chore(skills): drop dead SAFETY_PREAMBLE_INJECT marker

No injection code ever existed in this repository, so the marker was
inert in all 22 multi SKILL.md files. Remove it and correct the skill
authoring docs to describe safety rules as plain prose.

Co-authored-by: Cursor <cursoragent@cursor.com>

.github/PULL_REQUEST_TEMPLATE.md

@@ -5,6 +5,13 @@
 
 ## Verification
 
+For an exact in-place `CHANGELOG.md`-only pull request, the full-suite checks
+may be marked `N/A`, but the targeted CHANGELOG check is required. For every
+other pull request, mark the targeted check `N/A` and complete the applicable
+full-suite checks.
+
+- [ ] Exact `CHANGELOG.md`-only check (otherwise `N/A`):
+  `./scripts/policy/check-changelog-pr.sh --fast-path "$(git merge-base HEAD origin/main)" HEAD`
 - [ ] `make build`
 - [ ] `make lint`
 - [ ] `make test`

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

@@ -1,8 +1,11 @@
-name: AI Behavior Check
+name: Code Admission — AI Behavior
 
 on:
   pull_request_target:
     types: [opened, synchronize, reopened, labeled, unlabeled]
+  push:
+    branches:
+      - main
 
 permissions:
   contents: read
@@ -11,7 +14,7 @@ permissions:
 
 jobs:
   ai-behavior-check:
-    name: AI Behavior Policy Evaluator
+    name: AI Behavior
     runs-on: ubuntu-latest
     timeout-minutes: 5
     steps:
@@ -21,75 +24,108 @@ jobs:
         uses: actions/github-script@v7
         with:
           script: |
-            const sha = context.payload.pull_request.head.sha;
+            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 Check',
+                context: 'AI Behavior',
                 description,
               });
 
             await setStatus('pending', 'Evaluating AI-generated PR boundaries');
 
-            const labels = context.payload.pull_request.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.');
+            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;
             }
 
-            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,
-            });
+            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;
+              }
 
-            const maxChangedFiles = 30;
-            if (files.length > maxChangedFiles) {
               await setStatus(
-                'failure',
-                `Changes ${files.length} files; limit is ${maxChangedFiles}`
+                'success',
+                `Passed with ${files.length} changed files (limit ${maxChangedFiles})`
               );
-              core.setFailed(
-                `AI-generated PR changes ${files.length} files; limit is ${maxChangedFiles}.`
+              core.notice(
+                `AI behavior check passed (${files.length} changed files; limit ${maxChangedFiles}).`
               );
-              return;
+            } catch (error) {
+              await setStatus('error', 'Could not evaluate the exact pull request revision');
+              throw error;
             }
-
-            const isProtectedPath = (filename) =>
-              typeof filename === 'string' &&
-              (
-                filename.startsWith('.github/workflows/') ||
-                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}).`
-            );

.github/workflows/ci.yml

@@ -18,48 +18,138 @@ jobs:
     name: Lint
     runs-on: ubuntu-latest
     timeout-minutes: 10
+    permissions:
+      contents: read
+      pull-requests: read
+    outputs:
+      changelog_only: ${{ steps.classify.outputs.changelog_only }}
+      changelog_changed: ${{ steps.classify.outputs.changelog_changed }}
+      platform_sensitive: ${{ steps.classify.outputs.platform_sensitive }}
     steps:
+      - name: Classify pull request scope
+        id: classify
+        uses: actions/github-script@v7
+        with:
+          script: |
+            let changelogOnly = false;
+            let changelogChanged = false;
+            let platformSensitive = context.eventName === 'push';
+            let files = [];
+
+            if (context.eventName === 'pull_request') {
+              const expectedHead = context.payload.pull_request.head.sha;
+              const expectedBase = context.payload.pull_request.base.sha;
+              const assertCurrentRevision = 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 assertCurrentRevision('pre-classification');
+              files = await github.paginate(github.rest.pulls.listFiles, {
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                pull_number: context.issue.number,
+                per_page: 100,
+              });
+              const after = await assertCurrentRevision('post-classification');
+              if (
+                before.changed_files !== files.length ||
+                after.changed_files !== files.length
+              ) {
+                throw new Error(
+                  `Pull request file list is incomplete: API reports ` +
+                  `${after.changed_files} changed files, pagination returned ${files.length}`
+                );
+              }
+
+              changelogOnly =
+                files.length === 1 &&
+                files[0].filename === 'CHANGELOG.md' &&
+                files[0].status === 'modified' &&
+                !files[0].previous_filename;
+              changelogChanged = files.some(
+                ({ filename, previous_filename }) =>
+                  filename === 'CHANGELOG.md' ||
+                  previous_filename === 'CHANGELOG.md'
+              );
+
+              const isPlatformSensitive = (filename) =>
+                typeof filename === 'string' &&
+                (
+                  filename.startsWith('internal/auth/') ||
+                  filename.startsWith('internal/keychain/') ||
+                  /_(darwin|windows|linux|unix)\.go$/.test(filename) ||
+                  filename.startsWith('scripts/release/') ||
+                  filename.startsWith('scripts/install') ||
+                  filename.startsWith('Formula/') ||
+                  filename.startsWith('build/npm/') ||
+                  filename === '.goreleaser.yaml' ||
+                  filename === '.github/workflows/release.yml'
+                );
+              platformSensitive = files.some(
+                ({ filename, previous_filename }) =>
+                  isPlatformSensitive(filename) ||
+                  isPlatformSensitive(previous_filename)
+              );
+            }
+
+            core.setOutput('changelog_only', String(changelogOnly));
+            core.setOutput('changelog_changed', String(changelogChanged));
+            core.setOutput('platform_sensitive', String(platformSensitive));
+            await core.summary
+              .addHeading('Code Admission scope')
+              .addRaw(`- Event: \`${context.eventName}\`\n`)
+              .addRaw(`- Exact modified CHANGELOG only: \`${changelogOnly}\`\n`)
+              .addRaw(`- CHANGELOG touched: \`${changelogChanged}\`\n`)
+              .addRaw(`- Native-platform risk paths touched: \`${platformSensitive}\`\n`)
+              .addRaw(`- Changed files: \`${files.length}\`\n`)
+              .write();
+
+      - name: Record CHANGELOG-only fast path
+        if: steps.classify.outputs.changelog_only == 'true'
+        run: echo "Lint is satisfied by the trusted CHANGELOG-only Policy path." >> "$GITHUB_STEP_SUMMARY"
+
       - name: Check out repository
+        if: steps.classify.outputs.changelog_only != 'true'
         uses: actions/checkout@v4
 
       - name: Set up Go
+        if: steps.classify.outputs.changelog_only != 'true'
         uses: actions/setup-go@v5
         with:
           go-version-file: go.mod
 
+      - name: Verify Test Package Plan
+        if: steps.classify.outputs.changelog_only != 'true'
+        run: make test-plan
+
       - name: Format Check
-        run: |
-          unformatted="$(find cmd internal test scripts/policy -name '*.go' -print0 | xargs -0r gofmt -l)"
-          test -z "$unformatted" || (printf '%s\n' "$unformatted" && exit 1)
+        if: steps.classify.outputs.changelog_only != 'true'
+        run: make format-check
 
       - name: Go Vet
+        if: steps.classify.outputs.changelog_only != 'true'
         run: go vet ./...
 
-      # golangci-lint temporarily disabled: v1.64.8 built with Go 1.24 is incompatible with Go 1.25
-      # - name: golangci-lint
-      #   uses: golangci/golangci-lint-action@v6
-      #   with:
-      #     version: v1.64.8
-      #     args: ./...
-
-  actionlint:
-    name: Actionlint
-    runs-on: ubuntu-latest
-    timeout-minutes: 10
-    steps:
-      - name: Check out repository
-        uses: actions/checkout@v4
-
-      - name: Set up Go
-        uses: actions/setup-go@v5
-        with:
-          go-version-file: go.mod
-
       - name: Check GitHub Actions workflows
+        if: steps.classify.outputs.changelog_only != 'true'
         run: go run github.com/rhysd/actionlint/cmd/actionlint@v1.7.12
 
   test-race:
     name: "Test (race: ${{ matrix.shard }})"
+    needs: lint
+    if: ${{ needs.lint.outputs.changelog_only != 'true' }}
     runs-on: ubuntu-latest
     timeout-minutes: 15
     strategy:
@@ -79,45 +169,29 @@ jobs:
         with:
           go-version-file: go.mod
 
-      - name: Install archive tooling
-        run: sudo apt-get update && sudo apt-get install -y zip unzip
-
       - name: Build
         if: ${{ matrix.shard == 'remaining' }}
         run: make build
 
