | write | `aitable +chart-delete` | high-risk-write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 50 | write | `aitable +chart-share-update` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 51 | write | `aitable +chart-update` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 52 | write | `aitable +dashboard-arrange` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 53 | write | `aitable +dashboard-delete` | high-risk-write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 54 | write | `aitable +dashboard-share-update` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 55 | write | `aitable +dashboard-update` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 56 | write | `aitable +field-delete` | high-risk-write | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 57 | write | `aitable +field-update` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 58 | write | `aitable +form-delete` | high-risk-write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 59 | write | `aitable +form-field-hide` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 60 | write | `aitable +form-field-update` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 61 | write | `aitable +form-share-update` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 62 | write | `aitable +form-update` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 63 | write | `aitable +import-data` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 64 | write | `aitable +import-upload` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 65 | write | `aitable +record-delete` | high-risk-write | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 66 | write | `aitable +record-primary-doc-create` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 67 | write | `aitable +record-update` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 68 | write | `aitable +record-upsert` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 69 | write | `aitable +role-create` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 70 | write | `aitable +role-delete` | high-risk-write | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 71 | write | `aitable +role-update` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 72 | write | `aitable +section-create` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 73 | write | `aitable +section-delete` | high-risk-write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 74 | write | `aitable +section-move-node` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 75 | write | `aitable +section-rename` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 76 | write | `aitable +section-reorder` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 77 | write | `aitable +table-delete` | high-risk-write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 78 | write | `aitable +table-update` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 79 | write | `aitable +view-delete` | high-risk-write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 80 | write | `aitable +view-duplicate` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 81 | write | `aitable +view-lock` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 82 | write | `aitable +view-set-fill-color-rule` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | 后端/MCP 服务返回内部错误；CLI 无法直接修复，但报告保留 trace/stdout 供服务端排查。 |
+| 83 | write | `aitable +view-set-frozen-cols` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 84 | write | `aitable +view-set-row-height` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 85 | write | `aitable +view-update` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 86 | write | `aitable +workflow-disable` | high-risk-write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 87 | write | `aitable +workflow-enable` | write | real-error | missing-real-aitable-fixture | not-cli-fixable-without-fixture | AI 表格命令需要真实 Base/Table/View/Record 等资源；安全负向 ID 只能验证调用链，不能让后端成功。 |
+| 88 | write | `attendance +boss-check` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 89 | write | `attendance +create-class` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 90 | write | `attendance +create-group` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 91 | write | `attendance +import-schedule` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 92 | write | `attendance +save-leave-balance` | write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 93 | write | `attendance +update-class` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 94 | write | `attendance +update-group` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 95 | write | `attendance +update-group-members` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 96 | write | `attendance +update-leave-type` | write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 97 | write | `calendar +respond-event` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 98 | write | `chat +category-add-conversation` | write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 99 | write | `chat +category-remove-conversation` | write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 100 | write | `chat +chat-add-bot` | write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 101 | write | `chat +chat-audit-join` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | dry-run 已证明 CLI 装配了 applicantUid/inviterUid；真实后端仍报 applicantUid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 102 | write | `chat +chat-mute-member` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 103 | write | `chat +chat-quit` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 104 | write | `chat +chat-remove-bot` | high-risk-write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 105 | write | `chat +chat-role-remove` | high-risk-write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 106 | write | `chat +chat-role-remove-user` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 107 | write | `chat +chat-transfer-owner` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 108 | write | `chat +chat-update-icon` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 109 | write | `chat +chat-update-settings` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 110 | write | `chat +conversation-clear-messages` | high-risk-write | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 111 | write | `chat +conversation-clear-red-point` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 112 | write | `chat +conversation-hide` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 113 | write | `chat +conversation-mark-read` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 114 | write | `chat +conversation-mark-unread` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 115 | write | `chat +conversation-mute` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 116 | write | `chat +conversation-mute-at-all` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 117 | write | `chat +conversation-mute-red-envelope` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 118 | write | `chat +conversation-set-top` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 119 | write | `chat +messages-add-emoji` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 120 | write | `chat +messages-add-text-emotion` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 121 | write | `chat +messages-batch-recall-by-bot` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 122 | write | `chat +messages-batch-send-by-bot` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 123 | write | `chat +messages-combine-forward` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 124 | write | `chat +messages-create-text-emotion` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 125 | write | `chat +messages-forward` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 126 | write | `chat +messages-forward-topic` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 127 | write | `chat +messages-recall` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 128 | write | `chat +messages-recall-by-bot` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 129 | write | `chat +messages-remove-emoji` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 130 | write | `chat +messages-remove-text-emotion` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 131 | write | `chat +messages-send-by-bot` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 132 | write | `chat +messages-send-card` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | dry-run 已证明 CLI 装配了 receiverUid；真实后端仍报 receiverUid/openConversationId 为空，优先按 MCP schema/服务端字段映射问题处理。 |
+| 133 | write | `chat +messages-set-pin` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 134 | write | `chat +messages-set-top` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 135 | write | `chat +messages-unset-pin` | write | real-error | backend-or-mcp-error | not-cli-fixable-first | fake MCP 已证明 CLI 已装配会话 ID 字段；真实后端仍报 openConversationId/openCid/cid 缺失，优先按 MCP schema/服务端字段映射问题处理。 |
+| 136 | write | `chat +messages-unset-top` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 137 | write | `devapp +event-subscribe` | write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 138 | write | `devapp +event-unsubscribe` | high-risk-write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 139 | write | `devapp +permission-add` | write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 140 | write | `devapp +permission-remove` | high-risk-write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 141 | write | `devapp +robot-config` | write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 142 | write | `devapp +robot-disable` | high-risk-write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 143 | write | `devapp +robot-enable` | write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 144 | write | `devapp +security-config` | write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 145 | write | `devapp +version-create` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 146 | write | `devapp +version-publish` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 147 | write | `ding +send-by-message` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 148 | write | `doc +comment-create-inline` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 149 | write | `doc +template-apply` | write | real-error | auth-or-permission | not-cli-fixable | 真实账号、应用 scope 或资源权限不足；CLI 只能如实暴露，不能在本仓库内修复权限。 |
+| 150 | write | `minutes +record-pause` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 151 | write | `minutes +record-resume` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 152 | write | `minutes +record-stop` | write | real-error | input-or-business-validation | test-input-or-backend-rule | 命令已真实进入本地/后端校验；若该项仍使用安全负向输入，则失败符合预期；若使用真实 fixture 仍失败，再作为 CLI bug 处理。 |
+| 153 | write | `oa +approve-by` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 154 | write | `wiki +node-copy` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 155 | write | `wiki +node-move` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |
+| 156 | write | `wiki +wiki-new-doc` | write | real-error | missing-real-resource | not-cli-fixable-without-fixture | 真实测试使用的资源/单据/消息/群/文档不存在；需要准备对应 fixture 后才能期望成功，不属于 shortcut 参数投影错误。 |

docs/shortcut-report.html

@@ -0,0 +1,186 @@
+<!DOCTYPE html>
+<html lang="zh-CN">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<title>DWS Shortcut 完成情况报告</title>
+<style>
+  :root{--bg:#0f1420;--card:#161d2c;--ink:#e6edf6;--muted:#93a1b5;--line:#26314a;
+    --blue:#4f9cff;--green:#3fb950;--yellow:#d5a429;--red:#f25c5c;--accent:#6ea8fe;--purple:#a371f7}
+  *{box-sizing:border-box}
+  body{margin:0;background:linear-gradient(180deg,#0d1220,#0f1420);color:var(--ink);
+    font:15px/1.7 -apple-system,BlinkMacSystemFont,"Segoe UI","PingFang SC","Microsoft YaHei",sans-serif;padding:0 0 80px}
+  .wrap{max-width:1080px;margin:0 auto;padding:0 22px}
+  header{padding:52px 22px 28px;text-align:center;border-bottom:1px solid var(--line);
+    background:radial-gradient(1200px 300px at 50% -60px,rgba(79,156,255,.16),transparent)}
+  h1{font-size:29px;margin:0 0 8px;letter-spacing:.5px}
+  .sub{color:var(--muted);font-size:14px}
+  .stats{display:flex;gap:13px;justify-content:center;flex-wrap:wrap;margin:26px 0 4px}
+  .stat{background:var(--card);border:1px solid var(--line);border-radius:14px;padding:15px 20px;min-width:118px}
+  .stat .n{font-size:27px;font-weight:700;color:var(--accent)}
+  .stat .l{color:var(--muted);font-size:12.5px;margin-top:2px}
+  h2{font-size:21px;margin:44px 0 14px;padding-bottom:8px;border-bottom:1px solid var(--line)}
+  h2 .ico{color:var(--accent);margin-right:8px}
+  h3{font-size:15.5px;margin:22px 0 9px;color:var(--accent)}
+  table{width:100%;border-collapse:collapse;margin:12px 0;font-size:13.5px;background:var(--card);
+    border:1px solid var(--line);border-radius:10px;overflow:hidden}
+  th,td{padding:9px 12px;text-align:left;border-bottom:1px solid var(--line);vertical-align:top}
+  th{background:#1b2536;color:var(--muted);font-weight:600;font-size:12.5px}
+  tr:last-child td{border-bottom:none}
+  td.c,th.c{text-align:center}
+  .num{color:var(--accent);font-weight:700;text-align:center}
+  .g{color:var(--green);font-weight:700}.s{color:var(--yellow);font-weight:700}.b{color:var(--red);font-weight:700}
+  .ok{color:var(--green)}.star{color:var(--yellow)}
+  code{background:#0c1120;border:1px solid var(--line);border-radius:5px;padding:1px 6px;font-size:12.5px;color:#cfe0ff}
+  .card{background:var(--card);border:1px solid var(--line);border-radius:12px;padding:15px 18px;margin:13px 0}
+  .two{display:grid;grid-template-columns:1fr 1fr;gap:14px}
+  .layer{border-radius:12px;padding:16px 18px}
+  .l-wrap{background:linear-gradient(180deg,rgba(79,156,255,.08),transparent);border:1px solid #234b6b}
+  .l-smart{background:linear-gradient(180deg,rgba(163,113,247,.10),transparent);border:1px solid #4a3a6b}
+  .layer h3{margin-top:0}
+  .pill{display:inline-block;background:#12283a;color:#7fc6ff;border:1px solid #234b6b;border-radius:6px;padding:1px 7px;font-size:12px;margin:2px 3px 2px 0}
+  .flow{display:flex;align-items:center;gap:7px;flex-wrap:wrap;font-size:13px;color:var(--muted)}
+  .flow b{color:var(--ink)}.flow .arw{color:var(--purple)}
+  .concl{background:linear-gradient(90deg,rgba(63,185,80,.10),transparent);border-left:3px solid var(--green);padding:14px 18px;border-radius:8px;margin-top:16px}
+  .keyfind{background:linear-gradient(90deg,rgba(213,164,41,.10),transparent);border-left:3px solid var(--yellow);padding:14px 18px;border-radius:8px;margin:14px 0}
+  footer{color:var(--muted);text-align:center;font-size:12.5px;margin-top:40px}
+  a{color:var(--accent)}
+</style>
+</head>
+<body>
+<header>
+  <h1>DWS Shortcut 完成情况报告</h1>
+  <div class="sub">对齐基准 larksuite/cli　·　执行底座 钉钉 MCP　·　随 loop 持续更新</div>
+  <div class="stats">
+    <div class="stat"><div class="n">366</div><div class="l">shortcut 总数</div></div>
+    <div class="stat"><div class="n">298</div><div class="l">1:1 封装层</div></div>
+    <div class="stat"><div class="n">68</div><div class="l">真·智能层</div></div>
+    <div class="stat"><div class="n">16</div><div class="l">覆盖服务</div></div>
+    <div class="stat"><div class="n">0</div><div class="l">失败/panic/编造</div></div>
+  </div>
+</header>
+<div class="wrap">
+
+  <h2><span class="ico">①</span>做了什么：两个层次</h2>
+  <p>诚实区分——shortcut 分两层，价值定位不同，不混为一谈。</p>
+  <div class="two">
+    <div class="layer l-wrap">
+      <h3>🔵 1:1 封装层 · 298 条</h3>
+      <div style="color:var(--muted);font-size:13.5px">一个 shortcut ≡ 一个 MCP tool。把裸 <code>dws mcp &lt;svc&gt; &lt;tool&gt; --json '{…}'</code> 收敛成命名 flag，附校验/风险确认/Intent。</div>
+      <div style="margin:10px 0"><b>价值</b>：DX 与 AI-agent 可发现性，<b>不是新能力</b>。</div>
+      <div><span class="pill">命名 flag</span><span class="pill">required/enum 校验</span><span class="pill">风险确认</span><span class="pill">自然语言 Intent</span><span class="pill">dry-run/format</span></div>
+    </div>
+    <div class="layer l-smart">
+      <h3>🟣 真·智能层 · 68 条</h3>
+      <div style="color:var(--muted);font-size:13.5px">照 lark-cli 范式的多步/编排/智能，<b>不是 1:1</b>。框架新增 <code>CallMCPData</code>（多步取数，对标 lark <code>CallAPITyped</code>）+ <code>resolveUser</code>（名→ID，对标 <code>ResolveOpenIDsTyped</code>）。</div>
+      <div style="margin:10px 0"><b>价值</b>：<b>这才是「shortcut 作为新能力」</b>。</div>
+      <div><span class="pill" style="background:#241a3a;color:#c9b3ff;border-color:#4a3a6b">按名解析+消歧</span><span class="pill" style="background:#241a3a;color:#c9b3ff;border-color:#4a3a6b">多工具编排</span><span class="pill" style="background:#241a3a;color:#c9b3ff;border-color:#4a3a6b">失败回滚</span><span class="pill" style="background:#241a3a;color:#c9b3ff;border-color:#4a3a6b">跨服务</span></div>
+    </div>
+  </div>
+
+  <h2><span class="ico">②</span>真·智能层 68 条明细（节选）</h2>
+  <table>
+    <thead><tr><th>shortcut</th><th>多步/智能逻辑</th><th class="c">验证</th></tr></thead>
+    <tbody>
+      <tr><td><code>chat +dm --to &lt;名&gt;</code></td><td>搜人→解析 userId→发单聊；多人消歧</td><td class="c ok">真机 dry-run</td></tr>
+      <tr><td><code>contact +lookup --name &lt;名&gt;</code></td><td>搜人→解析→取完整资料</td><td class="c ok">✅ 真机端到端</td></tr>
+      <tr><td><code>todo +assign --to &lt;名&gt;</code></td><td>解析人→建待办并设执行人</td><td class="c ok">真机 dry-run</td></tr>
+      <tr><td><code>contact +org --name &lt;名&gt;</code></td><td>解析人→取 deptId→查部门详情（3 步）</td><td class="c ok">✅ 真机端到端</td></tr>
+      <tr><td><code>contact +team --name &lt;名&gt;</code></td><td>解析人→取部门→列部门成员</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>calendar +free --who &lt;名&gt;</code></td><td>解析人→查其时段忙闲</td><td class="c ok">✅ 真机端到端</td></tr>
+      <tr><td><code>calendar +book [--with &lt;名CSV&gt;]</code></td><td>建日程→按名加参与者→<b>失败回滚删日程</b></td><td class="c ok">真机 dry-run</td></tr>
+      <tr><td><code>calendar +invite --event --with</code></td><td>解析多人→加入已有日程</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>calendar +suggest-time --with</code></td><td>解析多人→推荐可开会时间</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>calendar +today</code></td><td>算今天范围→列我今天日程</td><td class="c ok">✅ 真机端到端</td></tr>
+      <tr><td><code>calendar +next-event</code></td><td>近 7 天→取最近一个日程</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>calendar +reschedule --event</code></td><td>查日程详情→改时间</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>chat +send-to-group --group &lt;群名&gt;</code></td><td>按群名搜群→消歧→发消息</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>chat +group-members --group &lt;群名&gt;</code></td><td>搜群→列群成员</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>chat +broadcast --to &lt;名CSV&gt;</code></td><td>多名逐一解析→群发单聊，失败汇总</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>todo +todo-done --task &lt;关键词&gt;</code></td><td>列我待办→按标题匹配→标完成</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>todo +remind --task --at</code></td><td>给自己建带提醒的待办</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>minutes +latest-minutes</code></td><td>列妙记→取最新一条详情</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>minutes +action-items</code></td><td>列妙记→取最新→取其待办</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>wiki +wiki-new-doc --space &lt;名&gt;</code></td><td>按名搜知识空间→建文档（跨 doc server 路由）</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>doc +doc-append --doc --text</code></td><td>文档末尾追加文本</td><td class="c ok">编译/挂载</td></tr>
+      <tr><td><code>doc +share-doc --to &lt;名&gt; --url</code></td><td>解析人→把文档链接私信 TA（跨服务）</td><td class="c ok">编译/挂载</td></tr>
+    </tbody>
+  </table>
+  <div class="card">
+    <b>质量亮点（agent 严守 ground truth）</b>：<code>+send-to-group</code> 纠正了「按群名搜群」的正确工具（<code>search_groups</code> 而非按成员昵称的 <code>search_common_groups</code>）；<code>+wiki-new-doc</code> 发现 <code>create_file</code> 在 doc server 并正确跨服务路由；<code>+mail-to</code> 因钉钉无 email 字段<b>主动 skip 拒绝编造</b>。
+  </div>
+
+  <h2><span class="ico">③</span>测试验证（零副作用全量）</h2>
+  <p>写/删命令不能真跑，用<b>假 Caller 拦截</b>——每条命令走完「解析→校验→确认→组装 MCP 调用」，捕获组装出的 <code>(product,tool,params)</code>，不真发网络。</p>
+  <table>
+    <thead><tr><th>验证项</th><th>范围</th><th>结果</th></tr></thead>
+    <tbody>
+      <tr><td><code>go build ./...</code> / gofmt / vet</td><td>全仓</td><td class="ok">✅ 0 告警</td></tr>
+      <tr><td>TestAllShortcutsAssemble</td><td>全部 366</td><td class="ok">✅ 322 组装真实MCP · 44 自校验 · 0 失败/panic</td></tr>
+      <tr><td>TestAllToolLiteralsAreReal</td><td>tool 字面量</td><td class="ok">✅ 0 编造（比对 helper ground truth）</td></tr>
+      <tr><td>TestAllHaveIntent</td><td>全部 366</td><td class="ok">✅ 每条均有自然语言描述</td></tr>
+      <tr><td>TestNoDuplicateCommands</td><td>全部</td><td class="ok">✅ 无重复 · 命名规范</td></tr>
+      <tr><td>usage / userdef 单测</td><td>埋点/沉淀</td><td class="ok">✅ 全通过</td></tr>
+      <tr><td>app 包全量回归</td><td>internal/app</td><td class="ok">✅ ~72s 通过（未破坏现有命令）</td></tr>
+      <tr><td>智能层真机验证（9 批）</td><td>只读/解析类 20+ 条</td><td class="ok">✅ 端到端返回真实数据、投影正确</td></tr>
+      <tr><td>真机抓修真实 bug</td><td>合成测试盖不住的投影/解析偏差</td><td class="ok">✅ 修复 8 处（resolve-dept/at-me/group-members/free/today/suggest-time/unread-chats/dept-members）</td></tr>
+    </tbody>
+  </table>
+  <p style="color:var(--muted);font-size:13px">真机验证补齐了 assemble 测试的盲区：合成响应验证不了「防御式投影是否匹配真实响应结构」。9 批真机验证抓到并修复 8 处解析/投影偏差（如 <code>+resolve-dept</code> 漏了真实容器 key <code>deptList</code>、<code>+at-me</code> 未拍平嵌套、<code>+today/+free</code> 直吐冗长 raw）。少数命令（chat 会话消息读、minutes）因 org/PAT 权限受限无法真机跑通，组装链路仍由 assemble 测试覆盖。</p>
+
+  <h2><span class="ico">④</span>效果评估 GSB（vs lark-cli）</h2>
+  <h3>4.1 能力覆盖 GSB（按 dws 实际暴露的 MCP tool 数 = helper∪shortcut，非 shortcut 数）</h3>
+  <table>
+    <thead><tr><th>lark 服务</th><th class="c">lark</th><th>dws</th><th class="c">dws</th><th class="c">GSB</th><th>说明</th></tr></thead>
+    <tbody>
+      <tr><td>im</td><td class="c">21</td><td>chat</td><td class="c">95</td><td class="c g">G</td><td>群/消息/机器人更全</td></tr>
+      <tr><td>mail</td><td class="c">21</td><td>mail</td><td class="c">43</td><td class="c g">G</td><td>覆盖更广</td></tr>
+      <tr><td>doc</td><td class="c">14</td><td>doc</td><td class="c">34</td><td class="c g">G</td><td>块级读写更细</td></tr>
+      <tr><td>minutes</td><td class="c">14</td><td>minutes</td><td class="c">25</td><td class="c g">G</td><td>录音/说话人更全</td></tr>
+      <tr><td>calendar</td><td class="c">12</td><td>calendar</td><td class="c">24</td><td class="c g">G</td><td>会议室/ACL 更全</td></tr>
+      <tr><td>contact</td><td class="c">2</td><td>contact</td><td class="c">15</td><td class="c g">G</td><td>部门/角色/花名册更全</td></tr>
+      <tr><td>wiki</td><td class="c">12</td><td>wiki</td><td class="c">16</td><td class="c g">G</td><td>略优</td></tr>
+      <tr><td>base</td><td class="c">87</td><td>aitable</td><td class="c">79</td><td class="c s">S</td><td>持平；helper 仅 16，<b>shortcut 补齐 +63</b>（真 gap-fill）</td></tr>
+      <tr><td>task</td><td class="c">18</td><td>todo</td><td class="c">20</td><td class="c s">S</td><td>持平</td></tr>
+      <tr><td>drive</td><td class="c">26</td><td>drive</td><td class="c">25</td><td class="c s">S</td><td>持平</td></tr>
+      <tr><td>apps</td><td class="c">63</td><td>devapp</td><td class="c">25</td><td class="c b">B</td><td><b>helper 无 devapp，25 全由 shortcut 补</b>，但仍少于 lark</td></tr>
+      <tr><td>sheets</td><td class="c">84</td><td>sheet</td><td class="c">60</td><td class="c b">B</td><td>钉钉表格 MCP 较少；helper 已覆盖</td></tr>
+      <tr><td>vc/okr/slides/markdown/whiteboard/note/event</td><td class="c">62</td><td>—</td><td class="c">0</td><td class="c b">B</td><td>钉钉无对应能力，<b>客观不可对齐</b></td></tr>
+    </tbody>
+  </table>
+  <div style="color:var(--muted);font-size:13px;margin:6px 0 2px">🟢 G 7 领域　·　🟡 S 3 领域　·　🔴 B（apps/sheets 少于 lark + 6 领域钉钉无能力）。base 的"持平"几乎全靠 shortcut gap-fill（helper 仅 16）。dws 独有：oa/attendance/report/ding/aisearch/live/devdoc。</div>
+
+  <h3>4.2 组合/智能层 GSB（关键发现）</h3>
+  <div class="keyfind">
+    <b>关键结论</b>：lark 有 ~104 个组合(≥2 次 API)shortcut，但 <b>lark 的组合性多源于飞书 REST API 太细粒度</b>（要先查 spreadsheetToken→sheetId→再操作）；<b>钉钉 MCP 是粗粒度的——一个 tool = 一个完整操作</b>，所以 lark 的组合在钉钉这边<b>大量塌缩成 1:1</b>（已被封装层覆盖），或<b>根本没有对应 tool</b>。
+  </div>
+  <table>
+    <thead><tr><th>lark 组合来源</th><th class="c">数量</th><th class="c">GSB</th><th>钉钉现实</th></tr></thead>
+    <tbody>
+      <tr><td>sheets（先解析 sheetId）</td><td class="c">41</td><td class="c s">S</td><td>钉钉直接吃 token → 1:1 层已覆盖</td></tr>
+      <tr><td>apps db-env/audit/log/trace</td><td class="c">17</td><td class="c b">B</td><td>钉钉无对应工具</td></tr>
+      <tr><td>drive/doc/im（上传/媒体/搜索）</td><td class="c">23</td><td class="c s">S</td><td>多为钉钉 1:1 已覆盖</td></tr>
+      <tr><td>calendar/contact/wiki/minutes/todo 编排</td><td class="c">~10</td><td class="c g">G</td><td>✅ 已建为真·智能 shortcut（+book/+lookup/+org/+wiki-new-doc/+reschedule…）</td></tr>
+      <tr><td>okr/whiteboard/slides/vc</td><td class="c">11</td><td class="c b">B</td><td>钉钉无对应能力</td></tr>
+    </tbody>
+  </table>
+  <div style="color:var(--muted);font-size:13px">→ 钉钉真正需要「组合」的场景（按名解析+多工具编排+跨服务），dws 已覆盖并<b>额外做了 lark 没有的</b>（+today/+broadcast/+share-doc/+action-items 等）。<b>不盲目复刻 lark 的机械多步</b>（在钉钉会成冗余假组合）。</div>
+
+  <h2><span class="ico">⑤</span>dws 差异化优势</h2>
+  <div class="card"><b>复用生产级 MCP 通道</b>（架构性）—— <code>CallMCP</code> 统一继承错误分类(auth/PAT/业务)、dry-run、<code>--format/--jq/--fields</code>；lark 每命令各自实现。</div>
+  <div class="card"><b>协作能力覆盖更全</b> —— chat 95/mail 43/doc 34/minutes 25（dws tool 覆盖）是 lark 对应 2–4 倍。</div>
+  <div class="card"><b>钉钉原生特有能力</b>（lark 完全没有）—— 审批/日志/考勤/DING/企业智能搜索，75 条差异化封装。</div>
+  <div class="card"><b>真·智能编排（22 条）</b> —— 按名解析+消歧、失败回滚、跨服务；<code>resolveUser</code>/<code>CallMCPData</code> 让新智能 shortcut 越写越快。</div>
+  <div class="card"><b>高频自动沉淀（P2 · lark 无此设计）</b>
+    <div class="flow" style="margin-top:9px"><b>高频使用</b><span class="arw">→</span><b>埋点</b><span class="arw">→</span><b>suggest</b><span class="arw">→</span><b>add 写 YAML</b><span class="arw">→</span><b>运行时加载可用</b></div>
+  </div>
+  <div class="card"><b>工程质量</b> —— 假 Caller 拦截，366 条（含写/删）零副作用全量验证，可复跑回归。</div>
+
+  <div class="concl">
+    <b>一句话结论</b>：dws 已把钉钉侧<b>能对齐的都对齐</b>（366 条 = 298 封装 + 68 智能 / 16 服务，1:1 层已去 213 条纯重复），即时协作显著优于 lark，拥有审批/考勤/DING 等原生差异化能力与「高频自动沉淀」独有闭环。lark 的组合优势多因飞书 API 细粒度、在钉钉粗粒度 MCP 下塌缩为 1:1（已覆盖），真正需编排的钉钉侧已建齐；受限项均为钉钉客观无对应能力，非工程遗漏。全部 366 条通过零副作用全量验证。
+  </div>
+
+  <footer>DWS Shortcut Report · 由持续精进 loop 维护 · 另见 <a href="shortcut-comparison.html">逐条三方对照 HTML</a> · <a href="shortcut-report.md">Markdown 版</a></footer>
+</div>
+</body>
+</html>

docs/shortcut-report.md

@@ -0,0 +1,289 @@
+# DWS Shortcut 能力整合与对齐报告
+
+> 版本：截至本轮 loop　|　对齐基准：larksuite/cli（lark-cli）　|　执行底座：钉钉 MCP
+> 相关文档：[总规划](shortcut-plan.md) · [P2 自动沉淀设计](shortcut-p2-design.md) · [HTML 报告](shortcut-report.html) · [**逐条三方对照 HTML**](shortcut-comparison.html)（每个 shortcut：dws +命令 vs lark-cli vs 原生 MCP 组合）
+
+---
+
+## 1. Shortcut 整合了哪些能力
+
+`dws` 现内建 **298 个 1:1 封装 shortcut** + **68 条 smart 智能编排命令**（§1.3）= **合计 366 条**，覆盖 **16 个钉钉服务**，以 `dws <service> +<command>` 形式提供。
+
+> ⚠️ **重要修订（去冗余）**：1:1 层原为 511 条，**复盘发现 `internal/helpers/` 早已把大量 MCP tool 封装成 `dws <svc> <verb>` 产品命令**——其中 **213 条 1:1 shortcut 只是把已被 helper 封装过的同一个 tool 用 `+` 前缀又封了一遍、且无输出投影增量，属纯重复**，已删除。**保留的 298 条 = 233 条填 helper 空白（helper 从没封装的 tool）+ 65 条虽 tool 重复但加了干净投影**。这是对"建 1:1 层前没先摸清 helper 已封装什么"的纠偏（详见 §5 复盘）。aisearch/live/devdoc 三个服务的 shortcut 全属纯重复、已整包移除（其 `dws <svc>` 命令仍由 helper 层提供）。
+
+### 1.1 能力清单（按服务，prune 后）
+
+| 服务 | 1:1 shortcut 数 | 覆盖能力（摘要） |
+|------|:---:|------|
+| chat（群聊/消息） | 79 | 群管理、群成员、群身份角色、消息收发/撤回/转发/表情/卡片、会话置顶/免打扰、消息分组、机器人 |
+| aitable（多维表 base） | 77 | 数据表/字段/记录/视图/表单/仪表盘/图表/角色/协作全生命周期 |
+| attendance（考勤）★ | 33 | 打卡记录、审批、排班、班次、考勤组、统计报表、请假 |
+| devapp（开放平台应用 apps） | 30 | 应用增删改查、成员、权限、版本发布、事件订阅、扩展机器人/H5 配置 |
+| doc（文档） | 16 | 文档/文件夹、正文块读写、权限、附件、节点 |
+| contact（通讯录） | 9 | 用户/部门搜索与详情、角色、花名册 |
+| drive（钉盘） | 8 | 文件/文件夹管理、下载、复制移动、权限、最近访问 |
+| calendar（日历） | 8 | 日程、参与人、会议室、忙闲、ACL、日历本 |
+| minutes（AI 听记） | 7 | 妙记详情/逐字稿、录音控制、说话人 |
+| oa（审批）★ | 6 | 审批实例、单据处理、模板、流程 |
+| mail（邮箱） | 6 | 邮件搜索/线程、标签、联系人、收信规则（投影类保留） |
+| wiki（知识库） | 5 | 知识空间、节点、成员 |
+| todo（待办 task） | 5 | 待办、子任务、执行人/参与人、附件 |
+| ding（DING）★ | 5 | 机器人/个人 DING 发送、撤回、接收状态 |
+| sheet（钉钉表格） | 2 | 区域读写（投影类保留） |
+| report（日志）★ | 2 | 日志收件箱/发件箱 |
+| **合计** | **298** | **16 个服务** |
+
+★ = 钉钉特有服务，lark-cli 无对应（详见 §3 GSB）。**注**：多数服务的 `shortcut 数` 已远小于该服务的 MCP tool 总数——因为 tool 的基础封装由 helper 层的 `dws <svc> <verb>` 命令承担，1:1 shortcut 只保留 helper 没覆盖的、或加了投影的。
+
+### 1.2 每个 shortcut 统一具备的能力（框架注入）
+
+不是简单命令别名，而是叠加在裸 MCP 之上的**精选薄封装**，统一获得：
+
+- **声明式定义**：`Shortcut{Service, Command, Product, Risk, Flags, Execute}`，一处声明、框架编译成 cobra 命令。
+- **自然语言 Intent**：每条 shortcut 均带一段自然语言描述（做什么/何时用/关键输入产出，写删类点明副作用），面向用户与 AI agent 的意图匹配；`--help` 展示为长描述，`dws shortcut list` 输出 `intent` 字段。全部 366 条覆盖（`TestAllHaveIntent` 强制校验）。
+- **参数收敛**：把裸 `dws mcp <svc> <tool> --json '{...}'` 的手拼 JSON，收敛成命名 flag（`--query`/`--group`…）。
+- **内建校验**：required / enum 声明式校验，结构化错误提示。
+- **风险确认**：read / write / high-risk-write 分级，写/删操作 `--yes` 前二次确认。
+- **复用生产级 MCP 通道**：错误分类（auth/PAT/业务）、`--dry-run` 预览、`--format`/`--jq`/`--fields` 输出，全部免费继承（详见 §4）。
+
+### 1.3 两个层次：1:1 封装层 vs 真·多步/智能层（重要澄清）
+
+诚实区分——上面 298 条**绝大多数是 1 shortcut ≡ 1 个 MCP tool 的 1:1 封装**，本质是「给 MCP 套命名 flag + 校验 + Intent 的友好外壳」，价值在 DX 与 agent 可发现性，**不是新能力**。
+
+真正的「shortcut 作为新能力」是 `internal/shortcut/smart/` 下的**多步/智能** shortcut——照 larksuite/cli 的实现范式（`CallAPITyped` 链式多步、按名解析 ID、Validate、DryRun 计划、失败回滚）落地。框架为此新增 `RuntimeContext.CallMCPData(product, tool, params)`（对应 lark 的 `CallAPITyped`：调用并返回 data 供下一步，跨服务）。
+
+已落地的真·智能 shortcut（`internal/shortcut/smart/`，共 68 条，下表为代表性节选）：
+
+| shortcut | 多步/智能逻辑 | 验证 |
+|----------|--------------|------|
+| `chat +dm --to <姓名> --text` | 搜人→解析唯一 userId→发单聊；多人消歧 | ✅ dry-run 真机 |
+| `contact +lookup --name <姓名>` | 搜人→解析 userId→取完整资料 | ✅ **真机端到端** |
+| `todo +assign --to <姓名> --task` | 解析人→建待办并把 TA 设为执行人 | ✅ dry-run 真机 |
+| `chat +send-to-group --group <群名> --text` | 按群名搜群(search_groups)→消歧→发消息 | ✅ 编译/挂载 |
+| `calendar +book --title --start --end [--with <姓名CSV>]` | 建日程→按名加参与者→**失败回滚删日程**（对标 lark `calendar +create`） | ✅ dry-run 真机 |
+| `calendar +free --who <姓名> --start --end` | 解析人→查其时段忙闲 | ✅ **真机端到端**（解析 202397→查忙闲） |
+| `chat +broadcast --to <姓名CSV> --text` | 多名逐一解析→群发单聊，失败汇总不中断 | ✅ 编译/挂载 |
+| `minutes +latest-minutes` | 列妙记→取最新一条详情 | ✅ 编译/挂载 |
+| `chat +group-members --group <群名>` | 按群名搜群→列群成员 | ✅ 编译/挂载 |
+| `contact +org --name <姓名>` | 解析人→取详情拿 deptId→查部门详情 | ✅ **真机端到端**（3 步：董鑫阳→模型算法/16人） |
+| `calendar +suggest-time --with <姓名CSV>` | 解析多人→推荐可开会时间 | ✅ 编译/挂载 |
+| `calendar +invite --event <id> --with <姓名CSV>` | 解析多人→加入已有日程 | ✅ 编译/挂载 |
+| `doc +share-doc --to <姓名> --url` | 解析人→把文档链接私信 TA | ✅ 编译/挂载 |
+| `calendar +today` | 算出今天时间范围→列我今天的日程 | ✅ **真机端到端**（返回真实日程+参会人） |
+| `calendar +next-event` | 近 7 天日程→按时间取最近一个 | ✅ 编译/挂载 |
+| `contact +team --name <姓名>` | 解析人→取部门→列部门直接成员 | ✅ 编译/挂载 |
+| `todo +remind --task --at` | 给自己建带截止/提醒时间的待办 | ✅ 编译/挂载 |
+| `todo +todo-done --task <关键词>` | 列我的待办→按标题匹配→标记完成 | ✅ 编译/挂载 |
+| `calendar +reschedule --event <id>` | 查日程详情→改时间（查→改机械多步） | ✅ 编译/挂载 |
+| `wiki +wiki-new-doc --space <名>` | 按名搜知识空间→在其下建文档（跨 doc server 路由） | ✅ 编译/挂载 |
+| `doc +doc-append --doc --text` | 文档末尾追加文本（update_document append 模式） | ✅ 编译/挂载 |
+| `minutes +action-items` | 列妙记→取最新→取其待办事项 | ✅ 编译/挂载 |
+| `minutes +detail --id <taskUuid>` | 一条命令聚合听记 basic/summary/keywords/transcript/todos，partial-failure 容错 | ✅ 全量测试 |
+| `minutes +replace-batch --id --pair "原文=>替换"…` | 多组批量替换文字，去重校验+逐组结果聚合 | ✅ 全量测试 |
+
+| `oa +approve-by --keyword` ★ | 列待审批→匹配→取 taskId→通过（钉钉原生，lark 无） | ✅ 编译/挂载 |
+| `attendance +my-attendance` ★ | 当前用户→算今天→查我打卡（路由 attendance-wukong server） | ✅ 编译/挂载 |
+| `todo +overdue` | 列我待办→本地过滤过期→投影输出 | ✅ **真机端到端** |
+| `report +report-latest` ★ | 列我日志→取最新→取详情 | ✅ 编译/挂载 |
+| `aitable +find-record --base --table` | 表内按关键词查记录 | ✅ 编译/挂载 |
+
+另有 gap-buildable 补齐（批6）：`chat +my-groups`（列群+类型过滤+投影）、`calendar +find-room`（时段找可用会议室）、`minutes +minutes-search`（关键词搜妙记）、`mail +search-mail`（搜邮件+自动解析绑定邮箱）、`drive +find-file`（搜钉盘文件+投影）。
+
+批7-8 续补（10 条）：`chat +at-me`（近期@我）、`calendar +cancel-event`（查→删，高危二次确认）、`todo +assign-multi`（多人指派）、`contact +dept-members`（搜部门→列成员）、`minutes +transcript`（最新妙记逐字稿）、`calendar +week`（本周日程）、`contact +by-mobile`（手机号→资料）、`todo +created-todos`（我创建的）、`chat +unread-chats`（未读会话）、`mail +unread-mail`（未读邮件）。
+
+批9 续补（3 条·手工，对齐 gap-buildable）：`minutes +detail`（单命令聚合一条听记的 basic/summary/keywords/transcript/todos，partial-failure 容错）、`minutes +replace-batch`（多组 `原文=>替换` 批量替换 + 去重校验 + 逐组结果聚合，补齐一次一组的 1:1 `+word-replace`）、`aitable +record-share-links`（>20 条记录分享链接：去重+分片(≤20/批)+跨 `aitable-helper` server fanout+合并，补齐单批 20 条上限）。
+
+批10 续补（3 条·多 agent 并行，对齐 gap-buildable）：`chat +thread-replies`（拉某条话题消息的全部回复 list_topic_replies + sender/text/time 投影）、`todo +related-tasks`（creator+executor+participant 三角色并集「与我相关的待办」+ taskId 去重 + 投影）、`doc +find-doc`（按关键词搜云文档 search_documents + title/url/type/token 投影）。
+
+批11 续补（3 条·多 agent 并行）：`aitable +resolve-base`（按名搜 Base 解析 baseId，0/1/多候选消歧 search_bases）、`chat +chat-messages`（群/单聊会话消息列表，list_conversation_message_v2 / list_individual_chat_message 互斥+投影）、`mail +find-mail-user`（按名/邮箱搜企业邮箱联系人 search_mail_users + 投影）。
+
+批12 续补（3 条·多 agent 并行，dws 原生 resolver 层，按名解析 ID）：`wiki +resolve-space`（search_wikiSpaces 名→spaceId）、`aitable +resolve-table`（get_tables 在 Base 内名→tableId，本地匹配）、`contact +resolve-dept`（search_dept_by_keyword 名→deptId，含数值 ID 兼容）。均 0/1/多候选消歧，对标 `resolveUser` 的各资源版。
+
+批28 续补（3 条·净新增便利读，dws 原生）：`oa +pending`（**只读**列待我审批，区别于会审批的 +approve-by）、`todo +due-today`（今天到期待办，planFinishDate 服务端过滤，区别于 +overdue 已过期）、`calendar +tomorrow`（明天日程，复用 +today/+week 投影）。均只读、真机验证（+tomorrow 返回真实明日日程；+pending/+due-today 空路径正确且复用已验证 helper）。
+
+批29 续补（3 条·净新增便利读，dws 原生）：`oa +done-approvals`（我已处理的审批历史 get_done_tasks；真机抓到并修复 pageSize=0 → 默认 20 的后端报错 bug）、`mail +recent-mail`（近期收件箱会话 list_mailbox_threads + 解析绑定邮箱/收件箱）、`attendance +this-month`（本月打卡 query_check_record on attendance-wukong，复用 +my-attendance 自身解析）。真机：+this-month 返回有效空、+done-approvals 修后走空路径、+recent-mail 正确报未绑定邮箱。
+
+批30 续补（1 条·净新增便利读）：`contact +me`（当前用户 get_current_user_profile + 投影 {name,userId,mobile,dept,org,email}，agent 的「我是谁」；区别于 1:1 +get-self 吐冗长 raw，真机验证 董鑫阳/202397/模型算法）。
+
+批31 续补（1 条·净新增便利读）：`calendar +my-free`（我自己的忙闲，自动解析当前 userId，默认今天，复用 +free 的 freebusySlots 投影；无需像 +free 传别人姓名，真机验证返回今日忙碌时段）。
+
+批32 续补（1 条·净新增 dws 原生编排，lark 也没有）：`calendar +conflicts`（检测某天日程时间冲突/双重预订，list_calendar_events + 本地两两重叠检测，默认今天/--in-days；真机验证抓到今日 2 处真实冲突）。这类纯 MCP-tool 的本地编排是复杂写死胡同之外仍有价值的方向。
+
+批33 续补（1 条·净新增 dws 原生编排，+conflicts 的互补品）：`calendar +free-slots`（找某天工作时段内的空闲时段"什么时候能安排会"，list_calendar_events + 合并忙碌区间 + 工作窗口内求补集，默认今天 09:00-18:00/--from/--to/--in-days；真机验证今日 4 段空档）。
+
+共 **68 条真·智能 shortcut**（多批多 agent 工作流并行生成 + 手工续补）。★=钉钉原生编排，lark 完全没有。
+
+**真机验证（登录态抽样，返回真实数据）**：`calendar +today/+week`（真实日程+投影）、`contact +org`（3 步→部门详情）、`contact +lookup/+free`、`todo +overdue`、`attendance +my-attendance` 等端到端可用。
+
+### 深度对齐矩阵（逐条分析 lark 361 条 shortcut）
+
+见 [`shortcut-lark-alignment.md`](shortcut-lark-alignment.md)——12 agent 逐条深读 lark 每个 shortcut 的智能实现（Validate/DryRun/ID解析/投影/多步/分页），映射钉钉：
+
+| dws_status | 数量 | 含义 |
+|---|:---:|---|
+| covered-1to1 | 144 (40%) | lark 组合在钉钉塌缩成 1:1，封装层已覆盖 |
+| no-dingtalk-tool | 127 (35%) | 钉钉无对应工具，客观不可对齐 |
+| **gap-buildable** | **42 (12%)** | 钉钉有工具、值得补成智能 shortcut（建设目标） |
+| covered-smart | 48 (13%) | 已建智能 shortcut / 部分覆盖 |
+
+**框架系统性能力已对齐 lark**：`resolveUser`（名→ID）· `CallMCPData`（多步取数）· `rt.Output`（输出投影）· `rt.MutuallyExclusive/AtLeastOne/ExactlyOne/RangeInt/RequireAll`（跨字段校验）。
+
+### 保真度升级（对齐 lark 96% 的输出投影）
+
+lark 96% 的 shortcut 都做**输出投影**（把原始 API 返回精简为干净字段列表）。已给 **~60 条列表/读类封装**升级到此保真度——从 `rt.CallMCP`（打印原始 MCP 返回）改为 `rt.CallMCPData` + 防御式投影 + `rt.Output`（自动吃 `--format/--jq/--fields`）：
+
+`contact +search-user/+search-mobile/+list-roles/+list-sub-depts` · `todo +get-my-tasks/+list-sub` · `calendar +book-list/+attendee-list` · `drive +list` · `wiki +node-list` · `chat +conversation-list/+category-list/+messages-list-unread-conversations/+messages-list-pin` · `doc +search/+list` · `mail +tag-list/+contact-list` · `aitable +base-list/+base-search` · `oa …`
+
+示例：`contact +search-user` 由原始 MCP 返回 → 干净 `{count, users:[{name,userId,flowerName,openDingTalkId,title}]}`（真机验证）。这是把封装层往 lark 高保真水平系统性拉升的开始。另有 `mail +to`（按名发邮件）被 agent **正确 skip**——钉钉 contact 无 email 字段、mail `send_email` 需发件人邮箱 `from` 无法解析，宁缺勿错不编造。关键复用：「按名解析人」抽成共享 helper `resolveUser`（对标 lark `ResolveOpenIDsTyped`，带 0/多人消歧，不瞎猜）；多步靠 `CallMCPData`（对标 lark `CallAPITyped`）。其中 5 条由多 agent 工作流并行生成——各自以 helper 为 ground truth 研究参数、`send-to-group` 的 agent 还主动纠正了「按群名搜群」的正确工具（`search_groups` 而非按成员昵称的 `search_common_groups`）。
+
+> 定位：1:1 层是「MCP 友好外壳」，smart 层才是「真 shortcut」。二者不混淆。
+
+---
+
+## 2. 测试验证报告
+
+**验证理念**：写/删命令不能真跑（会发消息、解散群、删数据），故用**假 Caller 拦截**——让每条命令（含写/删）真实走完「解析→校验→确认→组装 MCP 调用」全流程，捕获组装出的 `(product, tool, params)`，只是不真发网络。以此对**全部 366 条**做零副作用验证。
+
+### 2.1 结果总览（全绿）
+
+| 验证项 | 范围 | 结果 |
+|------|------|------|
+| `go build ./...` | 全仓 | ✅ 通过 |
+| `gofmt -l` / `go vet` | shortcut 全包 | ✅ 0 未格式化 / 0 告警 |
+| 框架单元测试（5） | 类型/挂载/校验/分组 | ✅ 全通过 |
+| **TestAllShortcutsAssemble** | **全部 366 条** | ✅ 322 组装真实 MCP · 44 自校验拦截 · **0 失败 · 0 panic** |
+| **TestAllToolLiteralsAreReal** | 全部 tool 字面量（逐条） | ✅ **0 编造**（tool 名逐一比对 helper ground truth） |
+| TestNoDuplicateCommands | 全部 366 条 | ✅ 无重复、命名规范（均 `+` 前缀） |
+| **TestAllHaveIntent** | 全部 366 条 | ✅ 每条均有自然语言 Intent 描述（无一遗漏） |
+| usage 包单测（4） | 埋点/脱敏/聚合/开关 | ✅ 全通过 |
+| app 包全量回归 | `internal/app` | ✅ 72.2s 通过（接线未破坏任何现有命令） |
+| 只读命令真机验证 | ~25 条（登录态，见 §2.4） | ✅ `contact +me`/`+org`/`+lookup`、`calendar +today/+week/+free/+conflicts/+free-slots`、`doc +find-doc`、`aitable +resolve-base/+resolve-table`、`drive +find-file`、`oa +my-initiated` 等端到端返回真实数据、投影核对 |
+
+### 2.2 关键指标解读
+
+- **322 「组装真实 MCP」**：喂合成参数后成功组装出 MCP 调用，且 tool 名经 helper ground truth 核验真实、非编造。
+- **44 「自校验拦截」**：这些命令有结构化/JSON/互斥输入（如多维表建记录需 JSON、DING 三选一接收人），dummy 值被其**自身校验正确拒绝**——证明校验链路健全。其 tool 名由静态测试 `TestAllToolLiteralsAreReal` 单独覆盖，无遗漏。
+- **0 编造 / 0 panic / 0 失败**：无幻觉工具名，无运行时崩溃，无死命令。
+
+> ⚠️ **验证强度分层（诚实口径，勿把"全绿"读成"真机全对"）**：`TestAllShortcutsAssemble` 的"0 失败"只证明**能正确组装 MCP 调用、零副作用**——它用**合成响应**，**验证不了防御式投影是否匹配真实响应结构**（真机验证正是靠这个抓到过 deptList 容器、pageSize=0 等 assemble 盖不住的 bug，见 §2.4）。按真机验证强度分三层：**(A) 真机正向验证** ~25 条只读/解析类（返回真实数据、投影核对）；**(B) 仅 assemble + 复用已验证 helper**（如 +due-today/+this-month 等，逻辑同构于已验证命令，但该条本身未在真机跑出正样本）；**(C) 未对真实后端跑过**——17 条写类 smart（不宜真跑，会发消息/建数据）、6 条 minutes smart（该 org 未开 CLI 数据访问）、mail +recent-mail（无绑定邮箱）。(C) 类**很可能仍有 assemble 盖不住的投影/参数 bug**，不应因"全绿"就当作"真机可用"。
+
+### 2.3 防幻觉机制（工作流生成时）
+
+生成阶段每个服务由独立 agent 负责，硬性规则：tool 名与参数 key **只能逐字取自 dws helper 的真实调用点**，无法确定参数的 tool 主动跳过并记录原因（如嵌套对象、时间戳转换、本地文件分片上传）。测试阶段再用 ground truth 二次核验，双重保险。
+
+### 2.4 真机验证战役（登录态打真实钉钉后端）
+
+assemble 测试用**合成响应**，能验证「调用是否组装正确」，但验证不了「防御式投影解析是否匹配真实响应结构」。为此做了 9 批真机验证（登录态 corp「钉钉」，token 有效期内），把只读/解析类 smart shortcut 打真实后端、逐条核对投影输出。
+
+**正向验证 20+ 条**（返回真实数据、投影正确）：`doc +find-doc`、`aitable +resolve-base`/`+resolve-table`/`+list-tables`/`+base-list`/`+find-record`、`mail +find-mail-user`、`chat +my-groups`/`+group-members`/`+at-me`、`contact +org`/`+lookup`、`calendar +today`/`+week`/`+next-event`/`+free`/`+suggest-time`、`todo +overdue`/`+related-tasks`、`attendance +my-attendance`、`report +report-latest`、`oa +my-initiated`、`drive +find-file`、`wiki +resolve-space` 等。
+
+**真机抓到并修复 8 个真实问题**（assemble 测试抓不到，只有真机能抓）：
+
+| shortcut | 真机发现的问题 | 修复 |
+|---|---|---|
+| `contact +resolve-dept` | 对任何真实部门名都「未找到」——真实响应容器 key 是 `deptList`（防御探测清单漏了），deptName 带 `<red>` 高亮、deptId 是数值 | 加 `deptList` 探测 + `stripHighlightTags` + 数值 coerce |
+| `contact +dept-members` | 消歧消息泄漏 `<red>` 标记 | 复用 `stripHighlightTags` |
+| `chat +unread-chats` | 每行吐 `unread: null`（底层不返回每会话未读数） | 仅当有值才带该字段 |
+| `chat +at-me` | 直吐原始两层嵌套、未拍平 | 新增 `atMeFlattenGroups`，拍平 43 条为 `{conversation,sender,text,time}` |
+| `chat +group-members` | 终结步 raw `CallMCP` 吐冗长 raw（含 avatar 媒体 ID + errorCode 噪音） | 升级 `CallMCPData`+投影 `{name,nick,role,openDingtalkId}` |
+| `calendar +free` | 吐冗长 `result[].scheduleItems[].{start,end}.dateTime` 嵌套 | 投影为 `{who,userId,free,busy:[{start,end}]}` |
+| `calendar +today` | 吐 17 字段冗长事件（含完整 attendees 数组），与 `+week` 不一致 | 投影为 `{title,start,end,location,eventId}`，对齐 `+week` |
+| `calendar +suggest-time` | `timeConflictAttendees:[null]` 噪音 + result 包裹 | 拍平 + 丢 null 冲突 → `{suggestions:[{start,end}]}` |
+
+**后端受限、无法真机正向验证的（非代码问题）**：`chat +chat-messages`/`+search-msg`（读会话消息需更高 PAT 权限 / org 未开 CLI 数据访问）、`minutes +*`（org 未开 CLI 数据访问 `TOKEN_VERIFIED_FAILED`）、`contact +team`（列部门成员 medium-risk 权限墙）。这些命令的**组装链路**经 assemble 测试验证正确，仅无法在本环境跑通后端。
+
+**一个已澄清的非 bug**：`resolveUser` 对组织内成员（如董鑫阳→userId 202397）真机端到端正常；对外部/资料受限联系人 `search_contact_by_key_word` 只返回 openDingTalkId（name/userId 全 null），此时正确报「没找到」而非瞎猜——钉钉数据模型现实，非代码缺陷。
+
+> **结论**：真机验证证明「防御式多候选 key 投影」在真实响应上整体成立，并纠正了 8 处「合成测试盖不住」的解析/投影偏差。这是把可用性从「组装正确」提升到「真机输出正确」的关键一环。
+
+---
+
+## 3. 效果评估 GSB（vs lark-cli）
+
+以 lark-cli 各服务领域为基准，评估 dws shortcut 的相对表现。**G**ood=优于/更全，**S**ame=持平，**B**ad=弱于/缺失。
+
+> ⚠️ 重要前提：两边是**不同 API**（飞书 vs 钉钉），数量不能机械 1:1；覆盖度受钉钉实际能力约束。
+>
+> **口径（prune 后重做）**：不再用 shortcut 数（会因去冗余失真），改用**「dws 该服务实际暴露的 distinct MCP tool 数」= helper 命令 ∪ shortcut 覆盖的 tool 合集**——这才代表真实能力，与 helper/shortcut 怎么分层无关。对比 lark 的 shortcut 数（不同 API，只作量级参考）。
+
+| lark 服务 | lark 数 | dws 对应 | dws tool 覆盖 | (其中 shortcut 补) | GSB | 说明 |
+|-----------|:---:|---------|:---:|:---:|:---:|------|
+| im | 21 | chat | **95** | +3 | 🟢 G | 群/消息/机器人能力更全 |
+| mail | 21 | mail | **43** | +0 | 🟢 G | 覆盖更广（几乎全由 helper 提供，shortcut 曾重复、已 prune） |
+| doc | 14 | doc | **34** | +4 | 🟢 G | 块级读写更细 |
+| minutes | 14 | minutes | **25** | +0 | 🟢 G | 录音控制/说话人更全 |
+| calendar | 12 | calendar | **24** | +5 | 🟢 G | 会议室/ACL/忙闲；shortcut 另补排期智能 |
+| contact | 2 | contact | **15** | +0 | 🟢 G | 部门/角色/花名册更全 |
+| wiki | 12 | wiki | **16** | +0 | 🟢 G | 略优 |
+| base | 87 | aitable | **79** | **+63** | 🟡 S | 基本持平；**helper 仅 16，shortcut 层补齐了绝大部分**（真·gap-fill） |
+| task | 18 | todo | **20** | +1 | 🟡 S | 持平 |
+| drive | 26 | drive | **25** | +4 | 🟡 S | 基本持平 |
+| apps | 63 | devapp | **25** | **+25** | 🔴 B | **helper 无 devapp 命令、25 个全由 shortcut 提供**（纯 gap-fill），但仍少于 lark |
+| sheets | 84 | sheet | **60** | +1 | 🔴 B | 钉钉表格 MCP 较少；helper 已覆盖，shortcut 曾重复、已 prune |
+| vc | 18 | — | 0 | — | 🔴 B | 钉钉 conference 无干净 MCP tool |
+| okr | 13 | — | 0 | — | 🔴 B | 钉钉无对应能力，**不可对齐** |
+| slides / markdown / whiteboard / note / event | 17 | — | 0 | — | 🔴 B | 钉钉无对应能力，**不可对齐** |
+
+**dws 独有（lark 无对应服务）**：oa 审批(~20 tool) · attendance 考勤(~38) · report 日志(~7) · ding(~8) · aisearch/live/devdoc（helper 层提供）——钉钉工作流核心，构成差异化。
+
+### GSB 汇总
+
+- 🟢 **G（7 领域）**：im/mail/doc/minutes/calendar/contact/wiki——dws tool 覆盖更全。
+- 🟡 **S（3 领域）**：base/task/drive 量级持平（base 的持平**几乎全靠 shortcut 层 gap-fill**，helper 只有 16）。
+- 🔴 **B（受限 2 + 不可对齐 6）**：apps/sheets 少于 lark（sheets 受钉钉 API 限，apps 全由 shortcut 补但仍少）；vc/okr/slides/markdown/whiteboard/note/event 客观不可对齐（非遗漏）。
+- **注**：这张表也印证了 1:1 层的真实价值分布——**base/apps 靠 shortcut 补了大量 helper 没有的 tool（gap-fill），而 mail/sheets 的 shortcut 基本是重复 helper（已 prune）**。
+
+---
+
+## 4. dws 差异化于 lark 的优势
+
+### 4.1 执行底座：复用生产级 MCP 通道（架构性优势）
+lark-cli 每个 shortcut 直连飞书 SDK，错误处理/输出各自实现。dws shortcut 的 `CallMCP` **委托统一的 MCP 调用路径**，一步继承：
+- **错误分类**：auth 过期 / 未登录 / PAT / 业务错误，自动给出可执行提示；
+- **`--dry-run` 预览**：不发网络，输出将执行的 tool + 参数；
+- **`--format`/`--jq`/`--fields`**：机器可解析输出，Agent 友好。
+
+lark 需在每个命令重复实现这些；dws 由框架统一注入，一致性与维护成本双赢。
+
+### 4.2 覆盖更全的高频协作能力
+按 **dws tool 覆盖**（helper∪shortcut，§3 口径）：chat 95 / mail 43 / doc 34 / minutes 25 / calendar 24 等即时协作场景，多为 lark 对应服务的 2–4 倍。
+
+### 4.3 钉钉原生特有能力（lark 完全没有）
+审批 oa、日志 report、考勤 attendance、DING、企业智能搜索 aisearch —— 这些是钉钉工作流的核心，构成差异化护城河。
+
+### 4.4 高频自动沉淀（P2，lark 无此设计）
+基于用户高频使用**主动把常用操作沉淀为自定义 shortcut**（`~/.dws/shortcuts/*.yaml` 运行时加载），让 CLI 越用越顺手。埋点**默认关（opt-in，`DWS_USAGE_TRACKING=1` 开启）**+开启后首跑告知，隐私优先（记形状不记值）。详见 [P2 设计](shortcut-p2-design.md)。**lark-cli 无任何等价能力。**
+
+### 4.5 AI Agent 友好
+`--yes` 跳过确认、结构化错误、`--dry-run` 预览、`--print-schema`（规划中）——为 Agent 自动化调用而设计。
+
+### 4.6 工程质量：全量自动化测试
+假 Caller 拦截，对全部 366 条（含写/删）做零副作用验证 + tool 名 ground truth 核验，可复跑、可回归。
+
+---
+
+## 5. 复盘与后续（loop 持续项）
+
+### 5.0 关键复盘：1:1 层去冗余（511 → 298）
+**问题**：建 1:1 shortcut 层之前，**没有先摸清 `internal/helpers/` 已经把哪些 MCP tool 封装成了 `dws <svc> <verb>` 产品命令**。结果对 **213 个 tool 重复封装**——同一个 tool，helper 有 `dws contact user search`、我又造了 `dws contact +search-user`，且这批无投影增量，纯重复。
+**纠偏**：按「tool 已被 helper 封装 且 shortcut 用 CallMCP 无投影」精确删除 213 条，1:1 层 511 → **298**（保留 233 填空白 + 65 有投影），总数 579 → **366**，服务 19 → **16**（aisearch/live/devdoc 整包移除）。全绿、真机复验保留命令仍可用。
+**教训**：**做封装层前先审已有封装**。对齐 lark「每一条 shortcut」时，应先问「dws 这边是不是已经有等价命令了」，而不是无脑对齐。这是本项目最大的方法论盲点。
+
+1. ~~补 apps（↔devapp）~~ ✅ 已完成：新增 30 个 devapp shortcut。
+2. **P2 落地**（差异化能力，lark 无）：
+   - ✅ **P2-1 usage 埋点**：装饰 MCP 调用唯一必经点记录 `~/.dws/usage.jsonl`（**记形状不记值**，敏感/自由文本字段脱敏；**默认关/opt-in**，`DWS_USAGE_TRACKING=1` 开启、开启后首跑告知）。端到端验证通过。
+   - ✅ **stats/list**：`dws shortcut list [--service]`、`dws shortcut stats [--top N] [--purge]`（按 `(product,tool,arg_keys)` 聚合、识别固定值 fixed_args）。
+   - ✅ **P2-2 suggest**：`dws shortcut suggest [--min N]` 把高频分组转成「建议沉淀的 +command」候选（含固定/可变参数拆分）。
+   - ✅ **P2-3 YAML 自定义 shortcut 沉淀闭环**：`dws shortcut add` 写 `~/.dws/shortcuts/*.yaml` → 下次运行 `userdef.Load()` 编译成 Shortcut 注册（复用同一 runner；`${flag}` 绑定+常量；与内建冲突自动跳过）。**端到端验证通过**：add→重载→`dws <svc> +<cmd>` 可用，dry-run 组装正确。
+   - ⏳ **P2-4 nudge**：命中高频候选时主动提示（TTY、频控、可永久关闭）。
+
+   > 至此，**「高频使用 → 建议 → 一键沉淀 → 自定义 shortcut 运行时生效」完整闭环已打通**——这是 lark-cli 完全没有的差异化能力。
+3. **深化 sheets**：随钉钉表格 MCP 能力增强补齐。
+4. **实机全读回归**：登录态下对全部只读 shortcut 做真实调用回归（需有效 token + 真实资源 ID）。
+5. **不可对齐项归档**：okr/slides/whiteboard/note/vc/event 明确标注为钉钉无能力，避免误解为遗漏。
+
+---
+
+## 附：一句话结论
+
+> dws 已把钉钉侧**能对齐的主要服务全部对齐**（16 服务 / **366 shortcut** = 298 封装 + 68 智能编排），在即时协作能力上显著优于 lark，并拥有审批/考勤/日志/DING 等钉钉原生差异化能力与「高频自动沉淀」独有设计；受限项均为钉钉客观无对应能力，非工程遗漏。全部 366 条通过零副作用全量自动化验证。