       - name: Test shard with Race Detection
+        shell: bash
         env:
+          DWS_PACKAGE_VERSION: 0.0.0-test
           TEST_SHARD: ${{ matrix.shard }}
         run: |
-          case "$TEST_SHARD" in
-            app)
-              packages=(./internal/app/...)
-              ;;
-            generators)
-              packages=(./internal/generator/...)
-              ;;
-            helpers)
-              packages=(./internal/helpers/...)
-              ;;
-            remaining)
-              mapfile -t packages < <(
-                go list ./cmd/... ./internal/... |
-                  grep -Ev '/internal/(app|generator|helpers)(/|$)'
-              )
-              ;;
-            *)
-              printf 'unknown test shard: %s\n' "$TEST_SHARD" >&2
-              exit 1
-              ;;
-          esac
+          set -euo pipefail
+          package_output="$(./scripts/ci/test-packages.sh list "$TEST_SHARD")"
+          test -n "$package_output"
+          mapfile -t packages <<< "$package_output"
           test "${#packages[@]}" -gt 0
           go test -v -race -count=1 -timeout=10m "${packages[@]}"
 
   test-release-scripts:
     name: Test (release scripts)
+    needs: lint
+    if: ${{ needs.lint.outputs.changelog_only != 'true' }}
     runs-on: ubuntu-latest
-    timeout-minutes: 10
+    timeout-minutes: 15
     steps:
       - name: Check out repository
         uses: actions/checkout@v4
@@ -131,27 +205,96 @@ jobs:
         run: sudo apt-get update && sudo apt-get install -y zip unzip
 
       - name: Test release scripts
-        run: go test -v -count=1 -timeout=5m ./test/scripts
+        shell: bash
+        env:
+          DWS_PACKAGE_VERSION: 0.0.0-test
+        run: |
+          set -euo pipefail
+          package_output="$(./scripts/ci/test-packages.sh list release-scripts)"
+          test -n "$package_output"
+          mapfile -t packages <<< "$package_output"
+          test "${#packages[@]}" -gt 0
+          go test -v -count=1 -timeout=10m "${packages[@]}"
+
+  test-cross-platform:
+    name: Test (cross-platform compile)
+    needs: lint
+    if: ${{ needs.lint.outputs.changelog_only != 'true' }}
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v4
+
+      - name: Set up Go
+        uses: actions/setup-go@v5
+        with:
+          go-version-file: go.mod
+
+      - name: Compile supported operating systems
+        shell: bash
+        run: |
+          set -eu
+          for target in darwin/amd64 darwin/arm64 windows/amd64 windows/arm64; do
+            goos="${target%/*}"
+            goarch="${target#*/}"
+            output="$RUNNER_TEMP/dws-${goos}-${goarch}"
+            if [ "$goos" = windows ]; then
+              output="${output}.exe"
+            fi
+            printf 'compile %s/%s\n' "$goos" "$goarch"
+            CGO_ENABLED=0 GOOS="$goos" GOARCH="$goarch" \
+              go build -o "$output" ./cmd
+          done
 
   test:
     name: Test
     needs:
+      - lint
       - test-race
       - test-release-scripts
-    if: ${{ always() }}
+      - test-cross-platform
+      - test-darwin
+      - test-windows
+    if: ${{ always() && needs.lint.result == 'success' }}
     runs-on: ubuntu-latest
     timeout-minutes: 5
     permissions: {}
     steps:
       - name: Verify test shards
         env:
+          CHANGELOG_ONLY: ${{ needs.lint.outputs.changelog_only }}
+          PLATFORM_SENSITIVE: ${{ needs.lint.outputs.platform_sensitive }}
           RACE_RESULT: ${{ needs.test-race.result }}
           RELEASE_SCRIPTS_RESULT: ${{ needs.test-release-scripts.result }}
+          CROSS_PLATFORM_RESULT: ${{ needs.test-cross-platform.result }}
+          DARWIN_RESULT: ${{ needs.test-darwin.result }}
+          WINDOWS_RESULT: ${{ needs.test-windows.result }}
         run: |
           failed=0
+          if [ "$CHANGELOG_ONLY" = true ]; then
+            for shard in \
+              "race shards:$RACE_RESULT" \
+              "release scripts:$RELEASE_SCRIPTS_RESULT" \
+              "cross-platform compile:$CROSS_PLATFORM_RESULT" \
+              "macOS native:$DARWIN_RESULT" \
+              "Windows native:$WINDOWS_RESULT"
+            do
+              name="${shard%%:*}"
+              result="${shard#*:}"
+              printf '%s: %s\n' "$name" "$result"
+              if [ "$result" != skipped ]; then
+                failed=1
+              fi
+            done
+            test "$failed" -eq 0
+            exit
+          fi
+
           for shard in \
             "race shards:$RACE_RESULT" \
-            "release scripts:$RELEASE_SCRIPTS_RESULT"
+            "release scripts:$RELEASE_SCRIPTS_RESULT" \
+            "cross-platform compile:$CROSS_PLATFORM_RESULT"
           do
             name="${shard%%:*}"
             result="${shard#*:}"
@@ -160,10 +303,28 @@ jobs:
               failed=1
             fi
           done
+
+          native_expected=skipped
+          if [ "$PLATFORM_SENSITIVE" = true ]; then
+            native_expected=success
+          fi
+          for native in \
+            "macOS native:$DARWIN_RESULT" \
+            "Windows native:$WINDOWS_RESULT"
+          do
+            name="${native%%:*}"
+            result="${native#*:}"
+            printf '%s: %s\n' "$name" "$result"
+            if [ "$result" != "$native_expected" ]; then
+              failed=1
+            fi
+          done
           test "$failed" -eq 0
 
   test-darwin:
     name: Test (macOS auth/keychain)
+    needs: lint
+    if: ${{ needs.lint.outputs.changelog_only != 'true' && needs.lint.outputs.platform_sensitive == 'true' }}
     runs-on: macos-latest
     timeout-minutes: 15
     steps:
@@ -180,6 +341,8 @@ jobs:
 
   test-windows:
     name: Test (Windows)
+    needs: lint
+    if: ${{ needs.lint.outputs.changelog_only != 'true' && needs.lint.outputs.platform_sensitive == 'true' }}
     runs-on: windows-latest
     timeout-minutes: 15
     steps:
@@ -202,6 +365,8 @@ jobs:
 
   coverage-darwin:
     name: Coverage (macOS)
+    needs: lint
+    if: ${{ needs.lint.outputs.changelog_only != 'true' && needs.lint.outputs.platform_sensitive == 'true' }}
     runs-on: macos-latest
     timeout-minutes: 20
     steps:
@@ -244,6 +409,8 @@ jobs:
 
   coverage-windows:
     name: Coverage (Windows)
+    needs: lint
+    if: ${{ needs.lint.outputs.changelog_only != 'true' && needs.lint.outputs.platform_sensitive == 'true' }}
     runs-on: windows-latest
     timeout-minutes: 20
     steps:
@@ -288,6 +455,8 @@ jobs:
 
   coverage-current:
     name: Coverage (current)
+    needs: lint
+    if: ${{ needs.lint.outputs.changelog_only != 'true' }}
     runs-on: ubuntu-latest
     timeout-minutes: 20
     steps:
@@ -321,6 +490,8 @@ jobs:
 
   coverage-supporting:
     name: Coverage (supporting)
+    needs: lint
+    if: ${{ needs.lint.outputs.changelog_only != 'true' }}
     runs-on: ubuntu-latest
     timeout-minutes: 10
     steps:
@@ -358,6 +529,8 @@ jobs:
 
   coverage-baseline:
     name: Coverage (baseline)
+    needs: lint
+    if: ${{ needs.lint.outputs.changelog_only != 'true' }}
     runs-on: ubuntu-latest
     timeout-minutes: 20
     steps:
@@ -421,20 +594,35 @@ jobs:
   coverage:
     name: Coverage
     needs:
+      - lint
       - coverage-current
       - coverage-supporting
       - coverage-baseline
-    if: ${{ always() }}
+      - coverage-darwin
+      - coverage-windows
+    if: ${{ always() && needs.lint.result == 'success' }}
     runs-on: ubuntu-latest
     timeout-minutes: 10
     steps:
       - name: Verify coverage profile jobs
         env:
+          CHANGELOG_ONLY: ${{ needs.lint.outputs.changelog_only }}
+          PLATFORM_SENSITIVE: ${{ needs.lint.outputs.platform_sensitive }}
           CURRENT_RESULT: ${{ needs.coverage-current.result }}
           SUPPORTING_RESULT: ${{ needs.coverage-supporting.result }}
           BASELINE_RESULT: ${{ needs.coverage-baseline.result }}
+          DARWIN_RESULT: ${{ needs.coverage-darwin.result }}
+          WINDOWS_RESULT: ${{ needs.coverage-windows.result }}
         run: |
           failed=0
+          expected=success
+          native_expected=skipped
+          if [ "$CHANGELOG_ONLY" = true ]; then
+            expected=skipped
+          elif [ "$PLATFORM_SENSITIVE" = true ]; then
+            native_expected=success
+          fi
+
           for profile in \
             "current:$CURRENT_RESULT" \
             "supporting:$SUPPORTING_RESULT" \
@@ -443,24 +631,42 @@ jobs:
             name="${profile%%:*}"
             result="${profile#*:}"
             printf '%s: %s\n' "$name" "$result"
-            if [ "$result" != "success" ]; then
+            if [ "$result" != "$expected" ]; then
+              failed=1
+            fi
+          done
+
+          for native in \
+            "macOS native:$DARWIN_RESULT" \
+            "Windows native:$WINDOWS_RESULT"
+          do
+            name="${native%%:*}"
+            result="${native#*:}"
+            printf '%s: %s\n' "$name" "$result"
+            if [ "$CHANGELOG_ONLY" = true ]; then
+              native_expected=skipped
+            fi
+            if [ "$result" != "$native_expected" ]; then
               failed=1
             fi
           done
           test "$failed" -eq 0
 