docs/skill-content-framework.md

@@ -0,0 +1,105 @@
+# DWS Skill 内容框架合同
+
+> 本分支权威合同：`skills/mono` / `skills/multi` 的**内容组织**与 zip 内容树形状。  
+> 不做安装/升级行为约定。质检见 [skill-mono-multi-qa.md](skill-mono-multi-qa.md)。  
+> 对齐调研：[skill-wukong-align-plan.md](skill-wukong-align-plan.md)。
+
+## 1. 两棵内容树
+
+| 树 | 路径 | 角色 |
+|---|---|---|
+| **mono**（单 skill） | `skills/mono/` | 单一 `SKILL.md` 入口 + `references/products/*` 产品面 + 全局协议 |
+| **multi**（多 skill） | `skills/multi/` | 平铺 `dingtalk-*` 产品 skill + 必选 `dingtalk-shared` |
+
+Agent / 安装面选哪棵树由**行为分支**决定；本文件只规定树内合同。
+
+## 2. Multi 目录合同（如何新增一个产品 skill）
+
+新建 `skills/multi/<name>/` 时必须满足：
+
+1. **命名**
+   - 产品 skill：`dingtalk-<product>`（小写、连字符）
+   - 共享 skill：仅允许 `dingtalk-shared`
+2. **根文件**
+   - 必有 `SKILL.md`（YAML frontmatter + 正文）
+   - `references/` 推荐；无 reference 的 skill（如极简 profile）须在质检 omit 表登记
+   - `scripts/` 可选；脚本须被本 skill 树内某 `.md` 引用，或进入 orphan allowlist
+3. **Frontmatter 最小集**（产品 / shared）
+   - `name`：与目录名一致
+   - `description`：非空，含触发意图与边界
+   - `metadata.category`：`product` 或 `shared`（允许历史写法把 `cli_version` 放在 frontmatter 顶层）
+   - `metadata.requires.bins`：含 `dws`
+4. **契约块**
+   - 产品 skill 推荐内嵌 `<!-- DWS_RUNTIME_CONTRACT_START -->…END -->` **或** 明确 PREREQUISITE 指向 `dingtalk-shared`
+   - `dingtalk-shared` 承载跨产品路由与全局协议落点
+5. **与 mono 映射**
+   - 每个 mono `references/products/<stem>`（文件或目录）必须在
+     `skills/content-qa/mono-multi-coverage.yaml` 有 `coverage` 或 `omit_coverage` 行
+
+### 2.1 推荐骨架
+
+```text
+skills/multi/dingtalk-example/
+├── SKILL.md
+├── references/
+│   ├── example.md          # 主产品面
+│   └── …                   # 子章节 / 意图表
+└── scripts/                # 可选；须被 md 引用
+    └── example_helper.py
+```
+
+## 3. Mono 目录合同（质检对照基准）
+
+```text
+skills/mono/
+├── SKILL.md
+├── references/
+│   ├── products/           # 覆盖质检主源
+│   ├── error-codes.md      # 全局协议示例
+│   ├── error-codes.md
+│   └── …
+└── scripts/
+```
+
+- `references/products/` 下每个顶层 stem（`.md` 去后缀或子目录名）计入覆盖索引。
+- 同 stem 的 `.md` + 子目录视为同一产品面（如 `doc.md` + `doc/`）。
+
+## 4. 共享内容（`dingtalk-shared`）
+
+| 职责 | 落点 |
+|---|---|
+| 跨产品路由 / 工作流 | `references/routing.md`、`workflow-routing.md`、`intent-guide.md` |
+| 运行时最小契约长文 | `references/runtime-contract.md`（受 context-budget 约束） |
+| 全局协议（确认门禁 / Schema 教学等） | `references/`；见质检基线 |
+| 与 mono 全局文同名迁移 | `error-codes`、`url-patterns`、`capability-limits`、`channel-login`、`global-reference`、`recipes/`（`conventions.md`、`meta.md`、`lite-catalog.md`） |
+
+产品专属规则（如 AI 表格 `field-rules`）允许下沉到对应 `dingtalk-*`，须在覆盖表注明。
+
+## 5. Zip 内容布局合同（形状，非安装默认）
+
+发布物 `dws-skills.zip`（及 embed 同源）内容树形状：
+
+| Zip 路径 | 含义 |
+|---|---|
+| `<root>/` | mono 内容副本（兼容旧面） |
+| `<root>/mono/` | 与 `skills/mono/` 同构 |
+| `<root>/multi/` | 与 `skills/multi/` 同构 |
+
+质检可断言源树形状；**不**断言安装器默认解压哪棵。
+
+## 6. 与悟空 `dingtalk-skills/` 对照（组织概念 only）
+
+| 维度 | DWS `skills/multi` | 悟空 `dingtalk-skills/`（develop） |
+|---|---|---|
+| 布局 | flat `dingtalk-*` + `dingtalk-shared` | 同构 flat |
+| 集合 | 产品 skill + shared（含 event/profile/…；dev/skill 等长尾落在 misc） | 更小产品集（如 attendance/report 独立目录） |
+| 质检权威 | **mono 单 skill 树** | 不作为 DWS 覆盖基准 |
+| 不移植 | `_install.sh` / bundle / dual / Qwen overlay | — |
+
+悟空独有命名（如 `dingtalk-attendance`）在 DWS 中由 `dingtalk-misc` 承接对应 mono `attendance*` / `report` / `oa` / `sheet` / `dev` 等面——见覆盖表。
+
+## 7. 变更流程
+
+1. 改 / 增内容 → 更新 `skills/content-qa/mono-multi-coverage.yaml`（coverage 或 omit）  
+2. 跑 `make skill-mono-multi-content`（该独立门禁不包含在默认 `make policy` 中）
+3. 失败则修内容或更新 reviewed omit（disposition + 原因），**禁止**用安装默认值绕过  

docs/skill-mono-multi-qa.md

@@ -0,0 +1,76 @@
+# Mono↔Multi Skill 内容质检规格
+
+> 对照基准：`skills/mono`（单 skill）。被测主体：`skills/multi`。  
+> 机读合同：`skills/content-qa/mono-multi-coverage.yaml`。  
+> 执行：`make skill-mono-multi-content`（独立门禁；默认 `make policy` 按设计不包含该检查）。
+
+## 1. 质检矩阵
+
+| ID | 类型 | 输入 | 通过准则 |
+|---|---|---|---|
+| **G1 形状** | 结构 | `skills/multi/*` | 仅 `dingtalk-*`（含必选 `dingtalk-shared`）；每目录有 `SKILL.md` |
+| **G2 结构** | 结构 | 各 `SKILL.md` frontmatter | `name`==目录名；非空 `description`；`category`∈{product,shared}；`requires.bins` 含 `dws` |
+| **G3 覆盖** | 覆盖 | mono `references/products/*` 顶层 stem | 每 stem ∈ `coverage` 或 `omit_coverage`；coverage 目标 skill/refs 存在 |
+| **G4 漂移** | 漂移 | scripts、成对文件、全局协议 | orphan 脚本 ∈ allowlist；paired 内容一致（允许合同声明的布局链接替换）；全局协议存在或 ∈ `omit_global` |
+| **G5 链接** | 可达性 | 合同覆盖的 paired Markdown | 内联相对链接目标文件或目录存在且不逃出仓库；外链、纯锚点和锚点内容不在检查范围 |
+
+已有门禁（继续复用，不替代本矩阵）：`check-skill-commands`、`check-skill-context-budget`、`check-multi-im-skill-chain`、`skill_docs_policy`、whiteboard 成对测试。
+
+## 2. 有意省略 / 延期登记格式
+
+YAML（见 coverage 文件）：
+
+```yaml
+omit_coverage:
+  - mono: simple
+    disposition: covered_by   # covered_by | defer | wontfix
+    via: dingtalk-misc        # optional
+    reason: "拆入 oa/devdoc…"
+
+omit_global:
+  - id: field-rules-global
+    mono_path: references/field-rules.md
+    expected_multi: dingtalk-aitable/references/field-rules.md
+    disposition: covered_by
+    reason: "AI 表格字段规则已下沉到 dingtalk-aitable；G3/coverage 不强制全局同名"
+
+orphan_scripts_allowlist:
+  - path: dingtalk-misc/scripts/report_received_today.py
+    disposition: defer
+    reason: "pending report.md reference"
+
+paired_files:
+  - mono: references/products/sheet.md
+    multi: dingtalk-misc/references/sheet.md
+    mode: link-normalized
+    link_substitutions:
+      - mono: "../url-patterns.md"
+        multi: "../../dingtalk-shared/references/url-patterns.md"
+      - mono: "../intent-guide.md"
+        multi: "sheet-intent-guide.md"
+```
+
+**处置原则**：质检失败 → 修**内容**或更新 reviewed omit；**不**改安装/升级默认。
+
+## 3. 缺口基线（相对 mono）
+
+| ID | 项 | disposition | 说明 |
+|---|---|---|---|
+| M1 | recovery-guide / RECOVERY_EVENT_ID 闭环 | **removed** | 已从 mono/multi skill 文档删除；不做移植 |
+| M2 | confirmation_required 全局协议 | **done** | `dingtalk-shared/references/confirmation.md` + SKILL 导航 |
+| M3 | Schema 渐进查询教学 | **done** | `dingtalk-shared/references/schema-usage.md` |
+| M4 | `report_inbox_today.py` | `defer` / orphan 侧 | 验证后迁 misc 或删 |
+| M5 | multi LICENSE/NOTICE | `defer` | 内容或打包注入 |
+| M6 | aiapp 路由 vs orphan 脚本 | **done（标明未产品化）** | mono 死链移除；`unsupported-scripts.md` |
+| X1 | yida/finance/aiapp orphan scripts | **done（登记）** | 由 unsupported-scripts 具名引用 |
+| X2 | chat 死链 `extract_media_id.py` | n/a | 现仅为反模式提及 |
+| X3 | routing → markdown 错路径 | **done** | 已指 `dingtalk-misc/references/markdown.md`；drive 尾链已修 |
+| X4 | event 缺 metadata | **done** | |
+| X5 | multi skill 横幅 /「优先 mono」文案 | **done** | 横幅已全部移除 |
+| X6 | SAFETY_PREAMBLE_INJECT 无注入器 | **done** | 标记已移除 |
+
+产品面覆盖：见 YAML `coverage`——mono products 均有 multi 承接（misc 聚合 attendance/oa/sheet/…）。
+
+## 4. 与悟空
+
+借鉴 frontmatter / 断链 / requires 等**检查维度**；不运行悟空 bundle zip 校验脚本。覆盖权威始终是 DWS mono。

docs/skill-wukong-align-plan.md

@@ -0,0 +1,272 @@
+# DWS multi-skill **内容框架**对齐方案（相对 dws-wukong develop）
+
+> 状态：**执行中** — Phase 1–3 已落地；M2/M3 已补；**M1 recovery 闭环已从 skill 删除（不做移植）**。  
+> 合同短文：[skill-content-framework.md](skill-content-framework.md)  
+> 质检规格：[skill-mono-multi-qa.md](skill-mono-multi-qa.md)  
+> 机读合同：`skills/content-qa/mono-multi-coverage.yaml`  
+> 门禁：`make skill-mono-multi-content`（独立门禁；默认 `make policy` 按设计不包含该检查）
+>
+> 撰写 / 收窄 / 质检增补 / 执行：2026-08-05  
+> 工作树：`/Users/john/GolandProjects/open-source/dws-multi-skill-align`  
+> 分支：`feat/multi-skill-framework-align`（自 `origin/main` @ `a37e6e68`）  
+> **本分支范围：只做 skill 内容的这个框架**（目录布局、文档契约、共享内容约定、zip 内容树合同、**相对 mono 的内容质检**）。  
+> **不做**安装/升级引擎、agent-home、脚本 skill-install 行为翻转。
+>
+> 对照仓：
+>
+> | 仓 | 路径 | 基线 |
+> |---|---|---|
+> | DWS OSS CLI（本工作树） | `dws-multi-skill-align` | `origin/main` |
+> | dws-wukong | `~/GolandProjects/open-source/dws-wukong` | `origin/develop` @ `ab76629a`（调研时） |
+> | 行为参考（**另一分支**） | `dws-skill-mode-migration` @ `402429ac`/`d5c8982c` | 安装默认 multi / upgrade 强制 multi —— **不在本分支排期** |
+> | 内容缺口留档（参考） | 同迁移分支 `docs/skill-capability-completion.md`（M1–M6 / X1 等） | **仅作质检目标线索**，非本分支权威 |
+
+---
+
+## 0. TL;DR
+
+1. **本分支 = skill 内容框架 + 相对 mono 的内容质检**：固化 `skills/multi` 组织合同，并用 **mono 单 skill 布局作对照基准**做覆盖/结构/漂移门禁（文档 + CI 内容护栏）。
+2. **对齐悟空**：只取内容树组织概念；质检以 **DWS-native** 设计为主（已有 policy/测试可复用）。悟空 `validate-multiskill-bundle.py` 仅借鉴「frontmatter / 断链 / requires」类检查思路，**不**移植 bundle/安装校验。
+3. **安装/升级行为**与 `402429ac`/`d5c8982c` → **单独 follow-up 分支**，本方案只登记。
+4. 质检 **不改**默认安装哪棵树；只保证 multi 内容相对 mono **可解释、可覆盖、可回归**。
+
+### 0.1 IN SCOPE
+
+| 类别 | 包含 |
+|---|---|
+| 内容树结构 | `skills/mono/` 与 `skills/multi/<name>/` 目录合同 |
+| 单 skill 约定 | `SKILL.md` frontmatter / 契约块 / Golden Route；`references/`；可选 `scripts/` |
+| 共享内容 | `dingtalk-shared` 职责与被引用方式；与 mono 全局文映射（文档级） |
+| 命名与集合 | `dingtalk-*` + `dingtalk-shared`；相对悟空的共有/独有清单（文档） |
+| Zip **内容布局合同** | `mono/` / `multi/` / 根 mono 副本的内容含义与树形状；不改安装默认 |
+| **Mono↔multi 内容质检** | 覆盖、结构、漂移三类门禁；复用/扩展现有 policy 与测试；缺口修复属内容编辑（另批或同分支内容 Phase） |
+| 内容架构文档 | 本文件 + 可选短文（架构合同 + 质检矩阵） |
+
+### 0.2 OUT OF SCOPE
+
+| 类别 | 去向 |
+|---|---|
+| 安装默认 multi、upgrade always-multi | Follow-up 分支（`402429ac`/`d5c8982c`） |
+| `LocateSkillsRoot` / `skill_setup` / `paths.go` / `skillhome` / install 脚本行为 | 同上 |
+| 安装/运行时 manifest、state.json、mode 切换、telemetry header | 拒绝或行为分支 |
+| 悟空 `_install.sh` / dual / Qwen / RewindDesktop / pod | 拒绝 |
+| 非 skill 内容的 CLI 功能（schema/shortcut 代码等） | 拒绝 |
+| 把质检做成「改安装默认值」的后门 | 拒绝 |
+
+---
+
+## 1. 内容现状盘点
+
+### 1.1 DWS `skills/mono`（质检对照基准 · 单 skill）
+
+```text
+skills/mono/
+├── SKILL.md
+├── references/
+│   ├── products/<area>.md|…/   # 产品能力面（质检「覆盖」主源）
+│   ├── error-codes.md、…  # 全局协议（无 recovery 闭环）
+│   └── best_practices/…
+└── scripts/
+```
+
+### 1.2 DWS `skills/multi`（内容主体）
+
+```text
+skills/multi/
+├── dingtalk-shared/     # 跨产品契约 / routing / 全局协议应落点
+└── dingtalk-*/     # 19 产品 + 各 references、scripts
+```
+
+仅 DWS 有（悟空无）：dev, event, hrbrain, markdown, pat, profile, skill。
+
+### 1.3 悟空 `dingtalk-skills/`（内容组织对照，非质检权威）
+
+Flat `dingtalk-*` + `dingtalk-shared`；单 skill 骨架同构。**不作为 mono 覆盖基准**（集合更小、不同源）。
+
+### 1.4 Zip 内容布局合同
+
+| Zip 路径 | 内容含义 |
+|---|---|
+| `<root>/` | mono 副本（兼容） |
+| `<root>/mono/` | 显式 mono 内容源 |
+| `<root>/multi/` | 与 `skills/multi/` 同构 |
+
+质检可断言「源树形状」；**不**断言安装面默认选哪棵。
+
+### 1.5 现有 DWS skill 内容质检资产（复用清单）
+
+| 资产 | 作用 | 与 mono↔multi 质检关系 |
+|---|---|---|
+| `scripts/policy/check-skill-commands.sh` + `skill-command-check/` | Skill 文内 `dws …` 命令路径存在性 | **复用**（命令真实性）；非覆盖映射 |
+| `scripts/policy/check-skill-context-budget.sh` | chat/event/mono/`dingtalk-shared` 上下文预算与冷启动约束 | **复用**（结构/预算）；可扩展 shared 引用规则 |
+| `scripts/policy/check-multi-im-skill-chain.sh` + `multi-im-skill-chain/` | IM 意图单默认路由、retired scripts、handoff | **复用**（chat/event 链）；面窄 |
+| `test/unit/skill_docs_policy_test.go` | 退役命令、event 扁平输出契约等 | **复用**；可加 mono↔multi 断言 |
+| `test/unit/whiteboard_skill_docs_test.go` | mono/multi whiteboard recipes **字节一致** | **样板**：产品面「同源文件」门禁范式 |
+| `test/skill_static`（`-tags skill_verify`） | 文内命令 vs Cobra；multi 查 flag | **复用**（opt-in 深度）；非 CI 默认全量时可保持 tags |
+| `test/skill_e2e` / `test/run_skill_tests.py` | 执行层 / 用例驱动 | **偏行为**；本分支质检默认不依赖 e2e |
+| `Makefile` → `policy` 含 context-budget、multi-im-skill-chain；`skill-command-integrity` 独立 | 已有 CI 钩子 | 新门禁优先挂同类 policy / `test/unit` |
+
+**缺口（尚无的门禁）**：系统的「mono `references/products/*` → multi 目录/文」覆盖表；frontmatter 全集完备性；orphan scripts。全局协议中 **确认门禁 / Schema 教学已补**；**recovery 闭环已从 skill 移除（不再作为缺口）**。
+
+### 1.6 悟空侧类比质检
+
+| 悟空 | 说明 | 本分支 |
+|---|---|---|
+| `scripts/validate-multiskill-bundle.py` | 校验 **已打好的 bundle zip**：frontmatter keys/category、`requires`、markdown 断链、scenario 编排 | **Adapt 思路** → DWS 源树（`skills/multi` + 对照 mono），不跑 zip 安装语义 |
+| `sync-monolith-to-multiskill.py` | mono→multi 派生 | **不**作默认质检手段；DWS 直接维护 multi |
+
+结论：**DWS-native mono↔multi 质检**；悟空仅参考检查维度。
+
+---
+
+## 2. Diff（内容组织 + 质检视角）
+
+### 2.1 已同构
+
+Flat `dingtalk-*` + `dingtalk-shared`；`SKILL.md` + `references/`（+ 可选 `scripts/`）。
+
+### 2.2 分叉与已知内容风险（质检要盯的）
+
+| 风险 ID | 现象（线索） | 质检类型 |
+|---|---|---|
+| **C-cov** | mono `products/*` 能力面在 multi 无对应 skill/reference，或未登记「有意省略」 | 覆盖 |
+| **C-struct** | multi 缺 frontmatter 字段、`references/`、`DWS_RUNTIME_CONTRACT`、对 `dingtalk-shared` 引用不一致 | 结构 |
+| **C-drift-global** | 曾关注 recovery / 确认 / Schema；现确认与 Schema 已在 `dingtalk-shared`，**recovery skill 文档已删除** | 漂移（协议） |
+| **C-drift-orphan** | multi（或 mono）scripts/refs 无文档引用；或 routing 指向无索引产品（留档 X1/M6） | 漂移（孤儿） |
+| **C-pair** | 应对齐的成对文件（如 whiteboard recipes）内容不一致 | 漂移（成对） |
+
+### 2.3 Reject
+
+悟空安装包校验整文件照搬、内容集 19→12 砍产品、安装行为门禁冒充内容质检。
+
+---
+
+## 3. Goals / Non-goals
+
+### 3.1 Goals
+
+1. 固化 multi **内容目录合同**与 mono↔multi **映射说明**。  
+2. 建立 **质检矩阵**（覆盖 / 结构 / 漂移）并以 mono 为对照基准；有意省略必须 reviewed 登记。  
+3. **复用** §1.5 资产；新增门禁走 `scripts/policy` 或 `test/unit`，内容-only。  
+4. （可选）纯内容元数据；**禁止**被安装引擎读取改行为。  
+5. 质检失败 → 修 **内容**或更新「有意省略」表，不改 setup/upgrade。
+
+### 3.2 Non-goals
+
+安装/升级翻转；cherry-pick 行为提交；取消产品；悟空客户端；非 skill CLI 功能；用质检驱动默认 multi 安装。
+
+---
+
+## 4. 分期（内容框架 + 质检 · 均无安装引擎）
+
+> 批准前 **零编码**（含不实现新 gates）。**已执行**：Phase 1–3 见文首状态。
+
+### Phase 0 — 方案冻结（本文）
+
+| | |
+|---|---|
+| **范围** | 本文件；§7（含质检轨）勾选 |
+| **验收** | owner 重新批准 → ✅「现在开始执行」 |
+
+### Phase 1 — Multi 内容目录合同 + 架构短文 ✅
+
+| | |
+|---|---|
+| **范围** | `skills/multi` 目录合同；与悟空内容树对照表；zip `multi/` 同构合同 |
+| **触达** | `docs/skill-content-framework.md` |
+| **验收** | 可指导「如何新增 dingtalk-* 内容目录」 |
+
+### Phase 2 — Mono↔multi **内容质检规格**（矩阵 + 缺口基线） ✅
+
+| | |
+|---|---|
+| **范围** | 质检规格 + 覆盖/omit 机读表 + 缺口 disposition |
+| **触达** | `docs/skill-mono-multi-qa.md`、`skills/content-qa/mono-multi-coverage.yaml` |
+| **验收** | 矩阵可人工抽查；缺口均有 disposition |
+
+### Phase 3 — 质检落地：CI 内容护栏（复用 + 新 gate） ✅
+
+| | |
+|---|---|
+| **范围** | G1–G4 自动门禁 |
+| **触达** | `test/unit/mono_multi_skill_content_test.go`、`scripts/policy/check-mono-multi-skill-content.sh`、`Makefile` |
+| **验收** | `make skill-mono-multi-content` 绿；已知缺口走 reviewed omit |
+
+### Phase 4 — 可选：内容包元数据 + 缺口修复波次
+
+| | |
+|---|---|
+| **范围 A** | 纯内容 layout/skill 列表元数据（人不读安装器） |
+| **范围 B** | 按 Phase 2 disposition **修内容**：确认 / Schema 已补；**recovery skill 文档已删除（wontfix 移植）**；orphan 脚本仍走 allowlist（M4 等） |
+| **验收** | 元数据不驱动安装；修复项关闭对应质检失败或转入 omit |
+
+### 延期登记（非本分支）
+
+| 主题 | 载体 |
+|---|---|
+| 默认 multi + upgrade always-multi | 行为分支 ← `402429ac`/`d5c8982c` |
+| skillhome / 安装面 bootstrap | 行为分支 |
+
+---
+
+## 5. Port / Adapt / Reject
+
+| 项 | 决策 | 说明 |
+|---|---|---|
+| flat + `dingtalk-shared` 内容模型 | **Port** | 已有；合同 + 质检加固 |
+| 悟空 bundle frontmatter/断链/requires 检查维度 | **Adapt** | 做成 DWS 源树门禁，不校验 bundle zip/安装 |
+| whiteboard 式 mono/multi 成对一致 | **Port（范式）** | 推广到 reviewed 文件对 |
+| `validate-multiskill-bundle.py` 整脚本 | **Reject** | 绑定悟空 zip/Qwen 语义 |
+| `_install.sh` / dual / overlay | **Reject** | 非内容 |
+| 行为 cherry-pick | **Defer** | 另分支 |
+
+---
+
+## 6. 与 `402429ac` / `d5c8982c`
+
+| | |
+|---|---|
+| 本分支 cherry-pick？ | **否** |
+| 质检是否替代行为翻转？ | **否** |
+| 行为分支 | 另开；可与内容/质检并行 |
+
+---
+
+## 7. 批准清单（请重新勾选）
+
+**范围**
+
+- [x] 本分支 = skill **内容**框架 + **mono↔multi 内容质检**（§0.1）；无安装/升级引擎  
+- [x] `402429ac`/`d5c8982c` 及 setup/paths/install 脚本行为 **不在本分支**  
+- [x] 取消产品与悟空客户端链路仍拒绝  
+
+**内容框架 Phase**
+
+- [x] **Phase 1**：multi 目录合同 + 悟空内容树对照短文  
+
+**质检轨 Phase**
+
+- [x] **Phase 2**：质检矩阵 + mono↔multi 覆盖/缺口基线规格（先文档，可执行）  
+- [x] **Phase 3**：CI 内容护栏（G1–G4）—— 本迭代做 / 拆 PR / 只要规格暂不落地  
+- [x] 质检失败处置原则：修内容或 reviewed omit，**不**改安装默认  
+
+**可选**
+
+- [ ] **Phase 4A** 纯内容元数据：做 / 不做 / 以后  
+- [x] **Phase 4B** recovery skill 文档 **removed/wontfix**；确认/Schema 已补；剩余 orphan（M4 等）仍 defer / allowlist  
+
+**Follow-up 知悉**
+
+- [ ] 安装默认 multi + upgrade always-multi → **另一分支**  
+
+---
+
+## 8. 下一步
+
+**Phase 1–3 已落地**（合同短文 + 质检规格 + `skills/content-qa` + CI 门禁）。  
+Phase 4B：recovery 已删除（不做移植）；确认/Schema 已补。剩余 defer：orphan scripts（M4 等）、LICENSE/NOTICE（M5）、Phase 4A 元数据。  
+安装默认 multi 等行为仍走 **另一分支**。
+
+---
+
+*锚点：`skills/mono`、`skills/multi`、§1.5 policy/测试、wukong `dingtalk-skills/`（组织对照 only）。*

docs/unified-command-framework-2.0.md

@@ -0,0 +1,108 @@
+# DWS 统一命令框架设计概要
+
+> 状态：Framework core 已实现，dingtalk-dev/devapp 首批命令渐进接入中。本文定义框架能力、集成边界和首批 pilot 的发布纪律；其余产品命令迁移、Skill 更新和真实服务复验继续由后续 PR 独立完成。
+
+## 1. 产品裁决
+
+1. 不公开 `--output-contract`，也不增加任何等价别名。
+2. Agent 继续只使用既有 `--format json`。
+3. 每条 terminal command 在一个 release 中只有一个 active wire contract：已迁移命令直接使用统一结果，未迁移命令保持 legacy。
+4. contract 不由用户参数、环境变量、会话能力协商或 Agent 选择。
+5. 回滚是命令声明与发布行为，不改变消费者 argv。
+6. 本 PR 只迁移完成命令级兼容审计的 dingtalk-dev/devapp pilot；其他命令路径、参数和输出保持不变。
+
+## 2. 渐进迁移
+
+内部状态机：
+
+```text
+legacy_only -> dual_validate -> unified_active -> unified_stable -> unified_only
+```
+
+- `legacy_only`：只构造、输出 legacy。
+- `dual_validate`：业务只执行一次；外部仍逐字输出 legacy；同一内存结果 shadow-build 统一结果并严格校验。
+- `unified_active`：`--format json` 直接返回统一结果信封，可按发布声明回退。
+- `unified_stable`：完成真实 Agent 消费观察和兼容窗口。
+- `unified_only`：清理仅服务 legacy 的产品 renderer。
+
+状态是每条 terminal command 的内部发布元数据。Help、Skill、Agent Schema 不展示迁移状态，也不让消费者选择协议。
+
+## 3. 统一结果
+
+统一命令框架表达四类结果：
+
+```text
+success          请求完成且命令认为操作已完成
+pending          请求被受理，但异步操作尚未终结
+partial_failure  批量操作有成功项，也有失败或未知项
+failure          请求或操作失败
+```
+
+JSON 基本形态：
+
+```json
+{
+  "ok": true,
+  "outcome": "success",
+  "data": {}
+}
+```
+
+硬不变量：
+
+```text
+ok == (outcome in {success, pending})
+process rc == 0 <=> ok == true
+top-level error present <=> outcome == failure
+one invocation emits exactly one primary result
+```
+
+框架负责 L1 request outcome 和 L2 operation outcome 的统一表达；L3 verification 必须由产品命令基于业务事实实现，框架不得自动推断 `changed/verified`。
+
+## 4. 输出与错误纪律
+
+- 统一 JSON primary result 写 stdout；stderr 只写诊断。普通命令不把
+  NDJSON 作为通用结果契约；持续事件流若需要逐事件输出，由 event 命令
+  自己声明专用流协议。
+- 分页统一输出到信封 `meta.pagination`，并在命令 Schema 中作为与 `result`
+  同级的 `pagination` 能力声明；`result.data_schema` 只描述业务 data，不再
+  混入分页控制字段。
+- 日志不得污染 stdout。
+- `ok`、`retryable`、`dry_run` 等必须是 JSON boolean。
+- 失败由框架根据 typed error 映射退出码；产品代码不能自报任意 rc。
+- `partial_failure` 保留 `succeeded[]/failed[]/unknown[]`，使用非零 rc 7。
+- `pending` 必须提供 operation id、state 和可执行的 `next_command`。
+- `endpoint_exhausted` 只表示观察到当前 endpoint 分页耗尽；false 必须带 `next_token`，不得扩大成索引健康或业务数据完整。
+- dry-run 是已经完成的无副作用预览，表达为 `success + dry_run:true`，不是 `pending`。
+
+## 5. 重试与超时边界
+
+- 框架只统一表达 `retryable`、`retry_after_seconds` 和 `execution_started`，不自动决定业务操作能否安全重放。
+- 写调用的模糊失败、HTTP timeout 和异步等待预算属于 transport/产品集成范围，不在本 PR 改动。
+- 产品迁移必须证明其重试声明与幂等性、安全等级一致。
+
+## 6. 集成范围
+
+- 产品命令通过 `corecmd.ResultInvoke` 构造 `CommandResult`，由 root 单一出口渲染。
+- 首批 dingtalk-dev/devapp 命令用于验证原子命令与 shortcut 的接入缝；未进入 pilot 的 shortcut、长连接、批量写和异步任务各自需要独立集成 PR。框架 core 不替产品推断 success、pending、partial 或分页事实。
+- 每条 terminal command 独立 rollout；不能整域一次切换，也不能通过 Agent 参数选择协议。
+- 已有命令在进入 `unified_active` 前必须保留 legacy byte golden，并完成真实 Agent 语义扫描。
+
+## 7. 对齐原则
+
+- 对齐 Lark CLI：统一 envelope/emitter、typed error、partial、pending、分页窄语义和强类型结果。
+- 对齐 GWS：机器结果稳定结构化、日志与数据分流、消费者不协商协议版本。
+- DWS 保留差异：声明式 Agent Schema、安全门禁、静态命令与 shortcut 共存，以及四 outcome 模型。
+
+## 8. 发布门禁
+
+命令晋级 `unified_active` 前至少满足：
+
+1. success/failure/dry-run golden；批量或异步命令另有 partial/pending golden。
+2. 业务请求 exactly once；dual validation 不得二次调用服务端。
+3. legacy 命令 stdout/stderr/rc 字节级回归不变。
+4. Help、Schema 和全仓示例不存在协议选择参数。
+5. `--format json` 输出单个合法统一结果文档，stdout 无日志污染。
+6. typed error、进程 rc 与信封 `error.exit_code` 一致。
+7. 安全声明、确认门禁与 dry-run 运行时行为同源。
+8. Agent 语义扫描记录命令级迁移证据；发布回滚无需修改 Agent argv。

go.mod

@@ -3,15 +3,18 @@ module github.com/DingTalk-Real-AI/dingtalk-workspace-cli
 go 1.25.9
 
 require (
+	github.com/Microsoft/go-winio v0.6.2
 	github.com/RealAlexandreAI/json-repair v0.0.15
 	github.com/charmbracelet/bubbletea v1.3.6
 	github.com/charmbracelet/huh v1.0.0
 	github.com/charmbracelet/lipgloss v1.1.0
 	github.com/fatih/color v1.18.0
 	github.com/google/uuid v1.6.0
+	github.com/gorilla/websocket v1.5.0
 	github.com/itchyny/gojq v0.12.18
+	github.com/mattn/go-isatty v0.0.20
 	github.com/muesli/termenv v0.16.0
-	github.com/open-dingtalk/dingtalk-stream-sdk-go v0.9.1
+	github.com/open-dingtalk/dingtalk-stream-sdk-go v0.9.2-0.20260705041131-325e7c1049ad
 	github.com/spf13/cobra v1.10.2
 	github.com/zalando/go-keyring v0.2.8
 	golang.org/x/crypto v0.49.0
@@ -35,11 +38,9 @@ require (
 	github.com/dustin/go-humanize v1.0.1 // indirect
 	github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f // indirect
 	github.com/godbus/dbus/v5 v5.2.2 // indirect
-	github.com/gorilla/websocket v1.5.0 // indirect
 	github.com/itchyny/timefmt-go v0.1.7 // indirect
 	github.com/lucasb-eyer/go-colorful v1.2.0 // indirect
 	github.com/mattn/go-colorable v0.1.13 // indirect
-	github.com/mattn/go-isatty v0.0.20 // indirect
 	github.com/mattn/go-localereader v0.0.1 // indirect
 	github.com/mattn/go-runewidth v0.0.19 // indirect
 	github.com/mitchellh/hashstructure/v2 v2.0.2 // indirect

go.sum

@@ -1,5 +1,7 @@
 github.com/MakeNowJust/heredoc v1.0.0 h1:cXCdzVdstXyiTqTvfqk9SDHpKNjxuom+DOlyEeQ4pzQ=
 github.com/MakeNowJust/heredoc v1.0.0/go.mod h1:mG5amYoWBHf8vpLOuehzbGGw0EHxpZZ6lCpQ4fNJ8LE=
+github.com/Microsoft/go-winio v0.6.2 h1:F2VQgta7ecxGYO8k3ZZz3RS8fVIXVxONVUPlNERoyfY=
+github.com/Microsoft/go-winio v0.6.2/go.mod h1:yd8OoFMLzJbo9gZq8j5qaps8bJ9aShtEA8Ipt1oGCvU=
 github.com/RealAlexandreAI/json-repair v0.0.15 h1:AN8/yt8rcphwQrIs/FZeki+cKaIERUNr25zf1flirIs=
 github.com/RealAlexandreAI/json-repair v0.0.15/go.mod h1:GKJi5borR78O8c7HCVbgqjhoiVibZ6hJldxbc6dGrAI=
 github.com/atotto/clipboard v0.1.4 h1:EH0zSVneZPSuFR11BlR9YppQTVDbh5+16AmcJi4g1z4=
@@ -86,8 +88,8 @@ github.com/muesli/cancelreader v0.2.2 h1:3I4Kt4BQjOR54NavqnDogx/MIoWBFa0StPA8ELU
 github.com/muesli/cancelreader v0.2.2/go.mod h1:3XuTXfFS2VjM+HTLZY9Ak0l6eUKfijIfMUZ4EgX0QYo=
 github.com/muesli/termenv v0.16.0 h1:S5AlUN9dENB57rsbnkPyfdGuWIlkmzJjbFf0Tf5FWUc=
 github.com/muesli/termenv v0.16.0/go.mod h1:ZRfOIKPFDYQoDFF4Olj7/QJbW60Ol/kL1pU3VfY/Cnk=
-github.com/open-dingtalk/dingtalk-stream-sdk-go v0.9.1 h1:Lb/Uzkiw2Ugt2Xf03J5wmv81PdkYOiWbI8CNBi1boC8=
-github.com/open-dingtalk/dingtalk-stream-sdk-go v0.9.1/go.mod h1:ln3IqPYYocZbYvl9TAOrG/cxGR9xcn4pnZRLdCTEGEU=
+github.com/open-dingtalk/dingtalk-stream-sdk-go v0.9.2-0.20260705041131-325e7c1049ad h1:Bb4I+suYd+ehQ8e22aimLLze+5XTN3+WTc/x2LafmH8=
+github.com/open-dingtalk/dingtalk-stream-sdk-go v0.9.2-0.20260705041131-325e7c1049ad/go.mod h1:ln3IqPYYocZbYvl9TAOrG/cxGR9xcn4pnZRLdCTEGEU=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/rivo/uniseg v0.4.7 h1:WUdvkW8uEhrYfLC4ZzdpI2ztxP1I582+49Oc5Mq64VQ=

internal/apiclient/client.go

@@ -49,6 +49,8 @@ var AllowedMethods = map[string]bool{
 	"GET": true, "POST": true, "PUT": true, "PATCH": true, "DELETE": true,
 }
 
+var newHTTPRequest = http.NewRequestWithContext
+
 // RawAPIRequest describes a raw API request to DingTalk OpenAPI.
 type RawAPIRequest struct {
 	Method string         // GET, POST, PUT, PATCH, DELETE
@@ -112,7 +114,7 @@ func (c *APIClient) Do(ctx context.Context, req RawAPIRequest) (*RawAPIResponse,
 		bodyReader = bytes.NewReader(data)
 	}
 
-	httpReq, err := http.NewRequestWithContext(ctx, method, fullURL, bodyReader)
+	httpReq, err := newHTTPRequest(ctx, method, fullURL, bodyReader)
 	if err != nil {
 		return nil, fmt.Errorf("creating HTTP request: %w", err)
 	}

internal/apiclient/coverage_edges_test.go

@@ -0,0 +1,293 @@
+package apiclient
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"io"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/output"
+)
+
+type failingReader struct{ err error }
+
+func (r failingReader) Read([]byte) (int, error) { return 0, r.err }
+
+type failingWriter struct{ err error }
+
+func (w failingWriter) Write([]byte) (int, error) { return 0, w.err }
+
+func TestCrossPlatformCoverageDryRunAndParseCoverageEdges(t *testing.T) {
+	for _, tc := range []struct {
+		base string
+		path string
+	}{
+		{DefaultBaseURL, "/v1.0/test"},
+		{LegacyBaseURL, "/topapi/test"},
+	} {
+		var out bytes.Buffer
+		err := PrintDryRun(&out, RawAPIRequest{
+			Method: "post", Path: tc.path,
+			Params: map[string]any{"page": 1}, Data: map[string]any{"name": "value"},
+		}, tc.base, "token-value")
+		if err != nil || !strings.Contains(out.String(), "Dry Run") || !strings.Contains(out.String(), "toke****") {
+			t.Fatalf("PrintDryRun(%s) = %q, %v", tc.base, out.String(), err)
+		}
+	}
+	var out bytes.Buffer
+	if err := PrintDryRun(&out, RawAPIRequest{Method: "get", Path: "/x", Params: map[string]any{"bad": make(chan int)}, Data: make(chan int)}, DefaultBaseURL, "tiny"); err != nil {
+		t.Fatalf("PrintDryRun unsupported preview: %v", err)
+	}
+
+	wantErr := errors.New("read failed")
+	if _, err := ParseJSONMap("-", "--params", failingReader{err: wantErr}); !errors.Is(err, wantErr) {
+		t.Fatalf("ParseJSONMap read error = %v", err)
+	}
+	if got, err := ParseJSONMap("-", "--params", strings.NewReader(" \n")); err != nil || got != nil {
+		t.Fatalf("ParseJSONMap empty stdin = %#v, %v", got, err)
+	}
+	if _, err := ParseOptionalBody("POST", "-", failingReader{err: wantErr}); !errors.Is(err, wantErr) {
+		t.Fatalf("ParseOptionalBody read error = %v", err)
+	}
+	if got, err := ParseOptionalBody("POST", "-", strings.NewReader(" \n")); err != nil || got != nil {
+		t.Fatalf("ParseOptionalBody empty stdin = %#v, %v", got, err)
+	}
+	if _, err := ParseOptionalBody("POST", "{", strings.NewReader("")); err == nil {
+		t.Fatal("invalid optional body should fail")
+	}
+}
+
+func TestCrossPlatformCoverageResponseHandlingCoverageEdges(t *testing.T) {
+	jsonHeader := http.Header{"Content-Type": []string{"application/json"}}
+	textHeader := http.Header{"Content-Type": []string{"text/plain"}}
+	var out, errOut bytes.Buffer
+	opts := ResponseOptions{Format: output.FormatJSON, Out: &out, ErrOut: &errOut}
+
+	if err := HandleResponse(&RawAPIResponse{StatusCode: 500, Header: textHeader, Body: []byte(" failed ")}, opts); err == nil {
+		t.Fatal("plain HTTP error should fail")
+	}
+	for _, body := range [][]byte{nil, []byte("{")} {
+		if err := HandleResponse(&RawAPIResponse{StatusCode: 200, Header: jsonHeader, Body: body}, opts); err == nil {
+			t.Errorf("invalid JSON body %q should fail", body)
+		}
+	}
+	out.Reset()
+	if err := HandleResponse(&RawAPIResponse{StatusCode: 200, Header: jsonHeader, Body: []byte(`{"ok":true}`)}, opts); err != nil || !strings.Contains(out.String(), "ok") {
+		t.Fatalf("successful JSON response = %q, %v", out.String(), err)
+	}
+	for _, payload := range []string{
+		`{"errcode":1}`,
+		`{"message":"message failure"}`,
+		`{"error":"error failure"}`,
+		`{}`,
+	} {
+		status := 200
+		if !strings.Contains(payload, "errcode") {
+			status = 500
+		}
+		if err := HandleResponse(&RawAPIResponse{StatusCode: status, Header: jsonHeader, Body: []byte(payload)}, opts); err == nil {
+			t.Errorf("business/HTTP payload %s should fail", payload)
+		}
+	}
+	if err := checkDingTalkError([]any{1}, 200); err != nil || checkDingTalkError(map[string]any{"errcode": 0}, 200) != nil {
+		t.Fatal("successful DingTalk response classified as error")
+	}
+
+	if err := HandleResponse(&RawAPIResponse{StatusCode: 200, Header: textHeader, Body: []byte("binary")}, opts); err == nil {
+		t.Fatal("binary response without filename should fail")
+	}
+	invalidCD := http.Header{"Content-Type": []string{"application/octet-stream"}, "Content-Disposition": []string{`attachment; filename="unterminated`}}
+	if inferFilename(invalidCD) != "" {
+		t.Fatal("invalid content disposition should not infer filename")
+	}
+	if inferFilename(http.Header{}) != "" {
+		t.Fatal("missing content disposition should not infer filename")
+	}
+
+	dir := t.TempDir()
+	blockedParent := filepath.Join(dir, "file")
+	if err := os.WriteFile(blockedParent, []byte("x"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	opts.OutputPath = filepath.Join(blockedParent, "child.bin")
+	if err := handleBinaryResponse(&RawAPIResponse{Header: textHeader, Body: []byte("x")}, opts); err == nil {
+		t.Fatal("binary mkdir failure should fail")
+	}
+	opts.OutputPath = dir
+	if err := handleBinaryResponse(&RawAPIResponse{Header: textHeader, Body: []byte("x")}, opts); err == nil {
+		t.Fatal("binary write to directory should fail")
+	}
+	opts.OutputPath = ""
+	inferred := filepath.Join(dir, "inferred.bin")
+	header := http.Header{"Content-Type": []string{"application/octet-stream"}, "Content-Disposition": []string{`attachment; filename="` + inferred + `"`}}
+	if err := handleBinaryResponse(&RawAPIResponse{Header: header, Body: []byte("bytes")}, opts); err != nil {
+		t.Fatalf("inferred binary save: %v", err)
+	}
+	if !strings.Contains(errOut.String(), "已保存") {
+		t.Fatalf("binary status = %q", errOut.String())
+	}
+
+	for _, ct := range []string{" application/json; charset=utf-8 ", "text/json", "application/problem+json", "text/plain"} {
+		_ = isJSONContentType(ct)
+	}
+	for _, value := range []any{float64(1), 2, int64(3), json.Number("4"), json.Number("bad"), "5"} {
+		_ = toFloat64(value)
+	}
+}
+
+func TestCrossPlatformCoveragePaginationParsingAndInjectionEdges(t *testing.T) {
+	jsonHeader := http.Header{"Content-Type": []string{"application/json"}}
+	for _, resp := range []*RawAPIResponse{
+		{StatusCode: 200, Header: http.Header{"Content-Type": []string{"text/plain"}}, Body: []byte("x")},
+		{StatusCode: 200, Header: jsonHeader},
+		{StatusCode: 200, Header: jsonHeader, Body: []byte("{")},
+		{StatusCode: 500, Header: jsonHeader, Body: []byte(`{"message":"bad"}`)},
+	} {
+		if _, _, _, err := parsePaginatedResponse(resp); err == nil {
+			t.Errorf("parsePaginatedResponse(%#v) should fail", resp)
+		}
+	}
+	responses := []struct {
+		body  string
+		more  bool
+		token string
+	}{
+		{`{"result":{"has_more":true,"next_cursor":12}}`, true, "12"},
+		{`{"has_more":true,"next_cursor":13}`, true, "13"},
+		{`{"next_token":"next"}`, true, "next"},
+		{`{"result":[],"has_more":false}`, false, ""},
+	}
+	for _, tc := range responses {
+		_, more, token, err := parsePaginatedResponse(&RawAPIResponse{StatusCode: 200, Header: jsonHeader, Body: []byte(tc.body)})
+		if err != nil || more != tc.more || token != tc.token {
+			t.Errorf("pagination %s = %v, %q, %v", tc.body, more, token, err)
+		}
+	}
+
+	getCases := []RawAPIRequest{
+		{Method: "GET"},
+		{Method: "GET", Params: map[string]any{"cursor": "old"}},
+		{Method: "GET", Params: map[string]any{"next_token": "old"}},
+		{Method: "POST", Data: map[string]any{"cursor": "old"}},
+		{Method: "PUT", Data: map[string]any{}},
+		{Method: "POST", Data: "not-a-map"},
+	}
+	for _, req := range getCases {
+		_ = injectPageToken(req, "new")
+	}
+	logf(nil, "ignored")
+}
+
+type roundTripFunc func(*http.Request) (*http.Response, error)
+
+func (f roundTripFunc) RoundTrip(req *http.Request) (*http.Response, error) { return f(req) }
+
+func jsonHTTPResponse(body string) *http.Response {
+	return &http.Response{StatusCode: 200, Header: http.Header{"Content-Type": []string{"application/json"}}, Body: io.NopCloser(strings.NewReader(body))}
+}
+
+func TestCrossPlatformCoveragePaginationControlFlowEdges(t *testing.T) {
+	wantErr := errors.New("transport failed")
+	client := NewClient("token", DefaultBaseURL)
+	client.HTTPClient.Transport = roundTripFunc(func(*http.Request) (*http.Response, error) {
+		return &http.Response{StatusCode: 200, Header: http.Header{"Content-Type": []string{"text/plain"}}, Body: io.NopCloser(strings.NewReader("bad"))}, nil
+	})
+	if _, err := client.PaginateAll(context.Background(), RawAPIRequest{Method: "GET", Path: "/x"}, PaginationOptions{}); err == nil {
+		t.Fatal("first page parse error should fail")
+	}
+	client.HTTPClient.Transport = roundTripFunc(func(*http.Request) (*http.Response, error) { return nil, wantErr })
+	if _, err := client.PaginateAll(context.Background(), RawAPIRequest{Method: "GET", Path: "/x"}, PaginationOptions{}); !errors.Is(err, wantErr) {
+		t.Fatalf("first page transport error = %v", err)
+	}
+
+	calls := 0
+	client.HTTPClient.Transport = roundTripFunc(func(*http.Request) (*http.Response, error) {
+		calls++
+		if calls == 1 {
+			return jsonHTTPResponse(`{"next_token":"next"}`), nil
+		}
+		return nil, wantErr
+	})
+	if pages, err := client.PaginateAll(context.Background(), RawAPIRequest{Method: "GET", Path: "/x"}, PaginationOptions{PageDelay: 1}); err == nil || len(pages) != 1 {
+		t.Fatalf("later transport error pages=%d err=%v", len(pages), err)
+	}
+
+	calls = 0
+	var logs bytes.Buffer
+	client.HTTPClient.Transport = roundTripFunc(func(*http.Request) (*http.Response, error) {
+		calls++
+		if calls == 1 {
+			return jsonHTTPResponse(`{"next_token":"next"}`), nil
+		}
+		return &http.Response{StatusCode: 200, Header: http.Header{"Content-Type": []string{"text/plain"}}, Body: io.NopCloser(strings.NewReader("bad"))}, nil
+	})
+	if pages, err := client.PaginateAll(context.Background(), RawAPIRequest{Method: "GET", Path: "/x"}, PaginationOptions{PageDelay: 1, LogWriter: &logs}); err != nil || len(pages) != 1 || !strings.Contains(logs.String(), "解析失败") {
+		t.Fatalf("later parse failure pages=%d logs=%q err=%v", len(pages), logs.String(), err)
+	}
+
+	client.HTTPClient.Transport = roundTripFunc(func(*http.Request) (*http.Response, error) {
+		return jsonHTTPResponse(`{"next_token":"next"}`), nil
+	})
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+	if pages, err := client.PaginateAll(ctx, RawAPIRequest{Method: "GET", Path: "/x"}, PaginationOptions{PageDelay: 10}); !errors.Is(err, context.Canceled) || len(pages) != 1 {
+		t.Fatalf("pagination cancellation pages=%d err=%v", len(pages), err)
+	}
+	logs.Reset()
+	if pages, err := client.PaginateAll(context.Background(), RawAPIRequest{Method: "GET", Path: "/x"}, PaginationOptions{PageLimit: 1, PageDelay: 1, LogWriter: &logs}); err != nil || len(pages) != 1 || !strings.Contains(logs.String(), "安全上限") {
+		t.Fatalf("pagination safety cap pages=%d logs=%q err=%v", len(pages), logs.String(), err)
+	}
+}
+
+func TestCrossPlatformCoverageClientAndValidationFailureEdges(t *testing.T) {
+	client := NewClient("token", DefaultBaseURL)
+	if _, err := client.Do(context.Background(), RawAPIRequest{Method: "GET", Path: "https://api.dingtalk.com/%zz"}); err == nil {
+		t.Fatal("Do with invalid URL should fail")
+	}
+	if _, err := client.Do(context.Background(), RawAPIRequest{Method: "GET", Path: "https://example.test/x"}); err == nil {
+		t.Fatal("Do to untrusted host should fail")
+	}
+	if _, err := client.Do(context.Background(), RawAPIRequest{Method: "POST", Path: "/x", Data: make(chan int)}); err == nil {
+		t.Fatal("unmarshalable request body should fail")
+	}
+	if _, err := client.buildURL("https://api.dingtalk.com/%zz", nil); err == nil {
+		t.Fatal("invalid URL should fail")
+	}
+	oldNewRequest := newHTTPRequest
+	t.Cleanup(func() { newHTTPRequest = oldNewRequest })
+	wantCreateErr := errors.New("request creation failed")
+	newHTTPRequest = func(context.Context, string, string, io.Reader) (*http.Request, error) { return nil, wantCreateErr }
+	if _, err := client.Do(context.Background(), RawAPIRequest{Method: "GET", Path: "/x"}); !errors.Is(err, wantCreateErr) {
+		t.Fatalf("request creation error = %v", err)
+	}
+	newHTTPRequest = oldNewRequest
+	wantErr := errors.New("request failed")
+	client.HTTPClient.Transport = roundTripFunc(func(*http.Request) (*http.Response, error) { return nil, wantErr })
+	if _, err := client.Do(context.Background(), RawAPIRequest{Method: "GET", Path: "/x"}); !errors.Is(err, wantErr) {
+		t.Fatalf("HTTP transport error = %v", err)
+	}
+	client.HTTPClient.Transport = roundTripFunc(func(*http.Request) (*http.Response, error) {
+		return &http.Response{StatusCode: 200, Header: http.Header{}, Body: io.NopCloser(failingReader{err: wantErr})}, nil
+	})
+	if _, err := client.Do(context.Background(), RawAPIRequest{Method: "GET", Path: "/x"}); !errors.Is(err, wantErr) {
+		t.Fatalf("response read error = %v", err)
+	}
+
+	if ValidateTargetHost("http://%zz") == nil {
+		t.Fatal("invalid target URL should fail")
+	}
+	for _, r := range []rune{0x200B, 0xFEFF, 0x202A, 0x2028, 0x2066, 0x061C, 0xFDD0} {
+		if !isDangerousUnicode(r) || ValidateUserInput("x"+string(r), "field") == nil {
+			t.Errorf("dangerous rune %U was accepted", r)
+		}
+	}
+	if isDangerousUnicode('中') || ValidateUserInput("safe\t\n中文", "field") != nil {
+		t.Fatal("safe Unicode/input was rejected")
+	}
+}

internal/app/access_token_manager_test.go

@@ -0,0 +1,223 @@
+package app
+
+import (
+	"context"
+	"errors"
+	"os"
+	"path/filepath"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	authpkg "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/auth"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
+)
+
+type tokenManagerSnapshotProvider struct {
+	load func() (*authpkg.TokenData, error)
+}
+
+func (p tokenManagerSnapshotProvider) GetAccessToken(context.Context) (string, error) {
+	data, err := p.load()
+	if err != nil || data == nil {
+		return "", err
+	}
+	return data.AccessToken, nil
+}
+
+func (p tokenManagerSnapshotProvider) GetTokenSnapshot(context.Context) (*authpkg.TokenData, error) {
+	return p.load()
+}
+
+type tokenManagerLegacyGetter struct {
+	token string
+	err   error
+}
+
+func (g tokenManagerLegacyGetter) GetToken() (string, string, error) {
+	return g.token, "file", g.err
+}
+
+func installTokenManagerFakes(t *testing.T, load func() (*authpkg.TokenData, error)) {
+	t.Helper()
+	oldProvider, oldLegacy := newAccessTokenProvider, newLegacyTokenManager
+	oldEdition := edition.Get()
+	edition.Override(&edition.Hooks{})
+	newAccessTokenProvider = func(string) accessTokenGetter {
+		return tokenManagerSnapshotProvider{load: load}
+	}
+	newLegacyTokenManager = func(string) legacyTokenGetter {
+		return tokenManagerLegacyGetter{err: authpkg.ErrTokenDataNotFound}
+	}
+	t.Cleanup(func() {
+		newAccessTokenProvider, newLegacyTokenManager = oldProvider, oldLegacy
+		edition.Override(oldEdition)
+	})
+}
+
+func TestCrossPlatformCoverageTokenManagerCachesUntilMarkerRevisionChanges(t *testing.T) {
+	configDir := t.TempDir()
+	if err := authpkg.WriteTokenMarker(configDir); err != nil {
+		t.Fatal(err)
+	}
+	var calls atomic.Int32
+	token := "token-a"
+	installTokenManagerFakes(t, func() (*authpkg.TokenData, error) {
+		calls.Add(1)
+		return &authpkg.TokenData{AccessToken: token, ExpiresAt: time.Now().Add(time.Hour)}, nil
+	})
+
+	manager := NewTokenManager()
+	first, err := manager.Get(context.Background(), configDir, "")
+	if err != nil || first.AccessToken != "token-a" {
+		t.Fatalf("first token = %#v, %v", first, err)
+	}
+	second, err := manager.Get(context.Background(), configDir, "")
+	if err != nil || second.AccessToken != "token-a" || calls.Load() != 1 {
+		t.Fatalf("cached token = %#v, %v, calls=%d", second, err, calls.Load())
+	}
+
+	token = "token-b"
+	if err := authpkg.WriteTokenMarker(configDir); err != nil {
+		t.Fatal(err)
+	}
+	rotated, err := manager.Get(context.Background(), configDir, "")
+	if err != nil || rotated.AccessToken != "token-b" || calls.Load() != 2 {
+		t.Fatalf("rotated token = %#v, %v, calls=%d", rotated, err, calls.Load())
+	}
+}
+
+func TestCrossPlatformCoverageTokenManagerDoesNotCacheWithoutExpiryOrRevision(t *testing.T) {
+	configDir := t.TempDir()
+	var calls atomic.Int32
+	installTokenManagerFakes(t, func() (*authpkg.TokenData, error) {
+		calls.Add(1)
+		return &authpkg.TokenData{AccessToken: "token"}, nil
+	})
+	manager := NewTokenManager()
+	for range 2 {
+		if _, err := manager.Get(context.Background(), configDir, ""); err != nil {
+			t.Fatal(err)
+		}
+	}
+	if calls.Load() != 2 {
+		t.Fatalf("provider calls = %d, want 2", calls.Load())
+	}
+}
+
+func TestCrossPlatformCoverageTokenManagerTreatsMalformedMarkerAsUncacheable(t *testing.T) {
+	configDir := t.TempDir()
+	if err := os.WriteFile(filepath.Join(configDir, "token.json"), []byte("{"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	var calls atomic.Int32
+	installTokenManagerFakes(t, func() (*authpkg.TokenData, error) {
+		calls.Add(1)
+		return &authpkg.TokenData{AccessToken: "token", ExpiresAt: time.Now().Add(time.Hour)}, nil
+	})
+	manager := NewTokenManager()
+	for range 2 {
+		if snapshot, err := manager.Get(context.Background(), configDir, ""); err != nil || snapshot.AccessToken != "token" {
+			t.Fatalf("snapshot = %#v, error = %v", snapshot, err)
+		}
+	}
+	if calls.Load() != 2 {
+		t.Fatalf("provider calls = %d, want 2", calls.Load())
+	}
+}
+
+func TestCrossPlatformCoverageTokenManagerDoesNotCacheOpaqueEditionStorageWithProviderFallback(t *testing.T) {
+	configDir := t.TempDir()
+	if err := authpkg.WriteTokenMarker(configDir); err != nil {
+		t.Fatal(err)
+	}
+	var calls atomic.Int32
+	installTokenManagerFakes(t, func() (*authpkg.TokenData, error) {
+		calls.Add(1)
+		return &authpkg.TokenData{AccessToken: "token", ExpiresAt: time.Now().Add(time.Hour)}, nil
+	})
+	edition.Override(&edition.Hooks{
+		LoadToken: func(string) ([]byte, error) { return nil, nil },
+		TokenProvider: func(_ context.Context, fallback func() (string, error)) (string, error) {
+			return fallback()
+		},
+	})
+	manager := NewTokenManager()
+	for range 2 {
+		if _, err := manager.Get(context.Background(), configDir, ""); err != nil {
+			t.Fatal(err)
+		}
+	}
+	if calls.Load() != 2 {
+		t.Fatalf("provider calls = %d, want 2", calls.Load())
+	}
+}
+
+func TestCrossPlatformCoverageTokenManagerCoalescesConcurrentLoads(t *testing.T) {
+	configDir := t.TempDir()
+	if err := authpkg.WriteTokenMarker(configDir); err != nil {
+		t.Fatal(err)
+	}
+	var calls atomic.Int32
+	release := make(chan struct{})
+	installTokenManagerFakes(t, func() (*authpkg.TokenData, error) {
+		calls.Add(1)
+		<-release
+		return &authpkg.TokenData{AccessToken: "token", ExpiresAt: time.Now().Add(time.Hour)}, nil
+	})
+	manager := NewTokenManager()
+	const workers = 8
+	var wg sync.WaitGroup
+	wg.Add(workers)
+	errs := make(chan error, workers)
+	for range workers {
+		go func() {
+			defer wg.Done()
+			_, err := manager.Get(context.Background(), configDir, "")
+			errs <- err
+		}()
+	}
+	for calls.Load() == 0 {
+		time.Sleep(time.Millisecond)
+	}
+	close(release)
+	wg.Wait()
+	close(errs)
+	for err := range errs {
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+	if calls.Load() != 1 {
+		t.Fatalf("provider calls = %d, want 1", calls.Load())
+	}
+}
+
+func TestCrossPlatformCoverageTokenManagerPreservesProviderFailure(t *testing.T) {
+	configDir := t.TempDir()
+	want := errors.New("keychain permission denied")
+	installTokenManagerFakes(t, func() (*authpkg.TokenData, error) { return nil, want })
+	_, err := NewTokenManager().Get(context.Background(), configDir, "")
+	if !errors.Is(err, want) {
+		t.Fatalf("error = %v, want cause %v", err, want)
+	}
+	if errors.Is(err, authpkg.ErrTokenDataNotFound) {
+		t.Fatalf("provider failure was misclassified as missing credentials: %v", err)
+	}
+}
+
+func TestCrossPlatformCoverageTokenResolutionErrorOnlyClassifiesTrueMissingCredential(t *testing.T) {
+	missing := tokenResolutionError(authpkg.ErrTokenDataNotFound)
+	var typed interface{ Unwrap() error }
+	if !errors.As(missing, &typed) || !errors.Is(missing, authpkg.ErrTokenDataNotFound) {
+		t.Fatalf("missing error = %v", missing)
+	}
+	want := errors.New("decrypt failed")
+	if got := tokenResolutionError(want); !errors.Is(got, want) || errors.Is(got, authpkg.ErrTokenDataNotFound) {
+		t.Fatalf("storage error = %v", got)
+	}
+	if got := tokenResolutionError(context.Canceled); !errors.Is(got, context.Canceled) {
+		t.Fatalf("cancellation = %v", got)
+	}
+}

internal/app/access_token_resolve.go

@@ -21,63 +21,283 @@ import (
 	"log/slog"
 	"path/filepath"
 	"strings"
+	"sync"
+	"time"
 
 	authpkg "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/auth"
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
 )
 