       - name: Check out repository
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/checkout@v4
         with:
           fetch-depth: 0
           ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
 
       - name: Set up Go
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/setup-go@v5
         with:
           go-version-file: go.mod
 
       - name: Resolve authoritative coverage base
+        if: needs.lint.outputs.changelog_only != 'true'
         env:
           PR_HEAD_SHA: ${{ github.event.pull_request.head.sha }}
           PR_BASE_SHA: ${{ github.event.pull_request.base.sha }}
@@ -478,33 +684,40 @@ jobs:
           echo "COVERAGE_BASE_REF=$base_ref" >> "$GITHUB_ENV"
 
       - name: Download current coverage profile
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/download-artifact@v4
         with:
           name: coverage-current-profile
           path: .
 
       - name: Download supporting coverage profiles
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/download-artifact@v4
         with:
           name: coverage-supporting-profiles
           path: .
 
       - name: Download baseline coverage profile
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/download-artifact@v4
         with:
           name: coverage-baseline-profile
           path: .
 
       - name: Enforce coverage gate
+        if: needs.lint.outputs.changelog_only != 'true'
         env:
-          COVERAGE_TARGET: "80"
+          COVERAGE_TARGET: "100"
           COVERAGE_ENFORCE_OVERALL: "false"
-        run: COVERAGE_ADDITIONAL_PROFILE=coverage-shortcut.txt make coverage-gate BASE_REF="$COVERAGE_BASE_REF"
+          COVERAGE_OVERALL_TOLERANCE: "0"
+        run: COVERAGE_ADDITIONAL_DIFF_PROFILE=coverage-shortcut.txt make coverage-gate BASE_REF="$COVERAGE_BASE_REF"
 
       - name: Generate coverage report
+        if: needs.lint.outputs.changelog_only != 'true'
         run: go tool cover -html=coverage.txt -o coverage.html
 
       - name: Upload coverage artifact
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/upload-artifact@v4
         with:
           name: coverage-report
@@ -516,44 +729,114 @@ jobs:
             coverage.html
 
   policy:
-    name: Policy Check
+    name: Policy
+    needs: lint
     runs-on: ubuntu-latest
     timeout-minutes: 10
     steps:
       - name: Check out repository
         uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
 
       - name: Set up Go
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/setup-go@v5
         with:
           go-version-file: go.mod
 
+      - name: Verify pull request merge revision
+        if: github.event_name == 'pull_request'
+        env:
+          PR_BASE_SHA: ${{ github.event.pull_request.base.sha }}
+          PR_HEAD_SHA: ${{ github.event.pull_request.head.sha }}
+        run: |
+          set -eu
+          test "$(git rev-parse HEAD^1)" = "$PR_BASE_SHA" || {
+            echo "checked-out merge first parent does not match event base" >&2
+            exit 1
+          }
+          test "$(git rev-parse HEAD^2)" = "$PR_HEAD_SHA" || {
+            echo "checked-out merge second parent does not match event head" >&2
+            exit 1
+          }
+
+      - name: Validate changed CHANGELOG content
+        if: github.event_name == 'pull_request'
+        env:
+          CLASSIFIED_CHANGELOG_CHANGED: ${{ needs.lint.outputs.changelog_changed }}
+          CHANGELOG_ONLY: ${{ needs.lint.outputs.changelog_only }}
+          PR_BASE_SHA: ${{ github.event.pull_request.base.sha }}
+        run: |
+          set -eu
+          merge_changelog_changed=false
+          if git diff --no-ext-diff --find-renames --name-status \
+            "$PR_BASE_SHA" HEAD |
+            awk -F '\t' '
+              {
+                for (field = 2; field <= NF; field++) {
+                  if ($field == "CHANGELOG.md") found = 1
+                }
+              }
+              END { exit !found }
+            '
+          then
+            merge_changelog_changed=true
+          fi
+          test "$merge_changelog_changed" = "$CLASSIFIED_CHANGELOG_CHANGED" || {
+            echo "Files API and synthetic merge tree disagree on CHANGELOG scope" >&2
+            exit 1
+          }
+          if [ "$merge_changelog_changed" != true ]; then
+            exit 0
+          fi
+
+          mode=--content-only
+          if [ "$CHANGELOG_ONLY" = true ]; then
+            mode=--fast-path
+          fi
+          ./scripts/policy/check-changelog-pr.sh \
+            "$mode" "$PR_BASE_SHA" HEAD
+
+      - name: Record CHANGELOG-only fast path
+        if: needs.lint.outputs.changelog_only == 'true'
+        run: |
+          echo "Only the base-equivalent CHANGELOG validator ran; full Policy resumes on main." \
+            >> "$GITHUB_STEP_SUMMARY"
+
       - name: Build
+        if: needs.lint.outputs.changelog_only != 'true'
         run: make build
 
       - name: Policy
+        if: needs.lint.outputs.changelog_only != 'true'
         run: make policy
 
   interface-integrity:
     name: Interface Integrity
+    needs: lint
     runs-on: ubuntu-latest
     timeout-minutes: 15
     steps:
       - name: Check out repository
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/checkout@v4
         with:
           fetch-depth: 0
           ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
 
       - name: Set up Go
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/setup-go@v5
         with:
           go-version-file: go.mod
 
       - name: Build
+        if: needs.lint.outputs.changelog_only != 'true'
         run: make build
 
       - name: Resolve authoritative compatibility merge-base
+        if: needs.lint.outputs.changelog_only != 'true'
         env:
           PR_HEAD_SHA: ${{ github.event.pull_request.head.sha }}
           PR_BASE_SHA: ${{ github.event.pull_request.base.sha }}
@@ -578,6 +861,7 @@ jobs:
           echo "COMPATIBILITY_STABLE_REF=$stable_ref" >> "$GITHUB_ENV"
 
       - name: Check historical commands and help compatibility
+        if: needs.lint.outputs.changelog_only != 'true'
         run: |
           make authoritative-interface-integrity \
             BASE_REF="$COMPATIBILITY_BASE_REF"
@@ -587,143 +871,89 @@ jobs:
           fi
 
       - name: Check complete Schema compatibility
+        if: needs.lint.outputs.changelog_only != 'true'
         run: make schema-compatibility BASE_REF="$COMPATIBILITY_BASE_REF"
 
       - name: Check skill command references
+        if: needs.lint.outputs.changelog_only != 'true'
         run: make skill-command-integrity
 
+      - name: Record CHANGELOG-only fast path
+        if: needs.lint.outputs.changelog_only == 'true'
+        run: echo "Interface Integrity is unaffected by an exact CHANGELOG-only diff." >> "$GITHUB_STEP_SUMMARY"
+
   cli-smoke:
     name: CLI Smoke
+    needs: lint
     runs-on: ubuntu-latest
     timeout-minutes: 10
     steps:
       - name: Check out repository
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/checkout@v4
 
       - name: Set up Go
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/setup-go@v5
         with:
           go-version-file: go.mod
 
       - name: Build
+        if: needs.lint.outputs.changelog_only != 'true'
         run: make build
 
       - name: Check public top-level commands
+        if: needs.lint.outputs.changelog_only != 'true'
         run: make cli-smoke
 
+      - name: Record CHANGELOG-only fast path
+        if: needs.lint.outputs.changelog_only == 'true'
+        run: echo "CLI Smoke is unaffected by an exact CHANGELOG-only diff." >> "$GITHUB_STEP_SUMMARY"
+
   mock-mcp-smoke:
-    name: Mock MCP Smoke
+    name: Mock MCP
+    needs: lint
     runs-on: ubuntu-latest
     timeout-minutes: 10
     steps:
       - name: Check out repository
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/checkout@v4
 
       - name: Set up Go
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/setup-go@v5
         with:
           go-version-file: go.mod
 
       - name: Check HTTP and stdio MCP transport
+        if: needs.lint.outputs.changelog_only != 'true'
         run: make mock-mcp-smoke
 
+      - name: Record CHANGELOG-only fast path
+        if: needs.lint.outputs.changelog_only == 'true'
+        run: echo "Mock MCP is unaffected by an exact CHANGELOG-only diff." >> "$GITHUB_STEP_SUMMARY"
+
   edition-tests:
-    name: Edition Contract Tests
+    name: Edition
+    needs: lint
     runs-on: ubuntu-latest
     timeout-minutes: 10
     steps:
       - name: Check out repository
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/checkout@v4
 
       - name: Set up Go
+        if: needs.lint.outputs.changelog_only != 'true'
         uses: actions/setup-go@v5
         with:
           go-version-file: go.mod
 
       - name: Run edition contract tests
+        if: needs.lint.outputs.changelog_only != 'true'
         run: go test -v -count=1 ./pkg/editiontest/...
 