-// resolveAccessTokenFromDir loads OAuth then legacy token from configDir, applying
-// the same host compatibility hooks as MCP. It mirrors the former body of
-// getCachedRuntimeToken (excluding process-level cache and timing).
-func resolveAccessTokenFromDir(ctx context.Context, configDir string) (string, error) {
-	disc := slog.New(slog.NewTextHandler(io.Discard, nil))
-	provider := authpkg.NewOAuthProvider(configDir, disc)
-	configureOAuthProviderCompatibility(provider, configDir)
-	token, tokenErr := provider.GetAccessToken(ctx)
-	if tokenErr == nil && strings.TrimSpace(token) != "" {
-		return strings.TrimSpace(token), nil
-	}
-	if tokenErr != nil && errors.Is(tokenErr, authpkg.ErrTokenDecryption) {
-		return "", tokenErr
-	}
-	manager := authpkg.NewManager(configDir, nil)
-	configureLegacyAuthManagerCompatibility(manager)
-	if leg, _, err := manager.GetToken(); err == nil && strings.TrimSpace(leg) != "" {
-		return strings.TrimSpace(leg), nil
-	}
-	return "", nil
+const accessTokenRefreshWindow = 5 * time.Minute
+
+type legacyTokenGetter interface {
+	GetToken() (string, string, error)
 }
 
-// ResolveAuxiliaryAccessToken resolves a bearer token for HTTP clients that should
-// align with MCP tool calls. Non-empty explicitToken wins. When configDir matches
-// the active edition config directory, the same process-cached path as MCP is used.
-// Otherwise tokens are loaded from configDir with host compatibility hooks applied.
-func ResolveAuxiliaryAccessToken(ctx context.Context, configDir, explicitToken string) (string, error) {
-	if t := strings.TrimSpace(explicitToken); t != "" {
-		return t, nil
+type accessTokenSnapshotGetter interface {
+	GetTokenSnapshot(context.Context) (*authpkg.TokenData, error)
+}
+
+// AccessTokenSnapshot is the minimal bearer view needed by the process cache.
+// Refresh-token material never leaves the auth package.
+type AccessTokenSnapshot struct {
+	AccessToken string
+	ExpiresAt   time.Time
+	Source      string
+}
+
+type tokenManagerKey struct {
+	configDir string
+	profile   string
+}
+
+type tokenManagerEntry struct {
+	mu       sync.Mutex
+	snapshot AccessTokenSnapshot
+	revision string
+}
+
+// TokenManager is the only process cache for user access tokens. Cache entries
+// are isolated by config directory and profile, expiry-aware, and invalidated
+// by the credential publication marker written by auth storage.
+type TokenManager struct {
+	mu      sync.Mutex
+	entries map[tokenManagerKey]*tokenManagerEntry
+	now     func() time.Time
+}
+
+func NewTokenManager() *TokenManager {
+	return &TokenManager{entries: make(map[tokenManagerKey]*tokenManagerEntry), now: time.Now}
+}
+
+var runtimeTokenManager = NewTokenManager()
+
+var (
+	newAccessTokenProvider = func(configDir string) accessTokenGetter {
+		discard := slog.New(slog.NewTextHandler(io.Discard, nil))
+		provider := authpkg.NewOAuthProvider(configDir, discard)
+		configureOAuthProviderCompatibility(provider, configDir)
+		return provider
+	}
+	newLegacyTokenManager = func(configDir string) legacyTokenGetter {
+		manager := authpkg.NewManager(configDir, nil)
+		configureLegacyAuthManagerCompatibility(manager)
+		return manager
+	}
+)
+
+// Get resolves an access token for the active runtime profile.
+func (m *TokenManager) Get(ctx context.Context, configDir, explicitToken string) (AccessTokenSnapshot, error) {
+	if token := strings.TrimSpace(explicitToken); token != "" {
+		return AccessTokenSnapshot{AccessToken: token, Source: "explicit"}, nil
 	}
 	if strings.TrimSpace(configDir) == "" {
-		return "", fmt.Errorf("config directory is empty")
+		return AccessTokenSnapshot{}, fmt.Errorf("config directory is empty")
 	}
-	if filepath.Clean(configDir) == filepath.Clean(defaultConfigDir()) {
-		if tok := resolveRuntimeAuthToken(ctx, ""); tok != "" {
-			return tok, nil
+	key := tokenManagerKey{
+		configDir: canonicalTokenConfigDir(configDir),
+		profile:   strings.TrimSpace(authpkg.RuntimeProfile()),
+	}
+	entry := m.entry(key)
+	entry.mu.Lock()
+	defer entry.mu.Unlock()
+
+	now := time.Now()
+	if m != nil && m.now != nil {
+		now = m.now()
+	}
+	revision, present, err := authpkg.ReadTokenMarkerRevision(configDir)
+	if err != nil {
+		return AccessTokenSnapshot{}, err
+	}
+	if tokenSnapshotUsable(entry.snapshot, now) && present && revision != "" && revision == entry.revision {
+		return entry.snapshot, nil
+	}
+
+	// Treat the marker and credential as one optimistic snapshot. A concurrent
+	// login/refresh between the reads causes a retry instead of caching stale A
+	// under the publication marker for B.
+	for attempt := 0; attempt < 4; attempt++ {
+		beforeRevision, beforePresent, err := authpkg.ReadTokenMarkerRevision(configDir)
+		if err != nil {
+			return AccessTokenSnapshot{}, err
 		}
-		return "", noCredentialsError()
+		snapshot, err := resolveTokenSnapshotWithEdition(ctx, configDir, key.profile)
+		if err != nil {
+			return AccessTokenSnapshot{}, err
+		}
+		afterRevision, afterPresent, err := authpkg.ReadTokenMarkerRevision(configDir)
+		if err != nil {
+			return AccessTokenSnapshot{}, err
+		}
+		if beforePresent != afterPresent || beforeRevision != afterRevision {
+			continue
+		}
+		if strings.TrimSpace(snapshot.AccessToken) == "" {
+			return AccessTokenSnapshot{}, noCredentialsError()
+		}
+		if tokenSnapshotUsable(snapshot, now) && afterPresent && afterRevision != "" {
+			entry.snapshot = snapshot
+			entry.revision = afterRevision
+		} else {
+			entry.snapshot = AccessTokenSnapshot{}
+			entry.revision = ""
+		}
+		return snapshot, nil
 	}
-	tok, err := resolveAccessTokenFromDir(ctx, configDir)
+	return AccessTokenSnapshot{}, fmt.Errorf("token publication changed repeatedly while resolving credentials")
+}
+
+func (m *TokenManager) entry(key tokenManagerKey) *tokenManagerEntry {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	if m.entries == nil {
+		m.entries = make(map[tokenManagerKey]*tokenManagerEntry)
+	}
+	entry := m.entries[key]
+	if entry == nil {
+		entry = &tokenManagerEntry{}
+		m.entries[key] = entry
+	}
+	return entry
+}
+
+func (m *TokenManager) Invalidate() {
+	if m == nil {
+		return
+	}
+	m.mu.Lock()
+	m.entries = make(map[tokenManagerKey]*tokenManagerEntry)
+	m.mu.Unlock()
+}
+
+func resolveTokenSnapshotWithEdition(ctx context.Context, configDir, profile string) (AccessTokenSnapshot, error) {
+	hooks := edition.Get()
+	opaqueStorage := hooks.LoadToken != nil || hooks.SaveToken != nil || hooks.DeleteToken != nil
+	provider := hooks.TokenProvider
+	if provider == nil {
+		snapshot, err := resolveAccessTokenSnapshotFromDir(ctx, configDir, profile)
+		if err != nil {
+			return AccessTokenSnapshot{}, err
+		}
+		// Opaque edition storage hooks have no publication-revision contract.
+		// Resolve them on every logical request instead of caching a token that
+		// may be replaced outside the default auth store.
+		if opaqueStorage {
+			snapshot.ExpiresAt = time.Time{}
+		}
+		return snapshot, nil
+	}
+	var fallbackSnapshot AccessTokenSnapshot
+	var fallbackCalled bool
+	token, err := provider(ctx, func() (string, error) {
+		fallbackCalled = true
+		var fallbackErr error
+		fallbackSnapshot, fallbackErr = resolveAccessTokenSnapshotFromDir(ctx, configDir, profile)
+		if fallbackErr != nil {
+			return "", fallbackErr
+		}
+		return fallbackSnapshot.AccessToken, nil
+	})
+	if err != nil {
+		return AccessTokenSnapshot{}, fmt.Errorf("edition token provider: %w", err)
+	}
+	token = strings.TrimSpace(token)
+	if token == "" {
+		return AccessTokenSnapshot{}, noCredentialsError()
+	}
+	if fallbackCalled && token == fallbackSnapshot.AccessToken {
+		if opaqueStorage {
+			fallbackSnapshot.ExpiresAt = time.Time{}
+		}
+		return fallbackSnapshot, nil
+	}
+	// Edition providers expose no lifetime metadata, so resolve them on every
+	// logical request instead of recreating a process-lifetime string cache.
+	return AccessTokenSnapshot{AccessToken: token, Source: "edition"}, nil
+}
+
+func resolveAccessTokenSnapshotFromDir(ctx context.Context, configDir, profile string) (AccessTokenSnapshot, error) {
+	provider := newAccessTokenProvider(configDir)
+	if snapshotProvider, ok := provider.(accessTokenSnapshotGetter); ok {
+		data, err := snapshotProvider.GetTokenSnapshot(ctx)
+		if err == nil && data != nil && strings.TrimSpace(data.AccessToken) != "" {
+			return AccessTokenSnapshot{
+				AccessToken: strings.TrimSpace(data.AccessToken),
+				ExpiresAt:   data.ExpiresAt,
+				Source:      "oauth",
+			}, nil
+		}
+		if err != nil && !errors.Is(err, authpkg.ErrTokenDataNotFound) {
+			return AccessTokenSnapshot{}, err
+		}
+		if strings.TrimSpace(profile) != "" {
+			return AccessTokenSnapshot{}, authpkg.ErrTokenDataNotFound
+		}
+		return resolveLegacyToken(configDir, err)
+	}
+
+	token, err := provider.GetAccessToken(ctx)
+	if err == nil && strings.TrimSpace(token) != "" {
+		return AccessTokenSnapshot{AccessToken: strings.TrimSpace(token), Source: "oauth_compat"}, nil
+	}
+	if err != nil && !errors.Is(err, authpkg.ErrTokenDataNotFound) {
+		return AccessTokenSnapshot{}, err
+	}
+	if strings.TrimSpace(profile) != "" {
+		return AccessTokenSnapshot{}, authpkg.ErrTokenDataNotFound
+	}
+	return resolveLegacyToken(configDir, err)
+}
+
+func resolveLegacyToken(configDir string, oauthErr error) (AccessTokenSnapshot, error) {
+	token, source, err := newLegacyTokenManager(configDir).GetToken()
+	if err == nil && strings.TrimSpace(token) != "" {
+		return AccessTokenSnapshot{AccessToken: strings.TrimSpace(token), Source: source}, nil
+	}
+	if err != nil && !errors.Is(err, authpkg.ErrTokenDataNotFound) {
+		return AccessTokenSnapshot{}, err
+	}
+	if oauthErr != nil {
+		return AccessTokenSnapshot{}, oauthErr
+	}
+	return AccessTokenSnapshot{}, authpkg.ErrTokenDataNotFound
+}
+
+func resolveAccessTokenFromDir(ctx context.Context, configDir string) (string, error) {
+	snapshot, err := resolveAccessTokenSnapshotFromDir(ctx, configDir, authpkg.RuntimeProfile())
 	if err != nil {
 		return "", err
 	}
-	if tok != "" {
-		return tok, nil
+	return snapshot.AccessToken, nil
+}
+
+// ResolveAuxiliaryAccessToken resolves every non-runner bearer token through
+// the same TokenManager used by MCP tool calls.
+func ResolveAuxiliaryAccessToken(ctx context.Context, configDir, explicitToken string) (string, error) {
+	snapshot, err := runtimeTokenManager.Get(ctx, configDir, explicitToken)
+	if err != nil {
+		return "", err
 	}
-	return "", noCredentialsError()
+	return snapshot.AccessToken, nil
+}
+
+func tokenSnapshotUsable(snapshot AccessTokenSnapshot, now time.Time) bool {
+	return strings.TrimSpace(snapshot.AccessToken) != "" &&
+		!snapshot.ExpiresAt.IsZero() &&
+		now.Before(snapshot.ExpiresAt.Add(-accessTokenRefreshWindow))
+}
+
+func canonicalTokenConfigDir(configDir string) string {
+	if absolute, err := filepath.Abs(configDir); err == nil {
+		return filepath.Clean(absolute)
+	}
+	return filepath.Clean(configDir)
 }
 
 func noCredentialsError() error {
 	if edition.Get().IsEmbedded {
-		return fmt.Errorf("认证信息已失效，请重新认证")
+		return fmt.Errorf("认证信息已失效，请重新认证: %w", authpkg.ErrTokenDataNotFound)
 	}
-	return fmt.Errorf("no credentials found, run: dws auth login")
+	return fmt.Errorf("no credentials found, run: dws auth login: %w", authpkg.ErrTokenDataNotFound)
 }

internal/app/access_token_resolve_test.go

@@ -5,7 +5,15 @@ package app
 
 import (
 	"context"
+	"errors"
+	"net/http"
+	"path/filepath"
+	"strings"
 	"testing"
+	"time"
+
+	authpkg "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/auth"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/keychain"
 )
 
 func TestResolveAuxiliaryAccessToken_explicitToken(t *testing.T) {
@@ -24,3 +32,48 @@ func TestResolveAuxiliaryAccessToken_emptyConfigDir(t *testing.T) {
 		t.Fatal("expected error for empty config directory")
 	}
 }
+
+func TestResolveAccessTokenFromDirPreservesRefreshFailure(t *testing.T) {
+	root := t.TempDir()
+	configDir := filepath.Join(root, "config")
+	t.Setenv(keychain.DisableKeychainEnv, "1")
+	t.Setenv(keychain.StorageDirEnv, filepath.Join(root, "keychain"))
+
+	if err := authpkg.SaveTokenData(configDir, &authpkg.TokenData{
+		AccessToken:  "expired-access",
+		RefreshToken: "refresh-token",
+		ExpiresAt:    time.Now().Add(-time.Hour),
+		RefreshExpAt: time.Now().Add(24 * time.Hour),
+		CorpID:       "corp_refresh",
+		UserID:       "user_refresh",
+		ClientID:     "client_refresh",
+		Source:       "mcp",
+	}); err != nil {
+		t.Fatalf("SaveTokenData() error = %v", err)
+	}
+
+	originalTransport := http.DefaultTransport
+	t.Cleanup(func() {
+		http.DefaultTransport = originalTransport
+	})
+	http.DefaultTransport = refreshFailureRoundTripFunc(func(*http.Request) (*http.Response, error) {
+		return nil, errors.New("refresh endpoint rejected token")
+	})
+
+	token, err := resolveAccessTokenFromDir(context.Background(), configDir)
+	if token != "" {
+		t.Fatalf("token = %q, want empty", token)
+	}
+	if err == nil {
+		t.Fatal("resolveAccessTokenFromDir() error = nil")
+	}
+	if !strings.Contains(err.Error(), "refresh endpoint rejected token") {
+		t.Fatalf("error = %q, want original refresh failure", err)
+	}
+}
+
+type refreshFailureRoundTripFunc func(*http.Request) (*http.Response, error)
+
+func (f refreshFailureRoundTripFunc) RoundTrip(req *http.Request) (*http.Response, error) {
+	return f(req)
+}

internal/app/agent_host.go

@@ -0,0 +1,70 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package app
+
+import (
+	"regexp"
+	"strings"
+
+	apperrors "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/errors"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/configmeta"
+)
+
+const (
+	envDWSAgentHost    = "DWS_AGENT_HOST"
+	headerDWSAgentHost = "x-dws-agent-host"
+	maxAgentHostBytes  = 64
+)
+
+var agentHostPattern = regexp.MustCompile(`^[a-z0-9][a-z0-9_-]*$`)
+
+func init() {
+	configmeta.Register(configmeta.ConfigItem{
+		Name:        envDWSAgentHost,
+		Category:    configmeta.CategoryExternal,
+		Description: "调用 DWS 的 Agent 运行形态标识；作为 x-dws-agent-host 发送供下游观测，本客户端不使用该值改变 PAT、鉴权或路由",
+		Example:     "cloud",
+	})
+}
+
+// parseAgentHost normalizes and validates the caller-declared runtime-form
+// signal. Only surrounding ASCII spaces and tabs are trimmed; other control
+// or Unicode whitespace remains visible to validation and is rejected. An
+// unset or ASCII-whitespace-only value means "do not emit".
+func parseAgentHost(raw string) (string, error) {
+	if strings.ContainsAny(raw, "\r\n") {
+		return "", invalidAgentHostError()
+	}
+
+	value := strings.Trim(raw, " \t")
+	if value == "" {
+		return "", nil
+	}
+	if len(value) > maxAgentHostBytes {
+		return "", invalidAgentHostError()
+	}
+	if !agentHostPattern.MatchString(value) {
+		return "", invalidAgentHostError()
+	}
+	return value, nil
+}
+
+func invalidAgentHostError() error {
+	// Do not include the raw environment value in the error: it is an
+	// untrusted caller-controlled string and may contain sensitive data.
+	return apperrors.NewValidation(
+		"DWS_AGENT_HOST must be at most 64 bytes and match ^[a-z0-9][a-z0-9_-]*$",
+		apperrors.WithReason("invalid_agent_host"),
+	)
+}

internal/app/agent_host_test.go

@@ -0,0 +1,206 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package app
+
+import (
+	"errors"
+	"io"
+	"strings"
+	"testing"
+
+	authpkg "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/auth"
+	apperrors "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/errors"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/agentproduct"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
+	"github.com/spf13/cobra"
+)
+
+func TestParseAgentHost(t *testing.T) {
+	valid := []struct {
+		name string
+		raw  string
+		want string
+	}{
+		{name: "unset", raw: "", want: ""},
+		{name: "ASCII whitespace only", raw: " \t ", want: ""},
+		{name: "cloud", raw: "cloud", want: "cloud"},
+		{name: "desktop", raw: "desktop", want: "desktop"},
+		{name: "legacy combined label remains valid", raw: "qwenwork_cloud", want: "qwenwork_cloud"},
+		{name: "trim", raw: " \tcloud\t ", want: "cloud"},
+		{name: "generic", raw: "host-2_alpha", want: "host-2_alpha"},
+		{name: "leading digit", raw: "2nd_host", want: "2nd_host"},
+		{name: "maximum length", raw: strings.Repeat("a", maxAgentHostBytes), want: strings.Repeat("a", maxAgentHostBytes)},
+	}
+	for _, tc := range valid {
+		t.Run(tc.name, func(t *testing.T) {
+			got, err := parseAgentHost(tc.raw)
+			if err != nil {
+				t.Fatalf("parseAgentHost() error = %v", err)
+			}
+			if got != tc.want {
+				t.Fatalf("parseAgentHost() = %q, want %q", got, tc.want)
+			}
+		})
+	}
+
+	invalid := []struct {
+		name string
+		raw  string
+	}{
+		{name: "carriage return", raw: "qwenwork_cloud\r"},
+		{name: "line feed", raw: "\nqwenwork_cloud"},
+		{name: "uppercase", raw: "Qwenwork_cloud"},
+		{name: "internal space", raw: "qwenwork cloud"},
+		{name: "internal tab", raw: "qwenwork\tcloud"},
+		{name: "unicode", raw: "千问办公"},
+		{name: "leading dash", raw: "-qwenwork"},
+		{name: "leading underscore", raw: "_qwenwork"},
+		{name: "control character", raw: "qwenwork\x00cloud"},
+		{name: "vertical tab", raw: "\vcloud"},
+		{name: "form feed", raw: "cloud\f"},
+		{name: "next line", raw: "cloud\u0085"},
+		{name: "non-breaking space", raw: "\u00a0cloud"},
+		{name: "ideographic space", raw: "cloud\u3000"},
+		{name: "too long", raw: strings.Repeat("a", maxAgentHostBytes+1)},
+	}
+	for _, tc := range invalid {
+		t.Run(tc.name, func(t *testing.T) {
+			got, err := parseAgentHost(tc.raw)
+			if err == nil {
+				t.Fatalf("parseAgentHost(%q) = %q, want error", tc.raw, got)
+			}
+			var appErr *apperrors.Error
+			if !errors.As(err, &appErr) {
+				t.Fatalf("parseAgentHost() error type = %T, want *errors.Error", err)
+			}
+			if appErr.Category != apperrors.CategoryValidation {
+				t.Fatalf("category = %q, want validation", appErr.Category)
+			}
+			if appErr.Reason != "invalid_agent_host" {
+				t.Fatalf("reason = %q, want invalid_agent_host", appErr.Reason)
+			}
+			if tc.raw != "" && strings.Contains(err.Error(), tc.raw) {
+				t.Fatalf("error must not echo invalid value %q: %v", tc.raw, err)
+			}
+		})
+	}
+}
+
+func TestResolveIdentityHeadersAddsAgentHostBeforeEditionMerge(t *testing.T) {
+	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
+	t.Setenv(envDWSAgentHost, " qwenwork_desktop ")
+	t.Setenv(agentproduct.EnvName, "")
+	t.Setenv(envDWSChannel, "channel-test")
+	t.Setenv(envDingtalkAgent, "agent-test")
+	t.Setenv(authpkg.AgentCodeEnv, "agent-code-test")
+
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition) })
+
+	mergeSawAgentHost := ""
+	edition.Override(&edition.Hooks{
+		MergeHeaders: func(headers map[string]string) map[string]string {
+			mergeSawAgentHost = headers[headerDWSAgentHost]
+			headers["claw-type"] = "test-claw"
+			return headers
+		},
+	})
+
+	headers := resolveIdentityHeaders()
+	if got := mergeSawAgentHost; got != "qwenwork_desktop" {
+		t.Fatalf("MergeHeaders saw agent host %q, want qwenwork_desktop", got)
+	}
+	if got := headers[headerDWSAgentHost]; got != "qwenwork_desktop" {
+		t.Fatalf("%s = %q, want qwenwork_desktop", headerDWSAgentHost, got)
+	}
+	if got := headers["x-dingtalk-source"]; got != "github" {
+		t.Fatalf("x-dingtalk-source = %q, want github", got)
+	}
+	if got := headers["x-dingtalk-dws-agent-code"]; got != "agent-code-test" {
+		t.Fatalf("agentCode header = %q, want agent-code-test", got)
+	}
+	if got := headers["x-dws-channel"]; got != "channel-test" {
+		t.Fatalf("channel header = %q, want channel-test", got)
+	}
+	if got := headers["claw-type"]; got != "test-claw" {
+		t.Fatalf("claw-type = %q, want test-claw", got)
+	}
+}
+
+func TestResolveIdentityHeadersOmitsAbsentOrInvalidAgentHost(t *testing.T) {
+	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
+	t.Setenv(envDWSChannel, "channel-test")
+
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition) })
+	edition.Override(&edition.Hooks{
+		MergeHeaders: func(headers map[string]string) map[string]string {
+			return headers
+		},
+	})
+
+	for _, raw := range []string{"", " \t ", "DO_NOT_ECHO"} {
+		t.Setenv(envDWSAgentHost, raw)
+		headers := resolveIdentityHeaders()
+		if _, ok := headers[headerDWSAgentHost]; ok {
+			t.Fatalf("%s must be omitted for %q: %#v", headerDWSAgentHost, raw, headers)
+		}
+		if got := headers["x-dingtalk-source"]; got != "github" {
+			t.Fatalf("x-dingtalk-source = %q, want github", got)
+		}
+		if got := headers["x-dws-channel"]; got != "channel-test" {
+			t.Fatalf("channel header = %q, want channel-test", got)
+		}
+	}
+}
+
+func TestRootRejectsInvalidAgentHostBeforeEditionHook(t *testing.T) {
+	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
+	const invalidValue = "DO_NOT_ECHO"
+	t.Setenv(envDWSAgentHost, invalidValue)
+
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition) })
+
+	hookCalled := false
+	edition.Override(&edition.Hooks{
+		AfterPersistentPreRun: func(_ *cobra.Command, _ []string) error {
+			hookCalled = true
+			return nil
+		},
+	})
+
+	root := NewRootCommand()
+	root.SetOut(io.Discard)
+	root.SetErr(io.Discard)
+	root.SetArgs([]string{"version"})
+	err := root.Execute()
+	if err == nil {
+		t.Fatal("root command accepted invalid DWS_AGENT_HOST")
+	}
+	if hookCalled {
+		t.Fatal("edition AfterPersistentPreRun ran before DWS_AGENT_HOST validation")
+	}
+
+	var appErr *apperrors.Error
+	if !errors.As(err, &appErr) {
+		t.Fatalf("root error type = %T, want *errors.Error", err)
+	}
+	if appErr.Category != apperrors.CategoryValidation || appErr.Reason != "invalid_agent_host" {
+		t.Fatalf("root error = category %q reason %q", appErr.Category, appErr.Reason)
+	}
+	if strings.Contains(err.Error(), invalidValue) {
+		t.Fatalf("root error must not echo invalid value: %v", err)
+	}
+}

internal/app/agent_product.go

@@ -0,0 +1,75 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package app
+
+import (
+	apperrors "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/errors"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/agentproduct"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/configmeta"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
+)
+
+func init() {
+	configmeta.Register(configmeta.ConfigItem{
+		Name:         agentproduct.EnvName,
+		Category:     configmeta.CategoryExternal,
+		Description:  "调用方声明的 Agent 产品标识；作为 x-dws-agent-product 发送并用于 IM 小尾巴，本客户端不使用该值改变 HTTP claw-type/PAT",
+		DefaultValue: "未设置（请求头省略，IM 使用当前发行版默认值）",
+		Example:      "qwenwork",
+	})
+}
+
+// parseAgentProduct converts the reusable package error into the CLI's stable
+// structured validation error without exposing the untrusted raw value.
+func parseAgentProduct(raw string) (string, error) {
+	value, err := agentproduct.Parse(raw)
+	if err != nil {
+		return "", invalidAgentProductError()
+	}
+	return value, nil
+}
+
+func invalidAgentProductError() error {
+	return apperrors.NewValidation(
+		"DWS_AGENT_PRODUCT must be at most 64 bytes and match ^[A-Za-z0-9][A-Za-z0-9_-]*$",
+		apperrors.WithReason("invalid_agent_product"),
+	)
+}
+
+// resolveEditionClawType resolves the fixed routing/PAT identity supplied by
+// the active edition. DWS_AGENT_PRODUCT is deliberately not consulted.
+func resolveEditionClawType(headers map[string]string) string {
+	if value := headers["claw-type"]; value != "" {
+		return value
+	}
+	return edition.DefaultOSSClawType
+}
+
+// applyAgentProductHeader injects only a valid, non-empty caller-declared
+// product. Invalid values are omitted on library paths that bypass root
+// validation; normal CLI execution rejects them before network access.
+func applyAgentProductHeader(headers map[string]string) map[string]string {
+	value, err := agentproduct.ResolveFromEnv("")
+	if err != nil || value == "" {
+		if headers != nil {
+			delete(headers, agentproduct.HeaderName)
+		}
+		return headers
+	}
+	if headers == nil {
+		headers = make(map[string]string)
+	}
+	headers[agentproduct.HeaderName] = value
+	return headers
+}