-  ci-gate:
-    name: CI Gate
-    needs:
-      - lint
-      - actionlint
-      - test
-      - test-darwin
-      - test-windows
-      - coverage
-      - coverage-darwin
-      - coverage-windows
-      - policy
-      - interface-integrity
-      - cli-smoke
-      - mock-mcp-smoke
-      - edition-tests
-    if: ${{ always() }}
-    runs-on: ubuntu-latest
-    timeout-minutes: 5
-    permissions: {}
-    steps:
-      - name: Verify required checks
-        env:
-          LINT_RESULT: ${{ needs.lint.result }}
-          ACTIONLINT_RESULT: ${{ needs.actionlint.result }}
-          TEST_RESULT: ${{ needs.test.result }}
-          TEST_DARWIN_RESULT: ${{ needs.test-darwin.result }}
-          TEST_WINDOWS_RESULT: ${{ needs.test-windows.result }}
-          COVERAGE_RESULT: ${{ needs.coverage.result }}
-          COVERAGE_DARWIN_RESULT: ${{ needs.coverage-darwin.result }}
-          COVERAGE_WINDOWS_RESULT: ${{ needs.coverage-windows.result }}
-          POLICY_RESULT: ${{ needs.policy.result }}
-          INTERFACE_INTEGRITY_RESULT: ${{ needs.interface-integrity.result }}
-          CLI_SMOKE_RESULT: ${{ needs.cli-smoke.result }}
-          MOCK_MCP_SMOKE_RESULT: ${{ needs.mock-mcp-smoke.result }}
-          EDITION_TESTS_RESULT: ${{ needs.edition-tests.result }}
-        run: |
-          failed=0
-          for check in \
-            "Lint:$LINT_RESULT" \
-            "Actionlint:$ACTIONLINT_RESULT" \
-            "Test:$TEST_RESULT" \
-            "Test (macOS auth/keychain):$TEST_DARWIN_RESULT" \
-            "Test (Windows):$TEST_WINDOWS_RESULT" \
-            "Coverage:$COVERAGE_RESULT" \
-            "Coverage (macOS):$COVERAGE_DARWIN_RESULT" \
-            "Coverage (Windows):$COVERAGE_WINDOWS_RESULT" \
-            "Policy Check:$POLICY_RESULT" \
-            "Interface Integrity:$INTERFACE_INTEGRITY_RESULT" \
-            "CLI Smoke:$CLI_SMOKE_RESULT" \
-            "Mock MCP Smoke:$MOCK_MCP_SMOKE_RESULT" \
-            "Edition Contract Tests:$EDITION_TESTS_RESULT"
-          do
-            name="${check%%:*}"
-            result="${check#*:}"
-            printf '%s: %s\n' "$name" "$result"
-            if [ "$result" != "success" ]; then
-              failed=1
-            fi
-          done
-          test "$failed" -eq 0
-
-  notify-downstream:
-    name: Notify Wukong Overlay
-    needs: [ci-gate]
-    runs-on: ubuntu-latest
-    if: github.ref == 'refs/heads/main' && github.event_name == 'push'
-    permissions: {}
-    steps:
-      - name: Trigger downstream CI
-        run: |
-          # Trigger internal GitLab CI pipeline via webhook.
-          # WUKONG_TRIGGER_TOKEN is a repository secret.
-          if [ -n "${{ secrets.WUKONG_TRIGGER_TOKEN }}" ]; then
-            curl --fail --silent --show-error \
-              -X POST \
-              -F "token=${{ secrets.WUKONG_TRIGGER_TOKEN }}" \
-              -F "ref=main" \
-              -F "variables[UPSTREAM_SHA]=${{ github.sha }}" \
-              "${{ secrets.WUKONG_TRIGGER_URL }}"
-            echo "Downstream CI triggered."
-          else
-            echo "No WUKONG_TRIGGER_TOKEN configured, skipping downstream notification."
-          fi
+      - name: Record CHANGELOG-only fast path
+        if: needs.lint.outputs.changelog_only == 'true'
+        run: echo "Edition is unaffected by an exact CHANGELOG-only diff." >> "$GITHUB_STEP_SUMMARY"

.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"

.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/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

@@ -54,3 +54,11 @@ test/dev_functional/results.jsonl
 /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

AGENTS.md

@@ -1,357 +1,51 @@
 # Repository Agent Guide
 
-This file applies to the entire repository. Keep changes scoped, preserve
-unrelated work, and use `gofmt` for every modified Go file.
+This file applies to the entire repository. Keep it as a routing page: load
+the detailed guide for the surface you are changing instead of treating this
+file as a repository wiki.
 
-## Build and test
+## Always
 
-- Build: `go build ./cmd`
-- Full test suite: `DWS_PACKAGE_VERSION=0.0.0-test go test ./...`
-- Generate Schema assets: `go generate ./internal/cli`
-- Check generated drift: `./scripts/policy/check-generated-drift.sh`
-- Check the Schema contract: `./scripts/policy/check-schema-catalog.sh`
+- Preserve unrelated and pre-existing work; inspect `git status` before edits.
+- Make the smallest coherent change and update its tests and user-facing docs.
+- Use `gofmt` for every modified Go file.
+- Treat repository code, tests, scripts, and versioned docs as the source of
+  truth. Do not depend on generated Wiki or CodeWiki content.
+- Do not hand-edit generated Schema Catalog or Agent metadata. Change their
+  reviewed inputs or generators, then regenerate.
 
-Generated Schema JSON is committed. Change its source inputs and generators,
-then regenerate; do not hand-edit generated Catalog or Agent metadata files.
-`internal/cli/schema_command_registry.json` is different: it is a reviewed
-`CommandRegistry` source, not a generated snapshot. It is the single reviewed
-source of stable canonical identity,
-primary paths, aliases, and navigation. Edit it only when reviewed exposure,
-identity, primary path, or aliases change; parameter, Skill, and metadata-only
-changes must not rewrite it mechanically.
+## Read by task
 
-## Agent Schema contract
+| Change surface | Required guide |
+|---|---|
+| Any implementation or review | [`CONTRIBUTING.md`](CONTRIBUTING.md) and [`docs/coding-agent-guide.md`](docs/coding-agent-guide.md) |
+| Writing a task for a coding agent | [`docs/coding-agent-task-template.md`](docs/coding-agent-task-template.md) |
+| Overall architecture or package layering | [`docs/architecture.md`](docs/architecture.md) |
+| Product command handler behavior | [`internal/helpers/AGENTS.md`](internal/helpers/AGENTS.md) |
+| Helpers package/file layout or megafile splits | [`docs/helpers-structure-guide.md`](docs/helpers-structure-guide.md) |
+| Bundled skill authoring (`skills/`) | [`skills/AGENTS.md`](skills/AGENTS.md) and [`docs/skill-authoring-guide.md`](docs/skill-authoring-guide.md) |
+| CLI paths, flags, Schema, Agent metadata, or generated Catalog | [`docs/schema-contributor-guide.md`](docs/schema-contributor-guide.md) |
+| CI, release, packaging, or repository automation | [`docs/automation.md`](docs/automation.md) |
+| Agent identification headers or host integration | [`docs/agent-code.md`](docs/agent-code.md) |
 
-The Schema data flow is one way:
+Read the closest code and tests for the affected package as well. Nested
+`AGENTS.md` files take precedence for their subtrees.
 