internal/app/agent_product_test.go

@@ -0,0 +1,333 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package app
+
+import (
+	"errors"
+	"io"
+	"strings"
+	"testing"
+
+	authpkg "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/auth"
+	apperrors "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/errors"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/agentproduct"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
+	"github.com/spf13/cobra"
+)
+
+func TestUnsetAgentProductOmitsHeaderAndKeepsOpenSourceClawType(t *testing.T) {
+	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
+	t.Setenv(agentproduct.EnvName, "")
+
+	headers := resolveIdentityHeaders()
+	if got := headers["claw-type"]; got != edition.DefaultOSSClawType {
+		t.Fatalf("claw-type = %q, want %q", got, edition.DefaultOSSClawType)
+	}
+	if _, ok := headers[agentproduct.HeaderName]; ok {
+		t.Fatalf("unset Product must omit %s", agentproduct.HeaderName)
+	}
+}
+
+func TestParseAgentProductReturnsStableValidationError(t *testing.T) {
+	const invalidValue = "DO_NOT ECHO"
+
+	got, err := parseAgentProduct(invalidValue)
+	if got != "" {
+		t.Fatalf("parseAgentProduct() = %q, want empty", got)
+	}
+
+	var appErr *apperrors.Error
+	if !errors.As(err, &appErr) {
+		t.Fatalf("parseAgentProduct() error type = %T, want *errors.Error", err)
+	}
+	if appErr.Category != apperrors.CategoryValidation {
+		t.Fatalf("category = %q, want validation", appErr.Category)
+	}
+	if appErr.Reason != "invalid_agent_product" {
+		t.Fatalf("reason = %q, want invalid_agent_product", appErr.Reason)
+	}
+	if strings.Contains(err.Error(), invalidValue) {
+		t.Fatalf("error must not echo invalid value: %v", err)
+	}
+}
+
+func TestResolveIdentityHeadersSeparatesAgentProductFromClawType(t *testing.T) {
+	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
+
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition) })
+	edition.Override(&edition.Hooks{
+		MergeHeaders: func(headers map[string]string) map[string]string {
+			headers["claw-type"] = "wukong"
+			headers[agentproduct.HeaderName] = "merge-product-must-not-win"
+			headers["x-edition-header"] = "preserved"
+			return headers
+		},
+		EnterpriseCredentialHeaders: func(headers map[string]string) map[string]string {
+			headers["claw-type"] = "credential-must-not-win"
+			headers[agentproduct.HeaderName] = "credential-product-must-not-win"
+			headers["x-enterprise-header"] = "preserved"
+			return headers
+		},
+	})
+
+	t.Run("unset omits Product and keeps edition claw-type", func(t *testing.T) {
+		t.Setenv(agentproduct.EnvName, "")
+		headers := resolveIdentityHeaders()
+		if got := headers["claw-type"]; got != "wukong" {
+			t.Fatalf("claw-type = %q, want wukong", got)
+		}
+		if _, ok := headers[agentproduct.HeaderName]; ok {
+			t.Fatalf("unset Product must omit %s", agentproduct.HeaderName)
+		}
+	})
+
+	t.Run("valid Product is final without changing claw-type", func(t *testing.T) {
+		t.Setenv(agentproduct.EnvName, " qwenwork ")
+		headers := resolveIdentityHeaders()
+		if got := headers[agentproduct.HeaderName]; got != "qwenwork" {
+			t.Fatalf("%s = %q, want qwenwork", agentproduct.HeaderName, got)
+		}
+		if got := headers["claw-type"]; got != "wukong" {
+			t.Fatalf("claw-type = %q, want wukong", got)
+		}
+		if got := headers["x-edition-header"]; got != "preserved" {
+			t.Fatalf("edition header = %q, want preserved", got)
+		}
+		if got := headers["x-enterprise-header"]; got != "preserved" {
+			t.Fatalf("enterprise header = %q, want preserved", got)
+		}
+		if got := headers["x-dingtalk-source"]; got != "github" {
+			t.Fatalf("x-dingtalk-source = %q, want github", got)
+		}
+	})
+
+	t.Run("invalid library input omits Product and keeps edition claw-type", func(t *testing.T) {
+		t.Setenv(agentproduct.EnvName, "qwen work")
+		headers := resolveIdentityHeaders()
+		if got := headers["claw-type"]; got != "wukong" {
+			t.Fatalf("claw-type = %q, want wukong", got)
+		}
+		if _, ok := headers[agentproduct.HeaderName]; ok {
+			t.Fatalf("invalid Product must omit %s", agentproduct.HeaderName)
+		}
+	})
+}
+
+func TestApplyAgentProductHeader(t *testing.T) {
+	t.Run("valid value allocates headers", func(t *testing.T) {
+		t.Setenv(agentproduct.EnvName, "qwenwork")
+
+		headers := applyAgentProductHeader(nil)
+		if got := headers[agentproduct.HeaderName]; got != "qwenwork" {
+			t.Fatalf("%s = %q, want qwenwork", agentproduct.HeaderName, got)
+		}
+	})
+
+	for _, tc := range []struct {
+		name  string
+		value string
+	}{
+		{name: "empty value", value: ""},
+		{name: "invalid value", value: "qwen work"},
+	} {
+		t.Run(tc.name+" removes inherited header", func(t *testing.T) {
+			t.Setenv(agentproduct.EnvName, tc.value)
+			headers := applyAgentProductHeader(map[string]string{
+				agentproduct.HeaderName: "must-not-leak",
+				"x-preserved":           "yes",
+			})
+			if _, ok := headers[agentproduct.HeaderName]; ok {
+				t.Fatalf("%s must be omitted", agentproduct.HeaderName)
+			}
+			if got := headers["x-preserved"]; got != "yes" {
+				t.Fatalf("x-preserved = %q, want yes", got)
+			}
+		})
+	}
+}
+
+func TestRootRejectsInvalidAgentProductBeforeEditionHook(t *testing.T) {
+	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
+	const invalidValue = "DO_NOT ECHO"
+	t.Setenv(agentproduct.EnvName, invalidValue)
+
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition) })
+
+	hookCalled := false
+	edition.Override(&edition.Hooks{
+		AfterPersistentPreRun: func(_ *cobra.Command, _ []string) error {
+			hookCalled = true
+			return nil
+		},
+	})
+
+	root := NewRootCommand()
+	root.SetOut(io.Discard)
+	root.SetErr(io.Discard)
+	root.SetArgs([]string{"version"})
+	err := root.Execute()
+	if err == nil {
+		t.Fatal("root command accepted invalid DWS_AGENT_PRODUCT")
+	}
+	if hookCalled {
+		t.Fatal("edition AfterPersistentPreRun ran before DWS_AGENT_PRODUCT validation")
+	}
+
+	var appErr *apperrors.Error
+	if !errors.As(err, &appErr) {
+		t.Fatalf("root error type = %T, want *errors.Error", err)
+	}
+	if appErr.Category != apperrors.CategoryValidation || appErr.Reason != "invalid_agent_product" {
+		t.Fatalf("root error = category %q reason %q", appErr.Category, appErr.Reason)
+	}
+	if strings.Contains(err.Error(), invalidValue) {
+		t.Fatalf("root error must not echo invalid value: %v", err)
+	}
+}
+
+func TestEffectiveClawTypeDoesNotInvokeEnterpriseCredentialHeaders(t *testing.T) {
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition) })
+
+	hookCalled := false
+	edition.Override(&edition.Hooks{
+		MergeHeaders: func(headers map[string]string) map[string]string {
+			headers["claw-type"] = "wukong"
+			return headers
+		},
+		EnterpriseCredentialHeaders: func(headers map[string]string) map[string]string {
+			hookCalled = true
+			headers["claw-type"] = "enterprise-default"
+			return headers
+		},
+	})
+
+	t.Setenv(agentproduct.EnvName, "qwenwork")
+	if got := effectiveClawType(); got != "wukong" {
+		t.Fatalf("effectiveClawType() = %q, want wukong", got)
+	}
+	if hookCalled {
+		t.Fatal("EnterpriseCredentialHeaders hook ran during PAT error serialization")
+	}
+}
+
+func TestAgentProductControlsObservabilityHeaderAndMessageClawTypeOnly(t *testing.T) {
+	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
+	t.Setenv(agentproduct.EnvName, "qwenwork")
+
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition) })
+	edition.Override(&edition.Hooks{
+		ClawTypeValue: "message-brand",
+		MergeHeaders: func(headers map[string]string) map[string]string {
+			headers["claw-type"] = "wukong"
+			return headers
+		},
+	})
+
+	headers := resolveIdentityHeaders()
+	if got := headers[agentproduct.HeaderName]; got != "qwenwork" {
+		t.Fatalf("HTTP %s = %q, want qwenwork", agentproduct.HeaderName, got)
+	}
+	if got := headers["claw-type"]; got != "wukong" {
+		t.Fatalf("HTTP claw-type = %q, want wukong", got)
+	}
+	if got := edition.ClawType(); got != "qwenwork" {
+		t.Fatalf("message clawType = %q, want qwenwork", got)
+	}
+}
+
+func TestResolveIdentityHeadersRestoresIdentityAfterNilCredentialHeaders(t *testing.T) {
+	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
+	t.Setenv(agentproduct.EnvName, "qwenwork")
+
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition) })
+
+	credentialHookCalled := false
+	edition.Override(&edition.Hooks{
+		MergeHeaders: func(headers map[string]string) map[string]string {
+			headers["claw-type"] = "wukong"
+			return headers
+		},
+		EnterpriseCredentialHeaders: func(map[string]string) map[string]string {
+			credentialHookCalled = true
+			return nil
+		},
+	})
+
+	headers := resolveIdentityHeaders()
+	if !credentialHookCalled {
+		t.Fatal("EnterpriseCredentialHeaders hook was not called")
+	}
+	if got := headers[agentproduct.HeaderName]; got != "qwenwork" {
+		t.Fatalf("%s = %q, want qwenwork", agentproduct.HeaderName, got)
+	}
+	if got := headers["claw-type"]; got != "wukong" {
+		t.Fatalf("claw-type = %q, want wukong", got)
+	}
+}
+
+func TestResolveIdentityHeadersRestoresDefaultsAfterNilMergeHeaders(t *testing.T) {
+	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
+	t.Setenv(agentproduct.EnvName, "")
+
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition) })
+	edition.Override(&edition.Hooks{
+		MergeHeaders: func(map[string]string) map[string]string {
+			return nil
+		},
+	})
+
+	headers := resolveIdentityHeaders()
+	if got := headers["claw-type"]; got != edition.DefaultOSSClawType {
+		t.Fatalf("claw-type = %q, want %q", got, edition.DefaultOSSClawType)
+	}
+	if _, ok := headers[agentproduct.HeaderName]; ok {
+		t.Fatalf("unset Product must omit %s", agentproduct.HeaderName)
+	}
+}
+
+func TestEffectiveClawTypeIgnoresAgentProduct(t *testing.T) {
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition) })
+	edition.Override(&edition.Hooks{
+		MergeHeaders: func(headers map[string]string) map[string]string {
+			headers["claw-type"] = "wukong"
+			return headers
+		},
+	})
+
+	t.Setenv(agentproduct.EnvName, "qwenwork")
+	if got := effectiveClawType(); got != "wukong" {
+		t.Fatalf("effectiveClawType() = %q, want wukong", got)
+	}
+	t.Setenv(authpkg.AgentCodeEnv, "agent-code")
+	if got := apperrors.HostControlBlock()["clawType"]; got != "wukong" {
+		t.Fatalf("hostControl.clawType = %q, want wukong", got)
+	}
+
+	t.Setenv(agentproduct.EnvName, "")
+	if got := effectiveClawType(); got != "wukong" {
+		t.Fatalf("effectiveClawType() = %q, want wukong", got)
+	}
+
+	t.Setenv(agentproduct.EnvName, "invalid product")
+	if got := effectiveClawType(); got != "wukong" {
+		t.Fatalf("effectiveClawType() with invalid env = %q, want wukong", got)
+	}
+}

internal/app/api_command.go

@@ -38,6 +38,19 @@ type apiFlags struct {
 	baseURL   string
 }
 
+type appTokenGetter interface {
+	GetToken(context.Context) (string, error)
+}
+
+var newAppTokenProvider = func(configDir, appKey, appSecret string) appTokenGetter {
+	return &authpkg.AppTokenProvider{ConfigDir: configDir, AppKey: appKey, AppSecret: appSecret}
+}
+
+var (
+	apiClientID     = authpkg.ClientID
+	apiClientSecret = authpkg.ClientSecret
+)
+
 // newAPICommand creates the `dws api` subcommand for raw DingTalk OpenAPI calls.
 func newAPICommand(flags *GlobalFlags) *cobra.Command {
 	af := &apiFlags{}
@@ -271,10 +284,7 @@ func parseQueryStringToJSON(rawQuery string) string {
 		return "{}"
 	}
 
-	data, err := json.Marshal(paramsMap)
-	if err != nil {
-		return "{}"
-	}
+	data, _ := json.Marshal(paramsMap)
 	return string(data)
 }
 
@@ -289,8 +299,8 @@ func resolveRawAPIToken(ctx context.Context, explicitToken string) (string, erro
 	}
 
 	// Resolve app credentials (clientID/clientSecret).
-	appKey := authpkg.ClientID()
-	appSecret := authpkg.ClientSecret()
+	appKey := apiClientID()
+	appSecret := apiClientSecret()
 
 	if appKey == "" || appSecret == "" || strings.HasPrefix(appKey, "<") || strings.HasPrefix(appSecret, "<") {
 		return "", apperrors.NewAuth(
@@ -308,11 +318,7 @@ func resolveRawAPIToken(ctx context.Context, explicitToken string) (string, erro
 
 	// Use AppTokenProvider for automatic caching and refresh.
 	configDir := defaultConfigDir()
-	provider := &authpkg.AppTokenProvider{
-		ConfigDir: configDir,
-		AppKey:    appKey,
-		AppSecret: appSecret,
-	}
+	provider := newAppTokenProvider(configDir, appKey, appSecret)
 	token, err := provider.GetToken(ctx)
 	if err != nil {
 		return "", apperrors.NewAuth(fmt.Sprintf("获取应用级访问令牌失败: %v", err))

internal/app/app_small_coverage_test.go

@@ -0,0 +1,404 @@
+package app
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	authpkg "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/auth"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/executor"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/plugin"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/transport"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/agentproduct"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
+	"github.com/spf13/cobra"
+)
+
+type appFailWriter struct{ err error }
+
+func (w appFailWriter) Write([]byte) (int, error) { return 0, w.err }
+
+type fakeAccessTokenGetter struct {
+	token string
+	err   error
+}
+
+func (g fakeAccessTokenGetter) GetAccessToken(context.Context) (string, error) {
+	return g.token, g.err
+}
+
+func (g fakeAccessTokenGetter) ForceRefreshRejectedToken(context.Context, string) (string, error) {
+	return g.token, g.err
+}
+
+type fakeLegacyTokenGetter struct {
+	token string
+	err   error
+}
+
+func (g fakeLegacyTokenGetter) GetToken() (string, string, error) {
+	return g.token, "test", g.err
+}
+
+type fakeAppTokenGetter struct {
+	token string
+	err   error
+}
+
+func (g fakeAppTokenGetter) GetToken(context.Context) (string, error) { return g.token, g.err }
+
+type fakeSkillDirEntry struct{ dir bool }
+
+func (e fakeSkillDirEntry) Name() string               { return "entry" }
+func (e fakeSkillDirEntry) IsDir() bool                { return e.dir }
+func (e fakeSkillDirEntry) Type() os.FileMode          { return 0 }
+func (e fakeSkillDirEntry) Info() (os.FileInfo, error) { return nil, nil }
+
+func docPreflightServer(t *testing.T, result map[string]any) *httptest.Server {
+	t.Helper()
+	return httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		var request struct {
+			ID int `json:"id"`
+		}
+		_ = json.NewDecoder(r.Body).Decode(&request)
+		w.Header().Set("Content-Type", "application/json")
+		_ = json.NewEncoder(w).Encode(map[string]any{
+			"jsonrpc": "2.0",
+			"id":      request.ID,
+			"result":  result,
+		})
+	}))
+}
+
+func TestCrossPlatformCoverageDocDownloadPreflightCoverage(t *testing.T) {
+	runner := &runtimeRunner{}
+	base := executor.Invocation{CanonicalProduct: "doc", Tool: "download_file", Params: map[string]any{"nodeId": " node "}}
+	if err := runner.preflightDocDownload(context.Background(), transport.NewClient(nil), "", executor.Invocation{}); err != nil {
+		t.Fatal(err)
+	}
+	if err := runner.preflightDocDownload(context.Background(), transport.NewClient(nil), "", executor.Invocation{CanonicalProduct: "DOC", Tool: "download_file"}); err != nil {
+		t.Fatal(err)
+	}
+	if !isDocDownloadInvocation(base) || docDownloadNodeID(map[string]any{"node": " n "}) != "n" || docDownloadNodeID(map[string]any{"dentryUuid": " d "}) != "d" || docDownloadNodeID(map[string]any{"nodeId": 1}) != "" {
+		t.Fatal("doc download invocation helpers returned unexpected values")
+	}
+	if documentInfoExtension(map[string]any{"data": map[string]any{"extension": " doc "}}) != "doc" ||
+		documentInfoExtension(map[string]any{"extension": " pdf "}) != "pdf" ||
+		stringAtPath(map[string]any{"x": "value"}, "x", "nested") != "" ||
+		stringAtPath(map[string]any{"x": 1}, "x") != "" {
+		t.Fatal("document extension helpers returned unexpected values")
+	}
+	if unsupportedAXLSDownloadError() == nil {
+		t.Fatal("missing AXLS validation error")
+	}
+
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition) })
+	hookErr := errors.New("classified")
+	for _, tc := range []struct {
+		name   string
+		result map[string]any
+		hooks  *edition.Hooks
+		want   string
+	}{
+		{name: "ok", result: map[string]any{"content": map[string]any{"result": map[string]any{"extension": "docx"}}}, hooks: &edition.Hooks{}},
+		{name: "edition classifier", result: map[string]any{"content": map[string]any{}}, hooks: &edition.Hooks{ClassifyToolResult: func(map[string]any) error { return hookErr }}, want: "classified"},
+		{name: "pat", result: map[string]any{"content": map[string]any{"errorCode": "PAT_NO_PERMISSION"}}, hooks: &edition.Hooks{}, want: "PAT_NO_PERMISSION"},
+		{name: "mcp error", result: map[string]any{"isError": true, "content": []map[string]any{{"type": "text", "text": "mcp failed"}}}, hooks: &edition.Hooks{}, want: "mcp failed"},
+		{name: "business error", result: map[string]any{"content": map[string]any{"success": false, "errorMsg": "business failed"}}, hooks: &edition.Hooks{}, want: "business failed"},
+		{name: "axls", result: map[string]any{"content": map[string]any{"data": map[string]any{"extension": "AXLS"}}}, hooks: &edition.Hooks{}, want: "extension=axls"},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			edition.Override(tc.hooks)
+			server := docPreflightServer(t, tc.result)
+			defer server.Close()
+			client := transport.NewClient(nil)
+			client.TrustedDomains = []string{"127.0.0.1"}
+			err := runner.preflightDocDownload(context.Background(), client, server.URL, base)
+			if tc.want == "" && err != nil {
+				t.Fatalf("preflight error = %v", err)
+			}
+			if tc.want != "" && (err == nil || !strings.Contains(err.Error(), tc.want)) {
+				t.Fatalf("preflight error = %v, want %q", err, tc.want)
+			}
+		})
+	}
+	server := docPreflightServer(t, map[string]any{})
+	endpoint := server.URL
+	server.Close()
+	client := transport.NewClient(nil)
+	client.TrustedDomains = []string{"127.0.0.1"}
+	client.MaxRetries = 0
+	if err := runner.preflightDocDownload(context.Background(), client, endpoint, base); err == nil {
+		t.Fatal("network preflight failure succeeded")
+	}
+}
+
+func TestCrossPlatformCoverageRootHelpRemainingCoverage(t *testing.T) {
+	configureRootHelp(nil)
+	renderRootGlobalFlags(nil)
+	if visiblePersistentFlags(nil) != nil || formatRootFlag(nil) != "" || commandShort(nil) != "" || visibleMCPRootCommands(nil) != nil || visibleUtilityRootCommands(nil) != nil {
+		t.Fatal("nil root helper contract changed")
+	}
+
+	oldEdition := edition.Get()
+	edition.Override(&edition.Hooks{VisibleProducts: func() []string { return []string{"service"} }})
+	t.Cleanup(func() { edition.Override(oldEdition); SetDynamicServers(nil) })
+	root := &cobra.Command{Use: "root", Long: "long help"}
+	root.SetOut(io.Discard)
+	root.PersistentFlags().StringP("value", "x", "", "value")
+	root.PersistentFlags().Bool("hidden", false, "hidden")
+	_ = root.PersistentFlags().MarkHidden("hidden")
+	root.AddCommand(&cobra.Command{Use: "service", Short: "service"}, &cobra.Command{Use: "utility", Short: "utility"})
+	configureRootHelp(root)
+	if err := root.Commands()[0].Help(); err != nil {
+		t.Fatal(err)
+	}
+	root.SetArgs([]string{"help", "missing"})
+	if err := root.Execute(); err != nil {
+		t.Fatal(err)
+	}
+	root.SetArgs([]string{"help", "utility"})
+	if err := root.Execute(); err != nil {
+		t.Fatal(err)
+	}
+	if got := formatRootFlag(root.PersistentFlags().Lookup("value")); !strings.Contains(got, "-x") {
+		t.Fatalf("formatted flag = %q", got)
+	}
+	if got := formatRootFlag(root.PersistentFlags().Lookup("hidden")); !strings.Contains(got, "--hidden") {
+		t.Fatalf("formatted long flag = %q", got)
+	}
+	_ = commandShort(&cobra.Command{Use: "help", Short: "Help about any command"})
+	renderRootGlobalFlags(&cobra.Command{Use: "no-flags"})
+	edition.Override(&edition.Hooks{})
+	SetDynamicServers(nil)
+	_ = visibleMCPRootCommands(root)
+}
+
+func TestCrossPlatformCoverageConfigAndTokenSeamsCoverage(t *testing.T) {
+	oldHome, oldExe, oldEval := userHomeDir, executablePath, evaluateSymlink
+	oldEdition := edition.Get()
+	t.Cleanup(func() {
+		userHomeDir, executablePath, evaluateSymlink = oldHome, oldExe, oldEval
+		edition.Override(oldEdition)
+	})
+	t.Setenv("DWS_CONFIG_DIR", "")
+	edition.Override(&edition.Hooks{})
+	userHomeDir = func() (string, error) { return "", errors.New("home") }
+	executablePath = func() (string, error) { return "", errors.New("exe") }
+	if got := defaultConfigDir(); got != ".dws" {
+		t.Fatalf("fallback config dir = %q", got)
+	}
+	executablePath = func() (string, error) { return filepath.Join("", "tmp", "dws"), nil }
+	evaluateSymlink = func(string) (string, error) { return "", errors.New("link") }
+	if got := exeRelativeConfigDir(); !strings.HasSuffix(got, filepath.Join("tmp", ".dws")) {
+		t.Fatalf("executable config dir = %q", got)
+	}
+	userHomeDir = func() (string, error) { return "/home/test", nil }
+	if got := defaultConfigDir(); got != filepath.Join("/home/test", ".dws") {
+		t.Fatalf("home config dir = %q", got)
+	}
+	edition.Override(&edition.Hooks{ConfigDir: func() string { return "/edition" }})
+	if got := defaultConfigDir(); got != "/edition" {
+		t.Fatalf("edition config dir = %q", got)
+	}
+
+	oldProvider, oldManager := newAccessTokenProvider, newLegacyTokenManager
+	t.Cleanup(func() { newAccessTokenProvider, newLegacyTokenManager = oldProvider, oldManager })
+	newAccessTokenProvider = func(string) accessTokenGetter { return fakeAccessTokenGetter{err: authpkg.ErrTokenDecryption} }
+	if _, err := resolveAccessTokenFromDir(context.Background(), "unused"); !errors.Is(err, authpkg.ErrTokenDecryption) {
+		t.Fatalf("decryption error = %v", err)
+	}
+	newAccessTokenProvider = func(string) accessTokenGetter {
+		return fakeAccessTokenGetter{err: authpkg.ErrTokenDataNotFound}
+	}
+	newLegacyTokenManager = func(string) legacyTokenGetter { return fakeLegacyTokenGetter{token: " legacy "} }
+	if got, err := resolveAccessTokenFromDir(context.Background(), "unused"); err != nil || got != "legacy" {
+		t.Fatalf("legacy token = %q, %v", got, err)
+	}
+	authpkg.SetRuntimeProfile("corp:user")
+	t.Cleanup(func() { authpkg.SetRuntimeProfile("") })
+	if got, err := resolveAccessTokenFromDir(context.Background(), "unused"); got != "" || !errors.Is(err, authpkg.ErrTokenDataNotFound) {
+		t.Fatalf("explicit profile fallback = token %q error %v, want profile error", got, err)
+	}
+	authpkg.SetRuntimeProfile("")
+	newAccessTokenProvider = func(string) accessTokenGetter { return fakeAccessTokenGetter{err: errors.New("load")} }
+	newLegacyTokenManager = func(string) legacyTokenGetter { return fakeLegacyTokenGetter{err: errors.New("missing")} }
+	edition.Override(&edition.Hooks{})
+	other := filepath.Join(t.TempDir(), "other")
+	if _, err := ResolveAuxiliaryAccessToken(context.Background(), other, ""); err == nil {
+		t.Fatal("auxiliary provider failure succeeded")
+	}
+	newAccessTokenProvider = func(string) accessTokenGetter { return fakeAccessTokenGetter{err: authpkg.ErrTokenDecryption} }
+	if _, err := ResolveAuxiliaryAccessToken(context.Background(), other, ""); !errors.Is(err, authpkg.ErrTokenDecryption) {
+		t.Fatalf("auxiliary decryption error = %v", err)
+	}
+	t.Setenv("DWS_CONFIG_DIR", other)
+	ResetRuntimeTokenCache()
+	if _, err := ResolveAuxiliaryAccessToken(context.Background(), other, ""); err == nil {
+		t.Fatal("current config without credentials succeeded")
+	}
+	edition.Override(&edition.Hooks{IsEmbedded: true})
+	if !strings.Contains(noCredentialsError().Error(), "认证") {
+		t.Fatal("embedded credentials error changed")
+	}
+}
+
+func TestCrossPlatformCoverageForceRefreshAndStdioFailureCoverage(t *testing.T) {
+	oldLoad, oldFactory := loadRefreshTokenData, newRefreshProvider
+	oldStop := stopStdio
+	t.Cleanup(func() {
+		loadRefreshTokenData, newRefreshProvider = oldLoad, oldFactory
+		stopStdio = oldStop
+		stdioMu.Lock()
+		stdioClients = make(map[string]*transport.StdioClient)
+		stdioMu.Unlock()
+	})
+	fail := errors.New("failure")
+	_ = oldFactory(t.TempDir())
+	loadRefreshTokenData = func(string) (*authpkg.TokenData, error) { return nil, fail }
+	if _, err := ForceRefreshAccessToken(context.Background(), "config"); !errors.Is(err, fail) {
+		t.Fatalf("load rejected token error = %v", err)
+	}
+	loadRefreshTokenData = func(string) (*authpkg.TokenData, error) {
+		return &authpkg.TokenData{AccessToken: "rejected"}, nil
+	}
+	for _, tc := range []struct {
+		getter fakeAccessTokenGetter
+		want   string
+	}{
+		{getter: fakeAccessTokenGetter{err: fail}, want: "failure"},
+		{getter: fakeAccessTokenGetter{token: "  "}, want: "empty"},
+		{getter: fakeAccessTokenGetter{token: " refreshed "}},
+	} {
+		newRefreshProvider = func(string) rejectedAccessTokenRefresher { return tc.getter }
+		got, err := ForceRefreshAccessToken(context.Background(), "config")
+		if tc.want != "" && (err == nil || !strings.Contains(err.Error(), tc.want)) {
+			t.Fatalf("refresh error = %v, want %q", err, tc.want)
+		}
+		if tc.want == "" && (err != nil || got != "refreshed") {
+			t.Fatalf("refreshed token = %q, %v", got, err)
+		}
+	}
+
+	stopStdio = func(*transport.StdioClient) error { return fail }
+	RegisterStdioClient("all", transport.NewStdioClient("unused", nil, nil))
+	StopAllStdioClients()
+	RegisterStdioClient("one", transport.NewStdioClient("unused", nil, nil))
+	if !StopStdioClient("one") {
+		t.Fatal("registered stdio client not stopped")
+	}
+	RegisterStdioClient("plugin/server", transport.NewStdioClient("unused", nil, nil))
+	if got := StopStdioClientsByPlugin("plugin"); got != 1 {
+		t.Fatalf("stopped plugin clients = %d", got)
+	}
+}
+
+func TestCrossPlatformCoverageOverlayRecoveryHostAndHelperRemainingCoverage(t *testing.T) {
+	root := t.TempDir()
+	writeOverlay := filepath.Join(root, "overlay.json")
+	if err := os.WriteFile(writeOverlay, []byte(`{"toolOverrides":{"tool":{}}}`), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	for _, raw := range []json.RawMessage{
+		json.RawMessage(`"missing.json"`),
+		json.RawMessage(`"unterminated`),
+		json.RawMessage(`{`),
+		json.RawMessage(`"overlay.json"`),
+		json.RawMessage(`{"id":"","command":"","toolOverrides":{"tool":{}}}`),
+	} {
+		p := &plugin.Plugin{Root: root, Manifest: plugin.Manifest{Name: "plugin", Description: "description", MCPServers: map[string]*plugin.MCPServer{"server": {CLI: raw}}}}
+		overlay := resolveStdioOverlay(p, plugin.StdioServerClient{Key: "server", Client: transport.NewStdioClient("unused", nil, nil)})
+		if overlay.ID == "" || overlay.Command == "" {
+			t.Fatalf("overlay defaults missing: %#v", overlay)
+		}
+	}
+	p := &plugin.Plugin{Root: root, Manifest: plugin.Manifest{Name: "plugin", Description: "description", MCPServers: map[string]*plugin.MCPServer{"server": {CLI: json.RawMessage(`{}`)}}}}
+	if descriptor := registerStdioServerFromManifest(p, plugin.StdioServerClient{Key: "server"}); descriptor.Endpoint == "" {
+		t.Fatalf("empty overlay descriptor = %#v", descriptor)
+	}
+	p.Manifest.MCPServers["server"].CLI = json.RawMessage(`{"toolOverrides":{"tool":{}}}`)
+	if descriptor := registerStdioServerFromManifest(p, plugin.StdioServerClient{Key: "server", Client: transport.NewStdioClient("unused", nil, nil)}); descriptor.Endpoint == "" {
+		t.Fatalf("stdio overlay registration = %#v", descriptor)
+	}
+
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition); SetDynamicServers(nil) })
+	edition.Override(&edition.Hooks{ConfigDir: func() string { return "" }})
+	captureRuntimeFailure(executor.Invocation{}, nil, nil)
+	captureRuntimeFailure(executor.Invocation{}, errors.New("raw"), nil)
+
+	t.Setenv(authpkg.AgentCodeEnv, "")
+	if hostControlProviderFromEnv() != "" {
+		t.Fatal("host control enabled without agent code")
+	}
+	t.Setenv(authpkg.AgentCodeEnv, "agent")
+	t.Setenv(agentproduct.EnvName, "")
+	edition.Override(&edition.Hooks{MergeHeaders: func(headers map[string]string) map[string]string { return headers }})
+	if got := hostControlProviderFromEnv(); got != edition.DefaultOSSClawType {
+		t.Fatalf("default claw type = %q", got)
+	}
+	edition.Override(&edition.Hooks{MergeHeaders: func(map[string]string) map[string]string { return map[string]string{"claw-type": "custom"} }})
+	if got := effectiveClawType(); got != "custom" {
+		t.Fatalf("custom claw type = %q", got)
+	}
+}
+
+func TestCrossPlatformCoverageConfigAndCacheCommandRemainingCoverage(t *testing.T) {
+	for _, command := range []*cobra.Command{newConfigCommand(), newCacheCommand()} {
+		command.SetOut(io.Discard)
+		rootWrap := &cobra.Command{Use: "dws"}
+		rootWrap.PersistentFlags().String("format", "json", "")
+		rootWrap.AddCommand(command)
+		command.SetOut(io.Discard)
+		if err := command.RunE(command, nil); err != nil {
+			t.Fatal(err)
+		}
+	}
+	t.Setenv("DWS_CONFIG_DIR", "configured")
+	configCmd := &cobra.Command{Use: "config"}
+	var configOut bytes.Buffer
+	configCmd.SetOut(&configOut)
+	if err := writeConfigJSON(configCmd, filterVisible(nil), true); err != nil {
+		t.Fatal(err)
+	}
+	list := newConfigListCommand()
+	list.SetOut(io.Discard)
+	_ = list.Flags().Set("category", "core")
+	_ = list.Flags().Set("show-values", "true")
+	_ = list.Flags().Set("show-hidden", "true")
+	_ = list.Flags().Set("json", "true")
+	if err := runConfigList(list, nil); err != nil {
+		t.Fatal(err)
+	}
+
+	cacheRoot := &cobra.Command{Use: "root"}
+	cacheRoot.PersistentFlags().String("format", "", "")
+	cacheCmd := &cobra.Command{Use: "cache"}
+	cacheRoot.AddCommand(cacheCmd)
+	for _, format := range []string{"json", "pretty", "table"} {
+		_ = cacheRoot.PersistentFlags().Set("format", format)
+		cacheCmd.SetOut(io.Discard)
+		if err := printCacheCompatNotice(cacheCmd, "dws cache status"); err != nil {
+			t.Fatal(err)
+		}
+	}
+	fail := errors.New("write")
+	cacheCmd.SetOut(appFailWriter{err: fail})
+	_ = cacheRoot.PersistentFlags().Set("format", "pretty")
+	if err := printCacheCompatNotice(cacheCmd, "dws cache status"); !errors.Is(err, fail) {
+		t.Fatalf("pretty write error = %v", err)
+	}
+	_ = cacheRoot.PersistentFlags().Set("format", "table")
+	if err := printCacheCompatNotice(cacheCmd, "dws cache status"); !errors.Is(err, fail) {
+		t.Fatalf("table write error = %v", err)
+	}
+}