-```text
-1. app.NewRootCommand()
-   └─ builds the real Cobra command tree and flags
+## Common checks
 
-2. schema_command_registry.json
-   + schema_hints/metadata/<product>.json tool parameters (+ cli_path)
-   └─ forms EffectiveCommandRegistry
-      └─ binds exactly to real Cobra leaves and aliases
-
-3. Parameter resolution
-   Cobra flags
-   + schema_parameter_bindings.json
-   + metadata tool parameters
-   └─ produces ParameterSpec and constraints
-
-4. Agent and interface semantics
-   schema_hints/selection/<product>.json   (selection prose)
-   + schema_hints/metadata/<product>.json  (safety/interface/runtime_gate)
-   + pinned MCP metadata
-   └─ resolves Agent metadata by source precedence
-      Markdown is evidence only; it is not concatenated into final prose
-
-5. One typed hub
-   BoundCommandRegistry
-   + ParameterSpec
-   + Agent metadata
-   + Interface metadata
-   └─ resolves every command exactly once into ToolSpec
-      └─ aggregates SchemaRegistry + SchemaIndex
-
-6. One-way publication
-   SchemaRegistry
-   └─ internal/cli/schema_catalog.json
-      └─ dws schema list/product/group/leaf/--all
-```
-
-Parameter overlays from metadata are merged into `EffectiveCommandRegistry`
-*before* Cobra binding; after that point 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`.
-Build-time gates and the snapshot serializer consume that source-resolved typed
-registry/index. Runtime projections and delivery gates consume the typed
-registry/index returned by the production snapshot loader. Neither path may
-reopen annotations, merge source records, or use a previous Catalog or other
-generated JSON as a source. `schema_catalog.json` is output-only in the
-generation graph. The production loader decoding the embedded published
-snapshot is a delivery boundary, not source resolution; it must never create or
-repair a Cobra command, flag, registry entry, or later Catalog generation.
-
-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 reviewed `CommandRegistry` 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.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:
-
-- `internal/cli/schema_command_registry.json` for the reviewed
-  `CommandRegistry`: canonical identity, primary CLI path, aliases, and stable
-  navigation. It is the identity source and is not a generated artifact.
-- `internal/cli/schema_command_registry.schema.json` is its closed,
-  machine-readable editing contract. Preserve the local `$schema` reference;
-  unknown fields, invalid visibility values, stale paths, and collisions fail
-  Go validation and policy.
-- `internal/cli/schema_hints/metadata/<product>.json` for safety, interface,
-  `runtime_gate`, and optional parameter overlays (`parameters` / `cli_path`).
-- `internal/cli/schema_hints/selection/<product>.json` for reviewed Agent
-  selection prose (`agent_summary`, `use_when`, `avoid_when`, `examples`).
-- `internal/cli/schema_hints/index.json` only maps product IDs to those files.
-- 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.
-- Generated files under `internal/cli/schema_agent_metadata/` and
-  `internal/cli/schema_catalog.json` after running generation.
-
-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.
-
-Metadata parameter overlays must reference an exact public runnable Cobra leaf
-and real flags. They may override Schema description, interface-property/type
-mapping, `required`, and `required_when`; they must not create commands or
-flags, define an interface, or advertise an unknown RPC. Every authored entry
-requires `reviewed: true` and a non-empty review reason.
-
-For Agent-authored metadata or selection edits:
-
-1. Confirm the exact command and flag names in the current Cobra tree.
-2. Edit only the owning block (`metadata/` or `selection/`); do not mix fields.
-3. Add the smallest possible entry; do not copy generated Catalog fields into
-   the input.
-4. Describe user-visible semantics in `review_reason` and parameter
-   descriptions.
-5. Run generation, drift, Schema policy, and the focused CLI tests before
-   proposing the change.
-
-## Agent curation workflow (Schema hints)
-
-Use this workflow when refreshing Agent selection prose and confirmation
-alignment. Prefer **agent-authored review** over bulk merge scripts that dump
-`selection-review.json` or Skill Markdown into Catalog fields.
-
-Human-authored inputs are split into two blocks:
-
-| Block | Path | Owns |
-|---|---|---|
-| **metadata** | `internal/cli/schema_hints/metadata/<product>.json` | `effect` / `risk` / `confirmation` / `idempotency` / `interface_*` / `runtime_gate` / optional `parameters` |
-| **selection** | `internal/cli/schema_hints/selection/<product>.json` | `agent_summary` / `use_when` / `avoid_when` / `examples` (+ product routing) |
-
-`index.json` only maps product IDs to those files. Do not mix selection fields
-into metadata files or metadata fields into selection files.
-
-### 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 `reviewed_explicit` from
-   `selection/`.
-2. **Safety** follows Runtime: `confirmation=user_required` iff the tool's
-   metadata `runtime_gate != none` (for example `confirm_delete`, `typed_yes`,
-   `confirm_dangerous`).
-3. **Parameter overrides** (former Manual `commands`) live on metadata tools as
-   `parameters` (+ `cli_path`) and are applied into EffectiveCommandRegistry.
-
-### Authoring
-
-For every curated tool:
-
-1. Edit `metadata/<product>.json` for safety/interface/gates/parameters.
-2. Edit `selection/<product>.json` for selection prose (`reviewed: true`,
-   `review_reason`, `source_refs`).
-3. Run `make generate-schema`. Do not hand-edit generated
-   `schema_agent_metadata/` or `schema_catalog.json`.
-
-### Pull live MCP descriptions (personal token)
-
-Pinned `internal/cli/schema_mcp_metadata.json` is a sanitized baseline. Prefer
-live Schema from a logged-in personal session:
+Choose checks from the matrix in `docs/coding-agent-guide.md`; do not claim a
+check that was not run.
 
 ```bash
-dws auth status                 # token_valid should be true
-dws cache refresh               # refresh discovery / tools cache
-dws schema <mcp-canonical> -f json
-# or CLI path: dws schema --cli-path "drive copy" -f json
+make coding-agent-harness
+make build
+make format-check
+make test
+make policy
+git diff --check
 ```
 
-Resolve MCP identity via `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 + pinned MCP, and record evidence
-(for example `live-dws-schema:<path>#FAILED`). Never print or commit tokens.
-
-Precedence when sources disagree: **Runtime/Cobra > live MCP > pinned 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`
-  for its tools.
-- Hand-write selection + metadata; forbid wholesale JSON merges from review
-  dumps.
-- Edit only its `metadata/<product>.json` and `selection/<product>.json`.
-- **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
-```
-
-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 provenance is
-`reviewed_explicit` from `selection/`, and `user_required` count equals
-metadata `runtime_gate != none`.
-
-`make generate-schema` is a full deterministic snapshot rebuild, not an
-incremental patch over the previous Catalog. It rereads every reviewed input,
-removes stale generated product metadata, and rewrites the exact metadata and
-Catalog projections. Incremental work happens only when an Agent or human
-edits selected `metadata/` or `selection/` entries; the next publication still
-recomputes all outputs. Generated files must never be read back as merge input,
-and byte guards fail generation if it changes the hint inputs or CommandRegistry.
-
-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 TestManualAgentSelectionArkLive -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
-manual/hint overlays. Overlays may still raise an optional flag to required.
-`cli_required` continues to mirror the executable Cobra marker.
-
-For command text, reviewed `ToolSchemaHint` wins first, then command-specific
-Cobra Help, then MCP metadata. 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.
-
-## 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 -> product/group
-  -> leaf for routine Agent work. `--compact` is supported for context-saving
-  projections, but 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 leaf Schema defines Agent selection, parameter mapping
-  and constraints, and safety/confirmation semantics. 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.
+For Schema work, the minimum generation entry point is `make generate-schema`.
+For CLI path or flag changes, also run
+`./scripts/policy/check-command-surface.sh --strict`. Report failures,
+environment limits, and unrun checks explicitly in the handoff.

CHANGELOG.md

@@ -6,6 +6,89 @@ The format is inspired by [Keep a Changelog](https://keepachangelog.com/) and th
 
 ## [Unreleased]
 
+## [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.
@@ -13,6 +96,9 @@ The format is inspired by [Keep a Changelog](https://keepachangelog.com/) and th
 
 ### 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

CONTRIBUTING.md

@@ -10,9 +10,13 @@ under the project [Apache License 2.0](./LICENSE).
 ## Before You Start
 
 1. Read `README.md`.
-2. Read the relevant docs under `docs/`.
-3. Inspect the code and tests for the area you will change.
-4. Decide the smallest safe change that satisfies the request.
+2. Normalize the task and select checks with
+   [`docs/coding-agent-guide.md`](./docs/coding-agent-guide.md).
+3. Read the relevant docs under `docs/`. CLI/Schema/Agent metadata work must
+   also follow
+   [`docs/schema-contributor-guide.md`](./docs/schema-contributor-guide.md).
+4. Inspect the code and tests for the area you will change.
+5. Decide the smallest safe change that satisfies the request.
 