internal/app/app_tiny_coverage_test.go

@@ -0,0 +1,282 @@
+package app
+
+import (
+	"context"
+	"errors"
+	"io"
+	"io/fs"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/apiclient"
+	authpkg "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/auth"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/mcptypes"
+	"github.com/spf13/cobra"
+)
+
+func TestCrossPlatformCoverageAPIAndTimingRemainingCoverage(t *testing.T) {
+	oldProvider := newAppTokenProvider
+	oldClientID, oldClientSecret := apiClientID, apiClientSecret
+	oldMarshal, oldMkdir := timingMarshalIndent, timingMkdirAll
+	oldWrite, oldRemove, oldRename := timingWriteFile, timingRemove, timingRename
+	oldRead, oldHome := timingReadFile, timingUserHomeDir
+	t.Cleanup(func() {
+		newAppTokenProvider = oldProvider
+		apiClientID, apiClientSecret = oldClientID, oldClientSecret
+		timingMarshalIndent, timingMkdirAll = oldMarshal, oldMkdir
+		timingWriteFile, timingRemove, timingRename = oldWrite, oldRemove, oldRename
+		timingReadFile, timingUserHomeDir = oldRead, oldHome
+		authpkg.SetClientID("")
+		authpkg.SetClientSecret("")
+	})
+	fail := errors.New("failure")
+	apiClientID = func() string { return "" }
+	apiClientSecret = func() string { return "" }
+	if _, err := resolveRawAPIToken(context.Background(), ""); err == nil {
+		t.Fatal("missing raw API credentials succeeded")
+	}
+	apiClientID = func() string { return "<placeholder>" }
+	apiClientSecret = func() string { return "secret" }
+	if _, err := resolveRawAPIToken(context.Background(), ""); err == nil {
+		t.Fatal("placeholder raw API credentials succeeded")
+	}
+	apiClientID, apiClientSecret = authpkg.ClientID, authpkg.ClientSecret
+	authpkg.SetClientID("app-key")
+	authpkg.SetClientSecret("app-secret")
+	for _, tc := range []struct {
+		getter fakeAppTokenGetter
+		want   string
+	}{
+		{getter: fakeAppTokenGetter{err: fail}, want: "failure"},
+		{getter: fakeAppTokenGetter{token: " "}, want: "为空"},
+		{getter: fakeAppTokenGetter{token: " token "}},
+	} {
+		newAppTokenProvider = func(string, string, string) appTokenGetter { return tc.getter }
+		got, err := resolveRawAPIToken(context.Background(), "")
+		if tc.want != "" && (err == nil || !containsText(err.Error(), tc.want)) {
+			t.Fatalf("raw token error = %v, want %q", err, tc.want)
+		}
+		if tc.want == "" && (err != nil || got != "token") {
+			t.Fatalf("raw token = %q, %v", got, err)
+		}
+	}
+	server := httptest.NewServer(http.HandlerFunc(func(http.ResponseWriter, *http.Request) {}))
+	endpoint := server.URL
+	server.Close()
+	apiclient.AllowedHosts["127.0.0.1"] = true
+	t.Cleanup(func() { delete(apiclient.AllowedHosts, "127.0.0.1") })
+	cmd := &cobra.Command{Use: "api"}
+	cmd.SetContext(context.Background())
+	cmd.SetOut(io.Discard)
+	cmd.SetErr(io.Discard)
+	if err := runAPI(cmd, []string{"GET", "/path"}, &GlobalFlags{Token: "token", Timeout: 1}, &apiFlags{baseURL: endpoint}); err == nil {
+		t.Fatal("closed raw API endpoint succeeded")
+	}
+
+	collector := NewTimingCollector()
+	collector.Print(io.Discard)
+	t.Setenv(PerfReportEnv, t.TempDir()+"/report.json")
+	timingMarshalIndent = func(any, string, string) ([]byte, error) { return nil, fail }
+	collector.WriteReportIfEnabled("v", "cmd")
+	timingMarshalIndent = oldMarshal
+	timingUserHomeDir = func() (string, error) { return "", fail }
+	t.Setenv(PerfReportEnv, "auto")
+	collector.WriteReportIfEnabled("v", "cmd")
+	if defaultPerfReportPath() != "" {
+		t.Fatal("home-dir failure produced a report path")
+	}
+	t.Setenv(PerfReportEnv, t.TempDir()+"/report.json")
+	timingMkdirAll = func(string, os.FileMode) error { return fail }
+	collector.WriteReportIfEnabled("v", "cmd")
+	timingMkdirAll = oldMkdir
+	timingWriteFile = func(string, []byte, os.FileMode) error { return fail }
+	removed := false
+	timingRemove = func(string) error { removed = true; return nil }
+	collector.WriteReportIfEnabled("v", "cmd")
+	if !removed {
+		t.Fatal("failed temporary report was not removed")
+	}
+	timingWriteFile = oldWrite
+	timingRemove = oldRemove
+	renamed := false
+	timingRename = func(string, string) error { renamed = true; return fail }
+	collector.WriteReportIfEnabled("v", "cmd")
+	if !renamed {
+		t.Fatal("report rename was not attempted")
+	}
+	timingReadFile = func(string) ([]byte, error) { return []byte("{"), nil }
+	if _, err := LoadLatestReport(); err == nil {
+		t.Fatal("malformed performance report succeeded")
+	}
+}
+
+func TestCrossPlatformCoverageDirectRuntimeRemainingCoverage(t *testing.T) {
+	oldEdition := edition.Get()
+	t.Cleanup(func() {
+		edition.Override(oldEdition)
+		SetDynamicServers(nil)
+	})
+	configDir := t.TempDir()
+	t.Setenv("DWS_CONFIG_DIR", configDir)
+	for _, raw := range []string{
+		"not a url",
+		"https://mcp.dingtalk.com/path?q=1#fragment",
+		"https://pre-mcp.example.test:8443/path/",
+		"https://mcp.example.test/path/",
+	} {
+		if err := os.WriteFile(filepath.Join(configDir, "mcp_url"), []byte(raw), 0o600); err != nil {
+			t.Fatal(err)
+		}
+		if got := defaultPATGatewayBaseURL(); got == "" {
+			t.Fatalf("gateway for %q is blank", raw)
+		}
+	}
+
+	endpoints := map[string]string{}
+	products := map[string]bool{}
+	aliases := map[string]string{}
+	tools := map[string]string{}
+	registerDynamicServer(mcptypes.ServerDescriptor{CLI: mcptypes.CLIOverlay{Skip: true}}, endpoints, products, aliases, tools, false)
+	registerDynamicServer(mcptypes.ServerDescriptor{
+		Endpoint: "https://server.test",
+		CLI: mcptypes.CLIOverlay{
+			ID: "id", Command: "command", Aliases: []string{"alias", " "},
+			Tools:         []mcptypes.CLITool{{Name: "tool"}, {Name: " "}},
+			ToolOverrides: map[string]mcptypes.CLIToolOverride{"override": {}, " ": {}},
+		},
+	}, endpoints, products, aliases, tools, false)
+	if endpoints["command"] == "" || aliases["alias"] != "id" || tools["override"] == "" {
+		t.Fatalf("registered dynamic server = %#v %#v %#v", endpoints, aliases, tools)
+	}
+
+	SetDynamicServers(nil)
+	dynamicMu.Lock()
+	dynamicEndpoints = map[string]string{}
+	dynamicProducts = map[string]bool{}
+	dynamicAliases = map[string]string{}
+	dynamicToolEndpoints = map[string]string{}
+	dynamicMu.Unlock()
+	if got, ok := directRuntimeEndpoint(defaultPATProductID, ""); !ok || got == "" {
+		t.Fatal("cold-start PAT fallback did not resolve")
+	}
+	SetDynamicServers(nil)
+	if _, ok := directRuntimeEndpoint(" ", " "); ok {
+		t.Fatal("blank runtime endpoint resolved")
+	}
+	t.Setenv("DINGTALK_CUSTOM_MCP_URL", "https://override.test")
+	if got, ok := directRuntimeEndpoint("custom", ""); !ok || got != "https://override.test" {
+		t.Fatalf("environment runtime endpoint = %q, %v", got, ok)
+	}
+	if got, ok := directRuntimeEndpoint(devappProductID, ""); !ok || got == "" {
+		t.Fatal("devapp fallback did not resolve")
+	}
+	if got, ok := directRuntimeEndpoint(defaultPATProductID, ""); !ok || got == "" {
+		t.Fatal("PAT fallback did not resolve")
+	}
+	edition.Override(&edition.Hooks{
+		StaticServers: func() []edition.ServerInfo { return []edition.ServerInfo{{ID: "other", Endpoint: ""}} },
+		SupplementServers: func() []edition.ServerInfo {
+			return []edition.ServerInfo{{ID: "other", Endpoint: "https://other.test", Prefixes: []string{" ", "wanted"}}}
+		},
+	})
+	if got, ok := directRuntimeEndpoint("wanted", ""); !ok || got != "https://other.test" {
+		t.Fatalf("edition runtime endpoint = %q, %v", got, ok)
+	}
+
+	dynamicMu.Lock()
+	dynamicEndpoints, dynamicProducts, dynamicAliases, dynamicToolEndpoints = nil, nil, nil, nil
+	dynamicMu.Unlock()
+	AppendDynamicServer(mcptypes.ServerDescriptor{
+		Endpoint: "https://append.test",
+		CLI: mcptypes.CLIOverlay{
+			ID: "append", Command: "append-command", Aliases: []string{"append-alias"},
+			Tools: []mcptypes.CLITool{{Name: "append-tool"}},
+			ToolOverrides: map[string]mcptypes.CLIToolOverride{
+				"append-override": {}, "skip": {ServerOverride: "other"}, " ": {},
+			},
+		},
+	})
+	if got, ok := directRuntimeEndpoint("", "append-override"); !ok || got != "https://append.test" {
+		t.Fatalf("append override endpoint = %q, %v", got, ok)
+	}
+}
+
+func containsText(value, substring string) bool {
+	for i := 0; i+len(substring) <= len(value); i++ {
+		if value[i:i+len(substring)] == substring {
+			return true
+		}
+	}
+	return false
+}
+
+func TestCrossPlatformCoverageEmbeddedSkillAndTinyCommandsRemainingCoverage(t *testing.T) {
+	oldStat, oldTemp, oldRemove := embeddedSkillStat, embeddedSkillMkdirTemp, embeddedSkillRemoveAll
+	oldWalk, oldRead := embeddedSkillWalkDir, embeddedSkillReadFile
+	oldMkdir, oldWrite := embeddedSkillMkdirAll, embeddedSkillWriteFile
+	t.Cleanup(func() {
+		embeddedSkillStat, embeddedSkillMkdirTemp, embeddedSkillRemoveAll = oldStat, oldTemp, oldRemove
+		embeddedSkillWalkDir, embeddedSkillReadFile = oldWalk, oldRead
+		embeddedSkillMkdirAll, embeddedSkillWriteFile = oldMkdir, oldWrite
+	})
+	fail := errors.New("failure")
+	embeddedSkillStat = func(string) (os.FileInfo, error) { return nil, nil }
+	embeddedSkillMkdirTemp = func(string, string) (string, error) { return "", fail }
+	if _, _, err := materializeEmbeddedSkillSource("codex"); !errors.Is(err, fail) {
+		t.Fatalf("embedded mkdir error = %v", err)
+	}
+	embeddedSkillMkdirTemp = func(string, string) (string, error) { return t.TempDir(), nil }
+	removed := false
+	embeddedSkillRemoveAll = func(string) error { removed = true; return nil }
+	embeddedSkillWalkDir = func(_ string, fn fs.WalkDirFunc) error {
+		return fn("entry", nil, fail)
+	}
+	if _, _, err := materializeEmbeddedSkillSource("codex"); !errors.Is(err, fail) || !removed {
+		t.Fatalf("embedded walk error = %v, removed=%v", err, removed)
+	}
+	embeddedSkillWalkDir = func(_ string, fn fs.WalkDirFunc) error {
+		return fn("skills/codex/file", fakeSkillDirEntry{}, nil)
+	}
+	embeddedSkillReadFile = func(string) ([]byte, error) { return nil, fail }
+	if _, _, err := materializeEmbeddedSkillSource("codex"); !errors.Is(err, fail) {
+		t.Fatalf("embedded read error = %v", err)
+	}
+	embeddedSkillReadFile = func(string) ([]byte, error) { return []byte("skill"), nil }
+	embeddedSkillMkdirAll = func(string, os.FileMode) error { return fail }
+	if _, _, err := materializeEmbeddedSkillSource("codex"); !errors.Is(err, fail) {
+		t.Fatalf("embedded nested mkdir error = %v", err)
+	}
+	embeddedSkillWalkDir = func(_ string, fn fs.WalkDirFunc) error {
+		return fn("skills/codex/dir", fakeSkillDirEntry{dir: true}, nil)
+	}
+	if _, _, err := materializeEmbeddedSkillSource("codex"); !errors.Is(err, fail) {
+		t.Fatalf("embedded directory mkdir error = %v", err)
+	}
+	embeddedSkillWalkDir = func(_ string, fn fs.WalkDirFunc) error {
+		return fn("skills/codex/file", fakeSkillDirEntry{}, nil)
+	}
+	embeddedSkillMkdirAll = func(string, os.FileMode) error { return nil }
+	embeddedSkillWriteFile = func(string, []byte, os.FileMode) error { return fail }
+	if _, _, err := materializeEmbeddedSkillSource("codex"); !errors.Is(err, fail) {
+		t.Fatalf("embedded write error = %v", err)
+	}
+
+	merged := mergeTopLevelCommands([]*cobra.Command{nil, {}})
+	if len(merged) != 0 {
+		t.Fatalf("empty legacy commands = %#v", merged)
+	}
+	root := &cobra.Command{Use: "root"}
+	completion := newCompletionCommand(root)
+	if err := completion.RunE(completion, []string{"other"}); err != nil {
+		t.Fatal(err)
+	}
+	catalog := newCatalogCommand()
+	catalog.SetOut(io.Discard)
+	if err := catalog.RunE(catalog, nil); err != nil {
+		t.Fatal(err)
+	}
+}

internal/app/audit_command.go

@@ -0,0 +1,375 @@
+package app
+
+import (
+	"bufio"
+	"bytes"
+	"encoding/csv"
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/helpers"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/audit"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/corecmd/contract"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/output"
+	"github.com/spf13/cobra"
+)
+
+var (
+	auditCSVWrite = func(writer *csv.Writer, record []string) error { return writer.Write(record) }
+	auditCSVFlush = func(writer *csv.Writer) { writer.Flush() }
+	auditCSVError = func(writer *csv.Writer) error { return writer.Error() }
+	auditExit     = os.Exit
+	auditVerify   = audit.VerifyFile
+)
+
+func newAuditCommand() *cobra.Command {
+	// Product-level Agent routing Decl (migrated from selection/audit.json
+	// products.audit). Catalog assembly stamps provenance contract_final.
+	contract.RegisterProductDecl(contract.ProductDecl{
+		ID: "audit",
+		Selection: contract.ProductSelectionDecl{
+			AgentSummary: "查看、导出和校验本地操作审计日志",
+			UseWhen: []string{
+				"需要排查本机 CLI 操作审计记录，或验证审计文件完整性",
+			},
+			AvoidWhen: []string{
+				"查钉钉业务数据或发消息请用对应产品命令，不要用 audit",
+			},
+		},
+	})
+	cmd := &cobra.Command{
+		Use:   "audit",
+		Short: "操作审计日志管理",
+		Long:  "查看、导出和校验本地操作审计日志。",
+	}
+	cmd.AddCommand(
+		newAuditTailCommand(),
+		newAuditExportCommand(),
+		newAuditVerifyCommand(),
+	)
+	return cmd
+}
+
+func newAuditTailCommand() *cobra.Command {
+	var n int
+	cmd := &cobra.Command{
+		Use:   "tail",
+		Short: "查看最近的审计记录",
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if n < 1 {
+				return fmt.Errorf("--lines 必须为正整数，收到 %d", n)
+			}
+			dir := auditDir()
+			file, err := audit.LatestAuditFile(dir)
+			if err != nil {
+				return fmt.Errorf("无审计记录: %w", err)
+			}
+			lines, err := tailFile(file, n)
+			if err != nil {
+				return err
+			}
+			for _, line := range lines {
+				fmt.Println(line)
+			}
+			return nil
+		},
+	}
+	cmd.Flags().IntVarP(&n, "lines", "n", 20, "显示最近 N 条记录")
+	helpers.DeclareLeafMetadata(cmd, helpers.LeafSpec{
+		Safety: contract.SafetySpec{
+			Effect: "read", Risk: "low",
+			Confirmation: "not_required", Idempotency: "idempotent",
+		},
+		Contract: helpers.LeafContract{
+			Identity: contract.ToolIdentitySpec{
+				ProductID:      "audit",
+				Name:           "tail",
+				CanonicalPath:  "audit.tail",
+				CLIPath:        "audit tail",
+				PrimaryCLIPath: "audit tail",
+			},
+			Description: "查看本地操作审计日志最近 N 条记录",
+			Interface: &contract.InterfaceSpec{
+				Mode:         "local",
+				Availability: "available",
+				Reason:       "命令读取本地审计日志尾部，不绑定 pinned MCP RPC",
+			},
+			Selection: contract.SelectionSpec{
+				AgentSummary: "查看本地操作审计日志最近 N 条记录",
+				UseWhen:      []string{"需要快速查看最近写入的审计记录（默认最近 20 条）"},
+				AvoidWhen: []string{
+					"需要按日期范围整段导出用 audit export",
+					"需要校验哈希链用 audit verify",
+				},
+				Examples: []string{
+					"dws audit tail",
+					"dws audit tail --lines 50",
+				},
+			},
+		},
+	})
+	return cmd
+}
+
+func newAuditExportCommand() *cobra.Command {
+	var since, until, format string
+	cmd := &cobra.Command{
+		Use:   "export",
+		Short: "导出审计日志",
+		RunE: func(cmd *cobra.Command, args []string) error {
+			dir := auditDir()
+
+			sinceDate := strings.ReplaceAll(since, "-", "")
+			untilDate := strings.ReplaceAll(until, "-", "")
+
+			files, err := audit.AuditFilesInRange(dir, sinceDate, untilDate)
+			if err != nil {
+				return fmt.Errorf("查找审计文件失败: %w", err)
+			}
+			if len(files) == 0 {
+				return fmt.Errorf("指定范围内无审计文件")
+			}
+
+			switch format {
+			case "jsonl":
+				return exportJSONL(files)
+			case "csv":
+				return exportCSV(files)
+			default:
+				return fmt.Errorf("不支持的格式: %s（可选 jsonl, csv）", format)
+			}
+		},
+	}
+	cmd.Flags().StringVar(&since, "since", "", "起始日期 (YYYY-MM-DD)")
+	cmd.Flags().StringVar(&until, "until", "", "截止日期 (YYYY-MM-DD)")
+	cmd.Flags().StringVar(&format, "format", "jsonl", "输出格式: jsonl 或 csv")
+	helpers.DeclareLeafMetadata(cmd, helpers.LeafSpec{
+		Safety: contract.SafetySpec{
+			Effect: "read", Risk: "low",
+			Confirmation: "not_required", Idempotency: "idempotent",
+		},
+		Contract: helpers.LeafContract{
+			Identity: contract.ToolIdentitySpec{
+				ProductID:      "audit",
+				Name:           "export",
+				CanonicalPath:  "audit.export",
+				CLIPath:        "audit export",
+				PrimaryCLIPath: "audit export",
+			},
+			Description: "按日期范围导出本地操作审计日志（jsonl 或 csv）",
+			Interface: &contract.InterfaceSpec{
+				Mode:         "local",
+				Availability: "available",
+				Reason:       "命令读取并导出本地审计日志文件，不绑定 pinned MCP RPC",
+			},
+			Selection: contract.SelectionSpec{
+				AgentSummary: "按日期范围导出本地操作审计日志（jsonl 或 csv）",
+				UseWhen:      []string{"需要把本地审计日志导出为 jsonl/csv，或按 --since/--until 取一段时间"},
+				AvoidWhen: []string{
+					"只看最近几条用 audit tail",
+					"只校验哈希链完整性用 audit verify",
+				},
+				Examples: []string{
+					"dws audit export --format jsonl",
+					"dws audit export --since 2026-07-01 --until 2026-07-14 --format csv",
+				},
+			},
+		},
+	})
+	return cmd
+}
+
+func newAuditVerifyCommand() *cobra.Command {
+	var file string
+	cmd := &cobra.Command{
+		Use:   "verify",
+		Short: "校验审计日志哈希链完整性",
+		RunE: func(cmd *cobra.Command, args []string) error {
+			target := file
+			if target == "" {
+				dir := auditDir()
+				var err error
+				target, err = audit.LatestAuditFile(dir)
+				if err != nil {
+					return fmt.Errorf("无审计文件: %w", err)
+				}
+			}
+
+			valid, brokenAt, verifyErr := auditVerify(target)
+			if output.ResolveFormat(cmd, output.FormatTable) == output.FormatJSON {
+				if verifyErr != nil && brokenAt == 0 {
+					return fmt.Errorf("校验失败: %w", verifyErr)
+				}
+				payload := map[string]any{
+					"valid":    valid,
+					"file":     target,
+					"brokenAt": brokenAt,
+				}
+				if verifyErr != nil {
+					payload["reason"] = verifyErr.Error()
+				}
+				if err := output.WriteCommandPayload(cmd, payload, output.FormatTable); err != nil {
+					return err
+				}
+				if !valid {
+					auditExit(1)
+				}
+				return nil
+			}
+			if verifyErr != nil {
+				return fmt.Errorf("校验失败: %w", verifyErr)
+			}
+			if valid {
+				fmt.Printf("✓ %s 哈希链完整（全部通过）\n", filepath.Base(target))
+			} else {
+				fmt.Printf("✗ %s 哈希链在第 %d 行断裂\n", filepath.Base(target), brokenAt)
+				auditExit(1)
+			}
+			return nil
+		},
+	}
+	cmd.Flags().StringVar(&file, "file", "", "指定审计文件路径（默认最新文件）")
+	helpers.DeclareLeafMetadata(cmd, helpers.LeafSpec{
+		Safety: contract.SafetySpec{
+			Effect: "read", Risk: "low",
+			Confirmation: "not_required", Idempotency: "idempotent",
+		},
+		Contract: helpers.LeafContract{
+			Identity: contract.ToolIdentitySpec{
+				ProductID:      "audit",
+				Name:           "verify",
+				CanonicalPath:  "audit.verify",
+				CLIPath:        "audit verify",
+				PrimaryCLIPath: "audit verify",
+			},
+			Description: "校验本地审计日志文件的哈希链完整性",
+			Interface: &contract.InterfaceSpec{
+				Mode:         "local",
+				Availability: "available",
+				Reason:       "命令校验本地审计日志哈希链，不绑定 pinned MCP RPC",
+			},
+			Selection: contract.SelectionSpec{
+				AgentSummary: "校验本地审计日志文件的哈希链完整性",
+				UseWhen:      []string{"怀疑审计文件被篡改，或需要确认最新/指定文件哈希链是否完整"},
+				AvoidWhen:    []string{"只浏览或导出日志内容时用 audit tail / audit export"},
+				Examples: []string{
+					"dws audit verify",
+					"dws audit verify --file /path/to/audit.jsonl",
+				},
+			},
+		},
+	})
+	return cmd
+}
+
+func auditDir() string {
+	if dir := os.Getenv(audit.EnvAuditDir); dir != "" {
+		return dir
+	}
+	return filepath.Join(defaultConfigDir(), "audit")
+}
+
+func tailFile(path string, n int) ([]string, error) {
+	f, err := os.Open(path)
+	if err != nil {
+		return nil, err
+	}
+	defer f.Close()
+
+	var lines []string
+	scanner := bufio.NewScanner(f)
+	scanner.Buffer(make([]byte, 1024*1024), 1024*1024)
+	for scanner.Scan() {
+		lines = append(lines, scanner.Text())
+	}
+	if err := scanner.Err(); err != nil {
+		return nil, err
+	}
+
+	if len(lines) > n {
+		lines = lines[len(lines)-n:]
+	}
+	return lines, nil
+}
+
+func exportJSONL(files []string) error {
+	for _, file := range files {
+		f, err := os.Open(file)
+		if err != nil {
+			return err
+		}
+		scanner := bufio.NewScanner(f)
+		scanner.Buffer(make([]byte, 1024*1024), 1024*1024)
+		for scanner.Scan() {
+			fmt.Println(scanner.Text())
+		}
+		f.Close()
+		if err := scanner.Err(); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func exportCSV(files []string) error {
+	w := csv.NewWriter(os.Stdout)
+
+	header := []string{"timestamp", "execution_id", "user_id", "corp_id", "product", "command", "result", "duration_ms", "error_category"}
+	if err := auditCSVWrite(w, header); err != nil {
+		return fmt.Errorf("写入 CSV 表头失败: %w", err)
+	}
+
+	for _, file := range files {
+		f, err := os.Open(file)
+		if err != nil {
+			return err
+		}
+		scanner := bufio.NewScanner(f)
+		scanner.Buffer(make([]byte, 1024*1024), 1024*1024)
+		lineNum := 0
+		for scanner.Scan() {
+			lineNum++
+			line := scanner.Bytes()
+			if len(bytes.TrimSpace(line)) == 0 {
+				continue
+			}
+			var evt audit.Event
+			if err := json.Unmarshal(line, &evt); err != nil {
+				f.Close()
+				return fmt.Errorf("解析审计记录失败 %s:%d: %w", file, lineNum, err)
+			}
+			row := []string{
+				evt.Timestamp.Format(time.RFC3339),
+				evt.ExecutionID,
+				evt.Actor.UserID,
+				evt.Actor.CorpID,
+				evt.Product,
+				evt.Command,
+				evt.Result,
+				strconv.FormatInt(evt.DurationMs, 10),
+				evt.ErrCategory,
+			}
+			if err := auditCSVWrite(w, row); err != nil {
+				f.Close()
+				return fmt.Errorf("写入 CSV 记录失败: %w", err)
+			}
+		}
+		if err := scanner.Err(); err != nil {
+			f.Close()
+			return err
+		}
+		f.Close()
+	}
+
+	auditCSVFlush(w)
+	if err := auditCSVError(w); err != nil {
+		return fmt.Errorf("刷新 CSV 输出失败: %w", err)
+	}
+	return nil
+}