 Maintainers and automation authors should also read
 `docs/automation.md` for repo-local release and agent workflow
@@ -36,8 +40,8 @@ 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

Formula/dingtalk-workspace-cli-beta.rb

@@ -1,33 +1,33 @@
 class DingtalkWorkspaceCliBeta < Formula
   desc "Automate DingTalk workspace tasks from the terminal (beta channel)"
   homepage "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli"
-  version "1.0.53-beta.2"
+  version "1.0.54-beta.2"
   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/v1.0.53-beta.2/dws-darwin-arm64.tar.gz"
-      sha256 "47d3f470003a309f4a93a4dfe55ab39240018b7c698f7863d85834e1f0a3affa"
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.54-beta.2/dws-darwin-arm64.tar.gz"
+      sha256 "46b57bed1f6e9f7ba007d8a86a6f5eb280fdeb557fc9bb5946f14f9b1f8f0c9f"
     else
-      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.53-beta.2/dws-darwin-amd64.tar.gz"
-      sha256 "2dcf90b515d934e71715d95098d3b3cec34299cdbe6c858e1106a81d23d3d51a"
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.54-beta.2/dws-darwin-amd64.tar.gz"
+      sha256 "1b7fd08e64b1c86bbcee217604ffe07e0e8f1b3b5c4de518534386972bcf0f9b"
     end
   end
 
   on_linux do
     if Hardware::CPU.arm?
-      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.53-beta.2/dws-linux-arm64.tar.gz"
-      sha256 "b4692a3c2690460c039e641f75f6793daf3b8549b8c9a35d01153a908f6cc2b1"
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.54-beta.2/dws-linux-arm64.tar.gz"
+      sha256 "108d3861ef606519f9934530d29654eab55a73607d1ee6775461f98ef5a6acd4"
     else
-      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.53-beta.2/dws-linux-amd64.tar.gz"
-      sha256 "000d29d4c81589553e5b23b573002b7014b16c073793b8d7c5617e9b89488175"
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.54-beta.2/dws-linux-amd64.tar.gz"
+      sha256 "6cb96ee09419bbbcc1eb336218ac2aa1d9ca0ed5cbd5a80c79bc20e1e1f03ff7"
     end
   end
 
   resource "skills" do
-    url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.53-beta.2/dws-skills.zip"
-    sha256 "b55a4eaaa63073147c3b9efff48b4713be75577c8a1d2dc8c6e11dc30b4f91c8"
+    url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.54-beta.2/dws-skills.zip"
+    sha256 "572b93f04a10268d185ad1f8e70e0d412949ae056be8494a9949387076fd14bc"
   end
 
   def install

Formula/dingtalk-workspace-cli.rb

@@ -1,32 +1,33 @@
 class DingtalkWorkspaceCli < Formula
   desc "Automate DingTalk workspace tasks from the terminal"
   homepage "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli"
-  version "1.0.52"
+  version "1.0.54"
   license "Apache-2.0"
 
+
   on_macos do
     if Hardware::CPU.arm?
-      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.52/dws-darwin-arm64.tar.gz"
-      sha256 "4f6b4d064a76bcefac42feb5f356253fe43f9499b8cec9d2cdf202e7d3b9b60c"
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.54/dws-darwin-arm64.tar.gz"
+      sha256 "8ae0e52cf973f6fb3df61c67a41fd11e2df417a0c815762b6060cbcb5e600c08"
     else
-      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.52/dws-darwin-amd64.tar.gz"
-      sha256 "abc87128f4b98d0a01ea99235449031971db8fa4ce94167403e3b736c4b81e9a"
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.54/dws-darwin-amd64.tar.gz"
+      sha256 "11b711b9d70dea62304bf5f8206c56b4e7ea91148dafe97fb7c0f844a2a61da3"
     end
   end
 
   on_linux do
     if Hardware::CPU.arm?
-      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.52/dws-linux-arm64.tar.gz"
-      sha256 "0d357ef0535f99f2f63b5ecbfdee9c32448be2a2c24f3096c03126b3b7570bc5"
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.54/dws-linux-arm64.tar.gz"
+      sha256 "9c7ecb4c8cd55644b2faa73f6ce7843c0279b23793e23deb5061692ea71a0cf1"
     else
-      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.52/dws-linux-amd64.tar.gz"
-      sha256 "b7dfd9a4b3489211359261747ed0cb9c8c261434bb762ad3f76df33bdbabd5cb"
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.54/dws-linux-amd64.tar.gz"
+      sha256 "8a0bc245747fc3facf98c8103c06da46852a30bff31ac93b0aa874e8c7e46db7"
     end
   end
 
   resource "skills" do
-    url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.52/dws-skills.zip"
-    sha256 "0fa3c8dec500c1659e6480d6772ae901b2d12d24322dd5d7283f016024290c21"
+    url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.54/dws-skills.zip"
+    sha256 "7450fd0115c75bfe6820c7099f348973d9353cca9d8d647c9cddcd70978a7ec0"
   end
 
   def install
@@ -52,6 +53,7 @@ class DingtalkWorkspaceCli < Formula
     <<~EOS
       Agent Skills are bundled in #{pkgshare}/skills/dws.
       Run `dws skill setup` to install them into your Agent directories.
+
     EOS
   end
 

Makefile

@@ -1,12 +1,14 @@
 GO ?= go
+DWS_PACKAGE_VERSION ?= 0.0.0-test
 REMOTE ?=
 PUBLISH ?= 0
 YES ?= 0
 DWS_POLICY_TMPDIR ?= $(CURDIR)/.worktrees/policy-tmp
 POLICY_GOTMPDIR ?= $(DWS_POLICY_TMPDIR)/go
 POLICY_ENV = DWS_POLICY_TMPDIR="$(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 interface-integrity authoritative-interface-integrity coverage-gate coverage-gate-platform update-interface-baseline reset-interface-baseline schema-compatibility skill-command-integrity cli-smoke mock-mcp-smoke test-schema-agent-examples generate-schema generate-schema-agent-metadata generate-schema-catalog package release release-pre release-stable changelog-pre changelog-stable publish-homebrew-formula setup-hooks
+.PHONY: all help build rebuild test test-plan lint format-check fmt policy coding-agent-harness coding-agent-task edition-test interface-integrity authoritative-interface-integrity coverage-gate coverage-gate-platform update-interface-baseline reset-interface-baseline schema-compatibility skill-command-integrity cli-smoke mock-mcp-smoke test-schema-agent-examples generate-schema generate-schema-agent-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
 
@@ -14,13 +16,17 @@ 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 test-plan     - Verify every default Go package belongs to one CI test shard\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 coding-agent-harness - Validate coding-agent task intake, routing, and self-check contracts\n"
+	@printf "  make coding-agent-task TASK=<file> - Validate a filled coding-agent task contract\n"
 	@printf "  make interface-integrity - Check historical commands and help contracts still work\n"
 	@printf "  make authoritative-interface-integrity BASE_REF=<ref> - Check the Git-owned PR merge-base\n"
-	@printf "  make coverage-gate BASE_REF=<ref> - Enforce overall non-regression and changed-code coverage\n"
-	@printf "  make coverage-gate-platform BASE_REF=<ref> PROFILE=<file> - Enforce native-platform changed-code coverage\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 - Add new CLI contracts without removing history\n"
 	@printf "  make reset-interface-baseline - DANGEROUS: replace all CLI compatibility history\n"
 	@printf "  make schema-compatibility BASE_REF=<ref> - Check the complete Schema contract against the PR merge-base\n"
@@ -45,13 +51,32 @@ rebuild:
 	@./scripts/dev/build.sh
 
 test:
-	@./test/scripts/run_all_tests.sh --timeout 5m
+	@DWS_PACKAGE_VERSION="$(DWS_PACKAGE_VERSION)" $(GO) test -count=1 -timeout=10m ./...
+
+test-plan:
+	@./scripts/ci/test-packages.sh verify
 
 lint:
 	@./scripts/dev/lint.sh
 
+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
+
 fmt:
-	@find cmd internal test scripts/policy -name '*.go' -print0 2>/dev/null | xargs -0r gofmt -w
+	@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:
 	@mkdir -p "$(POLICY_GOTMPDIR)"
@@ -63,6 +88,13 @@ policy:
 	@$(POLICY_ENV) ./scripts/policy/check-schema-binary.sh
 	@$(POLICY_ENV) $(MAKE) test-schema-agent-examples
 
+coding-agent-harness:
+	@./scripts/policy/check-coding-agent-harness.sh
+
+coding-agent-task:
+	@test -n "$(TASK)" || { printf '%s\n' 'TASK is required, e.g. make coding-agent-task TASK=task.md' >&2; exit 2; }
+	@./scripts/policy/check-coding-agent-harness.sh -task "$(TASK)"
+
 edition-test:
 	$(GO) test -v -count=1 ./pkg/editiontest/...
 

README.md

@@ -476,6 +476,8 @@ Env vars: `DWS_SKILL_MODE=mono|multi` (also honored by `install.sh` / `install.p
 
 `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 currently covers messages that mention the current user, one-to-one messages with a specified user, and messages in a specified group.
 
+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:
@@ -487,19 +489,19 @@ curl -fsSL https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace
 ```bash
 # Inspect the public personal event catalog and schema
 dws event list
-dws event schema user_im_message_receive_o2o
+dws event schema user_im_message_receive_o2o --flatten
 
 # Listen for messages that mention the current user
-dws event consume user_im_message_receive_at -f ndjson
+dws event consume user_im_message_receive_at --flatten -f ndjson
 
 # Listen for one-to-one messages with a specified user
-dws event consume user_im_message_receive_o2o --user <userId> -f ndjson
+dws event consume user_im_message_receive_o2o --user <userId> --flatten -f ndjson
 
 # Listen by openDingtalkId (external contact, bot, or cross-organization identity)
-dws event consume user_im_message_receive_o2o --open-dingtalk-id <openDingtalkId> -f ndjson
+dws event consume user_im_message_receive_o2o --open-dingtalk-id <openDingtalkId> --flatten -f ndjson
 
 # Listen for messages in a specified group
-dws event consume user_im_message_receive_group --group <openConversationId> -f ndjson
+dws event consume user_im_message_receive_group --group <openConversationId> --flatten -f ndjson
 
 # Inspect local consumers and cancel a subscription
 dws event status

README_zh.md

@@ -470,6 +470,8 @@ DWS_SKILL_SOURCE=/path/to/skills dws skill setup --mode multi
 
 `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。当前公开目录包括：当前用户被 @ 的消息、与指定用户的单聊消息、指定群的消息。
 
+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：
@@ -481,19 +483,19 @@ curl -fsSL https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace
 ```bash
 # 查看公开个人事件目录和 schema
 dws event list
-dws event schema user_im_message_receive_o2o
+dws event schema user_im_message_receive_o2o --flatten
 
 # 监听当前用户被 @ 的消息
-dws event consume user_im_message_receive_at -f ndjson
+dws event consume user_im_message_receive_at --flatten -f ndjson
 
 # 监听与指定用户的单聊消息
-dws event consume user_im_message_receive_o2o --user <userId> -f ndjson
+dws event consume user_im_message_receive_o2o --user <userId> --flatten -f ndjson
 
 # 使用 openDingtalkId 监听外部联系人、机器人或跨组织身份
-dws event consume user_im_message_receive_o2o --open-dingtalk-id <openDingtalkId> -f ndjson
+dws event consume user_im_message_receive_o2o --open-dingtalk-id <openDingtalkId> --flatten -f ndjson
 
 # 监听指定群的消息
-dws event consume user_im_message_receive_group --group <openConversationId> -f ndjson
+dws event consume user_im_message_receive_group --group <openConversationId> --flatten -f ndjson
 
 # 查看本地 consume，并取消指定订阅
 dws event status

docs/architecture.md

@@ -2,6 +2,29 @@
 
 `dws` is a Go CLI with a versioned, static command surface for DingTalk MCP capabilities. Cobra help serves humans; the embedded Command Catalog serves AI agents.
 
+## Change Rules
+
+Prescriptive layering for new code. Keep descriptions here concise; scoped
+guides own the details.
+
+1. Dependencies point inward: `cmd` → `internal/app` → `internal/helpers` →
+   shared layers (`executor`, `transport`, `output`, `errors`, `safety`,
+   `cobracmd`). Shared layers never import `helpers` or `app`.
+2. New product commands go to `internal/helpers` following
+   [`helpers-structure-guide.md`](helpers-structure-guide.md); do not add
+   product logic to `internal/app`, `internal/cli`, or transport.
+3. New shared behavior joins the existing shared package that owns the
+   contract; do not create a new shared package for a single caller.
+4. Schema/Agent metadata changes start from reviewed inputs in `internal/cli`
+   per [`schema-contributor-guide.md`](schema-contributor-guide.md); never
+   hand-edit generated Catalog output.
+5. Bundled skill content under `skills/` follows
+   [`skill-authoring-guide.md`](skill-authoring-guide.md); skills are embedded
+   via `skills/embed.go` and ship with the binary.
+6. A new top-level package (under `internal/` or `pkg/`) requires a stated
+   boundary reason in its PR and an update to the Repository Structure list
+   below.
+
 ## High-Level Flow
 
 1. `cmd` is the CLI entrypoint, invoking `internal/app` to build the root Cobra command tree.
@@ -43,3 +66,38 @@
 - `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"] --> CA["CI"]
+  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
+```
+
+Complete Multi-profile E2E and the ordinary full native-platform matrix are
+downstream of PR admission. PRs still run primary-environment assurance and
+fast cross-platform compilation; auth, keychain, OS-specific, installer, and
+release changes additionally select native platform tests before merge. See
+[`docs/ci-pr-gates.md`](ci-pr-gates.md) for the exact context and ruleset
+contract.

docs/automation.md

@@ -108,6 +108,34 @@ commit, and failed tag-push run all match; it then reuses the normal release
 jobs. 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

@@ -1,133 +1,158 @@
-# Pull request quality gates
+# CI — PR 合入门禁
 
-The repository defines five focused checks in addition to its existing CI:
+The pull-request admission layer has exactly nine required external contexts:
 
-- **Interface Integrity** enforces backwards compatibility. Every historical
-  command path and alias must still resolve, every historical command must
-  still render `-h`, and historical flags must keep their type and shorthand.
-  New commands, aliases, and flags are allowed. The same job compares the full
-  complete `dws schema --all` contract with the PR merge-base, blocking removed
-  products/tools/parameters, incompatible parameter or interface mappings,
-  constraint drift, and safety-semantic drift. It also checks that executable
-  `dws ...` references in `skills/**/*.md` resolve to real commands.
-  Help compatibility covers command/alias/flag spelling, flag type and
-  shorthand; descriptive prose may evolve without breaking the gate.
-- **Coverage** runs unit tests on every pull request and prints both overall and
-  changed-code statement coverage. During the migration to the 80% repository
-  target, overall coverage may not regress from a profile generated from the
-  merge-base with the same test command, while changed production Go
-  statements must meet 80%. Linux, Windows, and macOS each generate a native
-  coverage profile for changed packages and enforce the threshold against
-  changed files buildable on that platform, so build-tagged source cannot be
-  hidden by an Ubuntu-only profile. Overall non-regression allows 0.1 percentage point of measurement
-  variance to avoid failing unchanged code on test-path noise. Set
-  `COVERAGE_ENFORCE_OVERALL=true` once repository coverage reaches 80% to make
-  the overall target fail closed as well.
-  CI generates the candidate, supporting, and merge-base profiles on three
-  independent runners, then downloads all profiles into the aggregate
-  `Coverage` job and applies the same fail-closed gate. The split changes only
-  scheduling: the tested packages, profile contents, merge-base comparison,
-  and final coverage thresholds remain unchanged.
-- **CLI Smoke** builds the release binary, reads the root command list from the
-  structured Interface contract, and renders offline help for every public
-  top-level command. It rejects Cobra's unknown-command root-help fallback and
-  fails when the checked-in development fixture is stale.
-- **Mock MCP Smoke** runs the existing HTTP and stdio MCP lifecycle tests
-  (`Initialize -> ListTools -> CallTool`).
-- **AI Behavior Check** applies to pull requests labeled `ai-generated`. It
-  limits the change to 30 files and blocks release/CI infrastructure changes,
-  including policy implementations and the checked-in Interface fixture.
-  It uses `pull_request_target` without checking out PR code, so the policy
-  cannot be bypassed by changing the workflow in the same pull request. The
-  evaluator writes an `AI Behavior Check` commit status to the PR head SHA so
-  GitHub rulesets can require it.
+| Required context | Contract |
+|---|---|
+| `Lint` | Stable PR revision classification, formatting, `go vet`, and Actionlint |
+| `Test` | Race/unit/release-script tests plus fast cross-platform compilation |
+| `Coverage` | 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 |
 
-## Running the compatibility gates
+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.
 