internal/app/audit_command_test.go

@@ -0,0 +1,106 @@
+package app
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func TestAuditTailRejectsNonPositiveLines(t *testing.T) {
+	for _, n := range []string{"0", "-1"} {
+		cmd := newAuditTailCommand()
+		cmd.SetArgs([]string{"--lines", n})
+		cmd.SilenceUsage = true
+		cmd.SilenceErrors = true
+		err := cmd.Execute()
+		if err == nil {
+			t.Fatalf("--lines %s: expected error, got nil", n)
+		}
+		if !strings.Contains(err.Error(), "正整数") {
+			t.Fatalf("--lines %s: unexpected error: %v", n, err)
+		}
+	}
+}
+
+func TestTailFileReturnsLastN(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "audit-20260101.jsonl")
+	if err := os.WriteFile(path, []byte("a\nb\nc\nd\ne\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	lines, err := tailFile(path, 2)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(lines) != 2 || lines[0] != "d" || lines[1] != "e" {
+		t.Fatalf("got %v, want [d e]", lines)
+	}
+}
+
+func TestExportCSVWritesHeaderAndRows(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "audit-20260101.jsonl")
+	rec := `{"timestamp":"2026-01-01T00:00:00Z","execution_id":"e1","actor":{"user_id":"u1","corp_id":"c1"},"product":"calendar","command":"event_list","result":"success","duration_ms":12,"hash":"h","prev_hash":""}`
+	if err := os.WriteFile(path, []byte(rec+"\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	stdout := os.Stdout
+	r, w, err := os.Pipe()
+	if err != nil {
+		t.Fatal(err)
+	}
+	os.Stdout = w
+	exportErr := exportCSV([]string{path})
+	w.Close()
+	os.Stdout = stdout
+
+	if exportErr != nil {
+		t.Fatalf("exportCSV error: %v", exportErr)
+	}
+	buf := make([]byte, 4096)
+	n, _ := r.Read(buf)
+	out := string(buf[:n])
+	if !strings.Contains(out, "timestamp,execution_id") {
+		t.Fatalf("missing CSV header, got: %q", out)
+	}
+	if !strings.Contains(out, "e1") || !strings.Contains(out, "event_list") {
+		t.Fatalf("missing CSV row data, got: %q", out)
+	}
+}
+
+// TestExportCSVFailsOnMalformedJSON guards the reviewer's V9 finding: a corrupt
+// JSONL line must surface an error with file/line evidence instead of being
+// silently skipped while the command exits 0.
+func TestExportCSVFailsOnMalformedJSON(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "audit-20260101.jsonl")
+	good := `{"timestamp":"2026-01-01T00:00:00Z","execution_id":"e1","actor":{"user_id":"u1"},"product":"calendar","command":"event_list","result":"success","duration_ms":1,"hash":"h","prev_hash":""}`
+	if err := os.WriteFile(path, []byte(good+"\nnot-json\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	stdout := os.Stdout
+	r, w, err := os.Pipe()
+	if err != nil {
+		t.Fatal(err)
+	}
+	os.Stdout = w
+	exportErr := exportCSV([]string{path})
+	w.Close()
+	os.Stdout = stdout
+	// Drain the pipe so the writer never blocks.
+	buf := make([]byte, 4096)
+	_, _ = r.Read(buf)
+
+	if exportErr == nil {
+		t.Fatal("expected error on malformed JSONL, got nil")
+	}
+	if !strings.Contains(exportErr.Error(), "解析审计记录失败") {
+		t.Fatalf("error missing parse context: %v", exportErr)
+	}
+	if !strings.Contains(exportErr.Error(), ":2") {
+		t.Fatalf("error missing line evidence: %v", exportErr)
+	}
+}

internal/app/audit_coverage_more_test.go

@@ -0,0 +1,272 @@
+package app
+
+import (
+	"encoding/csv"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/audit"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/auth"
+	apperrors "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/errors"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/executor"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/logging"
+)
+
+type auditCoverageSink struct {
+	events   []*audit.Event
+	emitErr  error
+	closeErr error
+}
+
+func (sink *auditCoverageSink) Emit(event *audit.Event) error {
+	sink.events = append(sink.events, event)
+	return sink.emitErr
+}
+
+func (sink *auditCoverageSink) Close() error { return sink.closeErr }
+
+func TestCrossPlatformCoverageAuditCommandsAndFileHelpersCoverage(t *testing.T) {
+	originalExit, originalVerify := auditExit, auditVerify
+	t.Cleanup(func() { auditExit, auditVerify = originalExit, originalVerify })
+	dir := t.TempDir()
+	t.Setenv(audit.EnvAuditDir, dir)
+	if auditDir() != dir {
+		t.Fatalf("auditDir() = %q", auditDir())
+	}
+
+	tail := newAuditTailCommand()
+	tail.SetArgs([]string{"--lines", "1"})
+	if err := tail.Execute(); err == nil || !strings.Contains(err.Error(), "无审计记录") {
+		t.Fatalf("audit tail(empty) error = %v", err)
+	}
+	path := filepath.Join(dir, "audit-20260101.jsonl")
+	if err := os.WriteFile(path, []byte("one\ntwo\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	tail = newAuditTailCommand()
+	tail.SetArgs([]string{"--lines", "1"})
+	if err := tail.Execute(); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := tailFile(filepath.Join(dir, "missing"), 1); err == nil {
+		t.Fatal("tailFile(missing) error = nil")
+	}
+	tailErrorDir := t.TempDir()
+	if err := os.Mkdir(filepath.Join(tailErrorDir, "audit-20260101.jsonl"), 0o700); err != nil {
+		t.Fatal(err)
+	}
+	t.Setenv(audit.EnvAuditDir, tailErrorDir)
+	tail = newAuditTailCommand()
+	if err := tail.Execute(); err == nil {
+		t.Fatal("audit tail(directory record) error = nil")
+	}
+	oversize := filepath.Join(dir, "oversize")
+	if err := os.WriteFile(oversize, []byte(strings.Repeat("x", 2*1024*1024)), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := tailFile(oversize, 1); err == nil {
+		t.Fatal("tailFile(oversize) error = nil")
+	}
+
+	exportDir := t.TempDir()
+	t.Setenv(audit.EnvAuditDir, exportDir)
+	export := newAuditExportCommand()
+	if err := export.Execute(); err == nil || !strings.Contains(err.Error(), "无审计文件") {
+		t.Fatalf("audit export(empty) error = %v", err)
+	}
+	if err := os.WriteFile(filepath.Join(exportDir, "audit-20260102.jsonl"), []byte("{}\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	for _, format := range []string{"jsonl", "csv"} {
+		export = newAuditExportCommand()
+		export.SetArgs([]string{"--since", "2026-01-01", "--until", "2026-01-03", "--format", format})
+		if err := export.Execute(); err != nil {
+			t.Fatalf("audit export(%s) error = %v", format, err)
+		}
+	}
+	export = newAuditExportCommand()
+	export.SetArgs([]string{"--format", "xml"})
+	if err := export.Execute(); err == nil || !strings.Contains(err.Error(), "不支持的格式") {
+		t.Fatalf("audit export(xml) error = %v", err)
+	}
+	t.Setenv(audit.EnvAuditDir, filepath.Join(exportDir, "missing"))
+	export = newAuditExportCommand()
+	if err := export.Execute(); err == nil || !strings.Contains(err.Error(), "查找审计文件失败") {
+		t.Fatalf("audit export(missing dir) error = %v", err)
+	}
+
+	if err := exportJSONL([]string{filepath.Join(dir, "missing")}); err == nil {
+		t.Fatal("exportJSONL(missing) error = nil")
+	}
+	if err := exportJSONL([]string{oversize}); err == nil {
+		t.Fatal("exportJSONL(oversize) error = nil")
+	}
+	if err := exportCSV([]string{filepath.Join(dir, "missing")}); err == nil {
+		t.Fatal("exportCSV(missing) error = nil")
+	}
+	blank := filepath.Join(dir, "blank")
+	if err := os.WriteFile(blank, []byte("\n \n{}\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if err := exportCSV([]string{blank}); err != nil {
+		t.Fatal(err)
+	}
+	if err := exportCSV([]string{oversize}); err == nil {
+		t.Fatal("exportCSV(oversize) error = nil")
+	}
+
+	originalWrite, originalFlush, originalError := auditCSVWrite, auditCSVFlush, auditCSVError
+	t.Cleanup(func() { auditCSVWrite, auditCSVFlush, auditCSVError = originalWrite, originalFlush, originalError })
+	auditCSVWrite = func(*csv.Writer, []string) error { return errors.New("write") }
+	if err := exportCSV(nil); err == nil || !strings.Contains(err.Error(), "表头") {
+		t.Fatalf("exportCSV(header error) = %v", err)
+	}
+	calls := 0
+	auditCSVWrite = func(writer *csv.Writer, row []string) error {
+		calls++
+		if calls > 1 {
+			return errors.New("row")
+		}
+		return originalWrite(writer, row)
+	}
+	if err := exportCSV([]string{blank}); err == nil || !strings.Contains(err.Error(), "记录") {
+		t.Fatalf("exportCSV(row error) = %v", err)
+	}
+	auditCSVWrite = originalWrite
+	auditCSVError = func(*csv.Writer) error { return errors.New("flush") }
+	if err := exportCSV(nil); err == nil || !strings.Contains(err.Error(), "刷新") {
+		t.Fatalf("exportCSV(flush error) = %v", err)
+	}
+
+	t.Setenv(audit.EnvAuditDir, t.TempDir())
+	verify := newAuditVerifyCommand()
+	if err := verify.Execute(); err == nil || !strings.Contains(err.Error(), "无审计文件") {
+		t.Fatalf("audit verify(empty) error = %v", err)
+	}
+	verify = newAuditVerifyCommand()
+	verify.SetArgs([]string{"--file", filepath.Join(dir, "missing")})
+	if err := verify.Execute(); err == nil || !strings.Contains(err.Error(), "校验失败") {
+		t.Fatalf("audit verify(missing) error = %v", err)
+	}
+	validDir := t.TempDir()
+	writer, err := audit.NewDateRotatingWriter(validDir, 1)
+	if err != nil {
+		t.Fatal(err)
+	}
+	sink := audit.NewFileSink(writer, audit.NewChain(validDir), nil)
+	if err := sink.Emit(&audit.Event{Timestamp: time.Now(), Product: "test", Command: "ok"}); err != nil {
+		t.Fatal(err)
+	}
+	if err := sink.Close(); err != nil {
+		t.Fatal(err)
+	}
+	validFile, err := audit.LatestAuditFile(validDir)
+	if err != nil {
+		t.Fatal(err)
+	}
+	verify = newAuditVerifyCommand()
+	verify.SetArgs([]string{"--file", validFile})
+	if err := verify.Execute(); err != nil {
+		t.Fatal(err)
+	}
+	broken := filepath.Join(t.TempDir(), "audit-broken.jsonl")
+	if err := os.WriteFile(broken, []byte(`{"prev_hash":"wrong","hash":"wrong"}`+"\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	exitCode := 0
+	auditExit = func(code int) { exitCode = code }
+	auditVerify = func(string) (bool, int, error) { return false, 1, nil }
+	verify = newAuditVerifyCommand()
+	verify.SetArgs([]string{"--file", broken})
+	if err := verify.Execute(); err != nil || exitCode != 1 {
+		t.Fatalf("audit verify(broken) = %v, exit=%d", err, exitCode)
+	}
+	t.Setenv(audit.EnvAuditDir, "")
+	if auditDir() == "" {
+		t.Fatal("default auditDir() is empty")
+	}
+}
+
+func TestCrossPlatformCoverageAuditRuntimeCoverage(t *testing.T) {
+	previousSink, previousLoader := sharedAuditSink, loadTokenForProfile
+	t.Cleanup(func() {
+		sharedAuditSink = previousSink
+		loadTokenForProfile = previousLoader
+		auditSinkOnce, auditCloseOnce = sync.Once{}, sync.Once{}
+		// The process-wide sink was initialized by TestMain. Preserve that
+		// initialized state when restoring it: leaving auditSinkOnce unused
+		// lets a later runner overwrite the live sink without closing its
+		// .audit.lock handle, which makes TestMain cleanup fail on Windows.
+		auditSinkOnce.Do(func() {})
+		if got := setupAuditSink(); got != previousSink {
+			t.Errorf("restored audit sink = %T, want original %T", got, previousSink)
+		}
+		resetAuditIdentityCache()
+	})
+
+	sharedAuditSink = nil
+	auditCloseOnce = sync.Once{}
+	CloseAuditSink()
+	failedClose := &auditCoverageSink{closeErr: errors.New("close")}
+	sharedAuditSink = failedClose
+	auditCloseOnce = sync.Once{}
+	CloseAuditSink()
+
+	bad := filepath.Join(t.TempDir(), "file")
+	if err := os.WriteFile(bad, []byte("x"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	t.Setenv(audit.EnvAudit, "1")
+	t.Setenv(audit.EnvAuditDir, filepath.Join(bad, "child"))
+	auditSinkOnce = sync.Once{}
+	sharedAuditSink = nil
+	if _, ok := setupAuditSink().(audit.NopSink); !ok {
+		t.Fatalf("setupAuditSink(error) = %T", sharedAuditSink)
+	}
+
+	t.Setenv(audit.EnvAuditDebug, "1")
+	fileLogger = logging.Setup(t.TempDir())
+	t.Cleanup(func() {
+		if fileLogger != nil {
+			fileLogger.Close()
+			fileLogger = nil
+		}
+	})
+	auditReport("coverage %d", 1)
+	loadTokenForProfile = func(string, string) (*auth.TokenData, error) { return nil, errors.New("identity") }
+	resetAuditIdentityCache()
+	if actor, _ := auditIdentity(); actor.UserID != "" {
+		t.Fatalf("auditIdentity(error) = %+v", actor)
+	}
+
+	invocation := executor.Invocation{CanonicalProduct: "calendar", Tool: "list", Params: map[string]any{"token": "secret"}}
+	emitAudit(nil, "nil", time.Now(), invocation, "https://example.com?token=secret", nil, "test")
+	emitAudit(audit.NopSink{}, "nop", time.Now(), invocation, "", nil, "test")
+	recording := &auditCoverageSink{}
+	emitAudit(recording, "ok", time.Now(), invocation, "https://example.com?token=secret", nil, "test")
+	if len(recording.events) != 1 || recording.events[0].Result != "success" {
+		t.Fatalf("successful audit events = %#v", recording.events)
+	}
+	typed := &apperrors.Error{Category: apperrors.CategoryAuth, Reason: "expired"}
+	emitAudit(recording, "typed", time.Now(), invocation, "", typed, "test")
+	if recording.events[1].ErrReason != "expired" {
+		t.Fatalf("typed audit event = %#v", recording.events[1])
+	}
+	recording.emitErr = errors.New("emit")
+	emitAudit(recording, "failed", time.Now(), invocation, "", errors.New("plain"), "test")
+	if category, reason := classifyAuditError(nil); category != "" || reason != "" {
+		t.Fatalf("classifyAuditError(nil) = %q, %q", category, reason)
+	}
+	if category, reason := classifyAuditError(fmt.Errorf("wrapped: %w", typed)); category != string(apperrors.CategoryAuth) || reason != "expired" {
+		t.Fatalf("classifyAuditError(typed) = %q, %q", category, reason)
+	}
+	if category, reason := classifyAuditError(errors.New("plain")); category != "unknown" || reason != "plain" {
+		t.Fatalf("classifyAuditError(plain) = %q, %q", category, reason)
+	}
+}

internal/app/audit_output_contract_test.go

@@ -0,0 +1,65 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+
+package app
+
+import (
+	"bytes"
+	"encoding/json"
+	"errors"
+	"testing"
+
+	"github.com/spf13/cobra"
+)
+
+func executeAuditVerifyJSON(t *testing.T, verify func(string) (bool, int, error)) (map[string]any, int, error) {
+	t.Helper()
+	previousVerify, previousExit := auditVerify, auditExit
+	t.Cleanup(func() {
+		auditVerify, auditExit = previousVerify, previousExit
+	})
+	auditVerify = verify
+	exitCode := 0
+	auditExit = func(code int) { exitCode = code }
+
+	root := &cobra.Command{Use: "dws"}
+	root.SilenceErrors = true
+	root.SilenceUsage = true
+	root.PersistentFlags().String("format", "json", "output format")
+	root.AddCommand(newAuditVerifyCommand())
+	var stdout bytes.Buffer
+	root.SetOut(&stdout)
+	root.SetArgs([]string{"verify", "--file", "/tmp/audit.jsonl"})
+	err := root.Execute()
+
+	var payload map[string]any
+	if decodeErr := json.Unmarshal(stdout.Bytes(), &payload); decodeErr != nil {
+		t.Fatalf("audit verify stdout must be one JSON document: %v\n%s", decodeErr, stdout.String())
+	}
+	return payload, exitCode, err
+}
+
+func TestCrossPlatformCoverageAuditVerifyJSONOutputIsSingleDocument(t *testing.T) {
+	payload, exitCode, err := executeAuditVerifyJSON(t, func(string) (bool, int, error) {
+		return true, 0, nil
+	})
+	if err != nil || exitCode != 0 {
+		t.Fatalf("audit verify returned err=%v exit=%d", err, exitCode)
+	}
+	if payload["valid"] != true || payload["file"] != "/tmp/audit.jsonl" || payload["brokenAt"] != float64(0) {
+		t.Fatalf("unexpected audit payload: %#v", payload)
+	}
+}
+
+func TestCrossPlatformCoverageAuditVerifyBrokenJSONIncludesReasonBeforeExit(t *testing.T) {
+	payload, exitCode, err := executeAuditVerifyJSON(t, func(string) (bool, int, error) {
+		return false, 3, errors.New("prev_hash mismatch")
+	})
+	if err != nil || exitCode != 1 {
+		t.Fatalf("broken audit verify returned err=%v exit=%d", err, exitCode)
+	}
+	if payload["valid"] != false || payload["brokenAt"] != float64(3) || payload["reason"] != "prev_hash mismatch" {
+		t.Fatalf("unexpected broken audit payload: %#v", payload)
+	}
+}

internal/app/audit_runtime.go

@@ -0,0 +1,164 @@
+package app
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"runtime"
+	"sync"
+	"time"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/audit"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/auth"
+	apperrors "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/errors"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/executor"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/logging"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/transport"
+)
+
+var (
+	auditSinkOnce   sync.Once
+	auditCloseOnce  sync.Once
+	sharedAuditSink audit.Sink
+
+	auditIDMu      sync.Mutex
+	cachedActor    audit.Actor
+	cachedAgentID  string
+	cachedProfile  string
+	identityLoaded bool
+
+	// loadTokenForProfile is the profile-scoped token loader. It is a package
+	// variable so profile-switch Actor attribution can be tested deterministically
+	// without touching the OS keychain.
+	loadTokenForProfile = auth.LoadTokenDataForProfile
+)
+
+// setupAuditSink builds the process-wide audit sink once and caches it so the
+// runner and the shutdown hook share a single writer/forwarder instance.
+func setupAuditSink() audit.Sink {
+	auditSinkOnce.Do(func() {
+		sink, err := audit.BuildSink(defaultConfigDir(), auditReport)
+		if err != nil {
+			auditReport("initialization failed, audit disabled for this session: %v", err)
+			sharedAuditSink = audit.NopSink{}
+			return
+		}
+		sharedAuditSink = sink
+	})
+	return sharedAuditSink
+}
+
+// CloseAuditSink flushes in-flight remote forwards and closes the audit writer.
+// It is invoked from an unconditional defer in Execute so the drain happens for
+// both successful and failed commands (Cobra skips PersistentPostRunE when RunE
+// returns an error). The sync.Once makes repeated calls safe.
+func CloseAuditSink() {
+	auditCloseOnce.Do(func() {
+		if sharedAuditSink == nil {
+			return
+		}
+		if err := sharedAuditSink.Close(); err != nil {
+			auditReport("close failed: %v", err)
+		}
+	})
+}
+
+// auditReport routes non-fatal audit-subsystem diagnostics to the structured
+// file log (always, when available) and to stderr when DWS_AUDIT_DEBUG is set,
+// so init/write/forward failures are observable instead of silently swallowed.
+func auditReport(format string, args ...any) {
+	msg := "audit: " + fmt.Sprintf(format, args...)
+	if l := FileLoggerInstance(); l != nil {
+		l.Warn(msg)
+	}
+	if audit.DebugEnabled() {
+		fmt.Fprintln(os.Stderr, "[dws] "+msg)
+	}
+}
+
+// auditIdentity resolves the Actor for the active runtime profile. The result
+// is cached per-profile so a profile switch within a long-running process (e.g.
+// serve mode) re-resolves rather than reusing a stale identity.
+func auditIdentity() (audit.Actor, string) {
+	profile := auth.RuntimeProfile()
+
+	auditIDMu.Lock()
+	defer auditIDMu.Unlock()
+	if identityLoaded && profile == cachedProfile {
+		return cachedActor, cachedAgentID
+	}
+
+	configDir := defaultConfigDir()
+	var actor audit.Actor
+	if td, err := loadTokenForProfile(configDir, profile); err == nil && td != nil {
+		actor = audit.Actor{
+			UserID:   td.UserID,
+			Name:     td.UserName,
+			CorpID:   td.CorpID,
+			CorpName: td.CorpName,
+		}
+	} else if err != nil {
+		auditReport("resolve actor for profile %q failed: %v", profile, err)
+	}
+
+	agentID := ""
+	if id := auth.Load(configDir); id != nil {
+		agentID = id.AgentID
+	}
+
+	cachedActor, cachedAgentID, cachedProfile, identityLoaded = actor, agentID, profile, true
+	return actor, agentID
+}
+
+func emitAudit(sink audit.Sink, execID string, invokeStart time.Time, invocation executor.Invocation, endpoint string, retErr error, cliVersion string) {
+	if sink == nil {
+		return
+	}
+	if _, ok := sink.(audit.NopSink); ok {
+		return
+	}
+
+	actor, agentID := auditIdentity()
+
+	result := "success"
+	var errCat, errReason string
+	if retErr != nil {
+		result = "error"
+		errCat, errReason = classifyAuditError(retErr)
+	}
+
+	paramsSummary := logging.SanitizeArguments(invocation.Params, 1024)
+
+	evt := &audit.Event{
+		Timestamp:     invokeStart,
+		ExecutionID:   execID,
+		AgentID:       agentID,
+		Actor:         actor,
+		Product:       invocation.CanonicalProduct,
+		Command:       invocation.Tool,
+		Endpoint:      transport.RedactURL(endpoint),
+		ParamsSummary: paramsSummary,
+		Result:        result,
+		ErrCategory:   errCat,
+		ErrReason:     errReason,
+		DurationMs:    time.Since(invokeStart).Milliseconds(),
+		CLIVersion:    cliVersion,
+		OS:            runtime.GOOS,
+		Arch:          runtime.GOARCH,
+	}
+
+	if err := sink.Emit(evt); err != nil {
+		auditReport("emit event failed (exec %s): %v", execID, err)
+	}
+}
+
+func classifyAuditError(err error) (category, reason string) {
+	if err == nil {
+		return "", ""
+	}
+	var typed *apperrors.Error
+	if errors.As(err, &typed) {
+		return string(typed.Category), typed.Reason
+	}
+	return "unknown", err.Error()
+}

internal/app/audit_runtime_test.go

@@ -0,0 +1,131 @@
+package app
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/audit"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/auth"
+)
+
+// TestAuditIdentityReresolvesOnProfileSwitch guards the reviewer's finding that a
+// long-running process (e.g. serve mode) must attribute events to the ACTIVE
+// runtime profile rather than reusing a process-global first Actor. It also
+// asserts the per-profile cache avoids redundant token loads within one profile.
+func TestAuditIdentityReresolvesOnProfileSwitch(t *testing.T) {
+	prevLoader := loadTokenForProfile
+	prevProfile := auth.RuntimeProfile()
+	t.Cleanup(func() {
+		loadTokenForProfile = prevLoader
+		auth.SetRuntimeProfile(prevProfile)
+		resetAuditIdentityCache()
+	})
+	resetAuditIdentityCache()
+
+	var mu sync.Mutex
+	calls := map[string]int{}
+	loadTokenForProfile = func(_ /*configDir*/, profile string) (*auth.TokenData, error) {
+		mu.Lock()
+		calls[profile]++
+		mu.Unlock()
+		switch profile {
+		case "orgA":
+			return &auth.TokenData{UserID: "ua", UserName: "Alice", CorpID: "ca", CorpName: "CorpA"}, nil
+		case "orgB":
+			return &auth.TokenData{UserID: "ub", UserName: "Bob", CorpID: "cb", CorpName: "CorpB"}, nil
+		default:
+			return nil, nil
+		}
+	}
+
+	auth.SetRuntimeProfile("orgA")
+	if actor, _ := auditIdentity(); actor.UserID != "ua" || actor.CorpName != "CorpA" {
+		t.Fatalf("orgA: got %+v, want Alice/CorpA", actor)
+	}
+	// Second call under the same profile must hit the cache (no extra load).
+	if actor, _ := auditIdentity(); actor.UserID != "ua" {
+		t.Fatalf("orgA cached: got %+v", actor)
+	}
+
+	auth.SetRuntimeProfile("orgB")
+	if actor, _ := auditIdentity(); actor.UserID != "ub" || actor.CorpName != "CorpB" {
+		t.Fatalf("orgB: got %+v, want Bob/CorpB (stale Actor reused?)", actor)
+	}
+
+	mu.Lock()
+	defer mu.Unlock()
+	if calls["orgA"] != 1 {
+		t.Fatalf("orgA loaded %d times, want 1 (cache miss?)", calls["orgA"])
+	}
+	if calls["orgB"] != 1 {
+		t.Fatalf("orgB loaded %d times, want 1", calls["orgB"])
+	}
+}
+
+func resetAuditIdentityCache() {
+	auditIDMu.Lock()
+	defer auditIDMu.Unlock()
+	cachedActor = audit.Actor{}
+	cachedAgentID = ""
+	cachedProfile = ""
+	identityLoaded = false
+}
+
+// TestCloseAuditSinkDrainsOnErrorPath guards the reviewer's V5 finding: when a
+// command's RunE returns an error, Cobra skips PersistentPostRunE, so the audit
+// drain must instead happen through the unconditional defer in Execute that calls
+// CloseAuditSink. This test wires a real forwarder-backed sink into the shared
+// slot and asserts CloseAuditSink flushes the queued forward exactly as the
+// error-path defer would, and that a second call is a harmless no-op.
+func TestCloseAuditSinkDrainsOnErrorPath(t *testing.T) {
+	var delivered int64
+	var releaseOnce sync.Once
+	release := make(chan struct{})
+	releaseFn := func() { releaseOnce.Do(func() { close(release) }) }
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		<-release // hold the request until the drain awaits it
+		atomic.AddInt64(&delivered, 1)
+		w.WriteHeader(http.StatusOK)
+	}))
+	defer srv.Close()
+	defer releaseFn() // LIFO: unblock any in-flight handler before srv.Close()
+
+	writer, err := audit.NewDateRotatingWriter(t.TempDir(), 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	fwd := audit.NewHTTPForwarder(srv.URL, "", audit.RedactNone, nil)
+	sink := audit.NewFileSink(writer, audit.NewChain(""), fwd)
+
+	prevSink := sharedAuditSink
+	t.Cleanup(func() {
+		sharedAuditSink = prevSink
+		auditCloseOnce = sync.Once{}
+	})
+	sharedAuditSink = sink
+	auditCloseOnce = sync.Once{}
+
+	if err := sink.Emit(&audit.Event{Timestamp: time.Unix(0, 0), Product: "calendar", Command: "event_list", Result: "error"}); err != nil {
+		t.Fatalf("emit: %v", err)
+	}
+	if got := atomic.LoadInt64(&delivered); got != 0 {
+		t.Fatalf("forward delivered before drain: %d", got)
+	}
+
+	// Let the held request complete, then drain via the same entry point the
+	// error-path defer uses. CloseAuditSink blocks until the forward goroutine
+	// observes the HTTP response, so the counter is settled when it returns.
+	releaseFn()
+	CloseAuditSink()
+
+	if got := atomic.LoadInt64(&delivered); got != 1 {
+		t.Fatalf("forward not drained on error path: delivered=%d, want 1", got)
+	}
+
+	// Idempotent: the success-path PersistentPostRunE and the defer both call it.
+	CloseAuditSink()
+}