-Run:
+`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. After merge, the protected `main` push executes the
+full admission suite.
+
+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.
+
+## Platform and downstream boundaries
+
+Ordinary PRs run the primary Linux assurance plus fast Darwin/Windows compile
+checks. Full native macOS/Windows tests and platform coverage run on a PR only
+when its diff touches auth, keychain, OS-specific Go files, installers,
+packaging, Formulae, or release automation. Protected `main` pushes run the
+complete native matrix.
+
+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"]
+```
+
+## Running focused gates locally
+
+Run the contracts relevant to the change:
 
 ```sh
 make build
+make policy
 make interface-integrity
 make authoritative-interface-integrity BASE_REF=<merge-base>
 make schema-compatibility BASE_REF=<merge-base>
 make skill-command-integrity
 make cli-smoke
-# Run on the corresponding native runner with its generated profile:
-make coverage-gate-platform BASE_REF=<merge-base> PROFILE=<coverage-profile>
+make mock-mcp-smoke
+go test -v -count=1 ./pkg/editiontest/...
 ```
 
-`make coverage-gate` is the enforcement step, not a profile generator. It
-expects the candidate, policy, shortcut, and merge-base profiles
-(`coverage.txt`, `coverage-policy.txt`, `coverage-shortcut.txt`, and
-`coverage-base.txt`) produced by the parallel CI profile jobs. A clean local
-checkout can reproduce the Linux/overall CI gate sequentially with:
+For an exact CHANGELOG-only branch:
 
 ```sh
 base_ref=$(git merge-base HEAD origin/main)
-root=$(pwd)
-base_worktree=$(mktemp -d "${TMPDIR:-/tmp}/dws-coverage-base.XXXXXX")
-rmdir "$base_worktree"
-cleanup() { git worktree remove --force "$base_worktree" >/dev/null 2>&1 || true; }
-trap cleanup EXIT HUP INT TERM
-
-go test -count=1 -p 1 -coverprofile=coverage.txt -covermode=atomic \
-  ./ ./cmd/... ./internal/... ./skills/...
-go test -count=1 -coverprofile=coverage-policy.txt -covermode=atomic \
-  ./pkg/... ./scripts/policy/...
-go test -count=1 \
-  -run '^(TestAllShortcuts|TestCrossPlatformCoverage)' \
-  -coverpkg=./internal/app,./internal/helpers,./internal/shortcut/... \
-  -coverprofile=coverage-shortcut.txt \
-  -covermode=atomic \
-  ./internal/app ./internal/helpers ./internal/shortcut/...
-git worktree add --detach "$base_worktree" "$base_ref"
-(
-  cd "$base_worktree"
-  go test -count=1 -p 1 \
-    -coverprofile="$root/coverage-base.txt" -covermode=atomic \
-    ./ ./cmd/... ./internal/... ./skills/...
-)
-COVERAGE_ADDITIONAL_PROFILE=coverage-shortcut.txt \
-  make coverage-gate BASE_REF="$base_ref"
+./scripts/policy/check-changelog-pr.sh --fast-path "$base_ref" HEAD
 ```
 
-The native-platform target likewise expects `PROFILE` to have already been
-generated on that operating system. CI owns those generation steps; copying
-only either enforcement command into a clean checkout is intentionally an
-incomplete invocation.
+`make coverage-gate` is an enforcement step, not a profile generator. CI
+generates the candidate, supporting, merge-base, and (when risk-selected)
+native profiles 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 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.
 
-CI derives the authoritative Interface snapshots from both the PR merge-base
-and the latest reachable stable release tag. The complete Schema snapshot comes
-from the PR merge-base, which contains the registry-first Schema introduced on
-`main`. The candidate branch cannot bless a breaking change by editing a
-fixture. Schema additions are allowed; historical products, tools, parameters,
-parameter mappings, positional execution fields, constraints, and safety
-semantics remain protected. Positional descriptions are documentation and may
-change without breaking compatibility.
-
-`make update-interface-baseline` still extends the local checked-in Interface
-fixture used by `make interface-integrity`. Updates are monotonic: they add new
-commands and flags without removing history.
-
-For an intentional compatibility reset at a major-version boundary, run
-`make reset-interface-baseline`. This replaces all CLI compatibility history
-with the current command tree and must receive explicit human review.
+Compatibility checks derive authoritative Interface snapshots from the PR
+merge-base and the latest reachable stable release. The candidate cannot bless
+a breaking change by editing a fixture. Schema additions are allowed;
+historical products, tools, parameters, mappings, positional execution fields,
+constraints, and safety semantics remain protected.
 
 ## Required GitHub repository settings
 
-Create a ruleset for `main` that requires pull requests and code-owner review,
-then mark these aggregate status checks as required:
+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:
 
-- `CI Gate`
-- `Multi Profile E2E`
-- `AI Behavior Check`
+- `Lint`
+- `Test`
+- `Coverage`
+- `Policy`
+- `Edition`
+- `Interface Integrity`
+- `AI Behavior`
+- `CLI Smoke`
+- `Mock MCP`
 
-`CI Gate` fails closed unless every first-layer CI job succeeds, including
-lint, tests, native Linux/Windows/macOS coverage, policy,
-Interface/Schema/Skill integrity, and smoke tests. Requiring the aggregate
-check keeps repository rules stable when an internal job is renamed or split.
-
-The `ai-generated` label must be applied by the PR-creation automation or by a
-maintainer; GitHub cannot infer reliably whether a human-authored PR contains
-AI-generated code.
+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.

docs/coding-agent-guide.md

@@ -0,0 +1,112 @@
+# Coding Agent Workflow
+
+This is the task intake and self-check contract for coding agents working in
+this repository. It is intentionally independent of external Wiki systems:
+the checked-out repository is the execution context and evidence source.
+
+## 1. Normalize the task input
+
+Start from the copyable
+[`coding-agent-task-template.md`](coding-agent-task-template.md). Keep one
+primary outcome per task. Fill unknown fields from the issue, nearby code,
+tests, and versioned docs; state any assumption that can affect behavior.
+
+For a saved, filled template, run `make coding-agent-task TASK=path/to/task.md`.
+The local checker rejects missing required fields, unsupported task kinds, and
+unresolved placeholders before implementation begins.
+
+Do not invent acceptance criteria that expand the requested behavior. Stop for
+user input only when the unresolved choice would change externally visible
+behavior, compatibility, destructive scope, credentials, or external state.
+
+## 2. Establish the baseline
+
+1. Run `git status --short` and identify pre-existing changes.
+2. Read the applicable guides linked from the root `AGENTS.md`.
+3. Locate the implementation and its closest tests with `rg`/`rg --files`.
+4. Reproduce the bug or capture the current contract before changing it.
+5. Pick the smallest owning layer; avoid duplicating policy in a caller when a
+   shared typed layer already owns it.
+
+Never clean, overwrite, stage, or reformat unrelated user changes. If a
+required file is already modified, inspect the overlap and preserve both
+intents or stop with the exact conflict.
+
+## 3. Implement from authoritative inputs
+
+- Go behavior belongs in the package that owns the contract, with focused
+  tests beside it.
+- Product handlers under `internal/helpers` follow
+  [`helpers-structure-guide.md`](helpers-structure-guide.md): thin
+  `{product}.go` wiring plus `{product}_{resource}.go` files; do not enlarge
+  megafiles such as `chat.go`.
+- Public CLI paths and flags must match the live Cobra tree and compatibility
+  policies.
+- Schema and Agent-facing changes start from reviewed source inputs; generated
+  outputs are publication artifacts.
+- Documentation describes behavior that exists in the same change.
+- Secrets, tokens, local identities, and private endpoints must not enter code,
+  fixtures, logs, or handoff output.
+
+For generated files, run the repository generator and inspect the resulting
+diff. A large or unrelated generated diff is a signal to stop and find the
+wrong input or nondeterminism, not something to accept automatically.
+
+## 4. Select validation by change surface
+
+Run the narrow check while iterating, then the applicable admission checks
+before handoff. `make help` is the authoritative target list.
+
+| Changed surface | Focused check | Admission checks |
+|---|---|---|
+| Documentation only | inspect links/examples | `git diff --check` |
+| Go implementation | `go test ./path/to/package` | `make format-check`, `make test` |
+| CLI paths or flags | focused command/help tests | `make build`, `./scripts/policy/check-command-surface.sh --strict`, `make interface-integrity` |
+| Schema registry, hints, or generators | focused generator/app tests | `make generate-schema`, `./scripts/policy/check-generated-drift.sh`, `./scripts/policy/check-schema-catalog.sh`, `make test-schema-agent-examples` |
+| Skill command examples | inspect referenced `dws` help | `make skill-command-integrity` |
+| CI or test sharding | run the affected script/test | `make test-plan`, `make lint`, and the CI workflow's pinned actionlint command when workflows change |
+| Packaging or installers | focused release-script tests | `make package`, `./scripts/release/verify-package-managers.sh` |
+| Authentication, transport, or OS-specific code | focused tests, including failure paths | `make test`; run relevant platform checks or disclose the unavailable platform |
+
+`make policy` is the combined policy gate and is appropriate for command,
+Schema, generated-asset, or broad cross-cutting changes. Platform credentials
+and live services are not prerequisites for ordinary unit tests; never turn a
+missing credential into permission to skip deterministic checks.
+
+The guide contract itself is executable. Run `make coding-agent-harness` after
+changing `AGENTS.md`, this guide, the Schema contributor guide, helpers
+structure guide, or their routed commands and paths. This remains a local,
+opt-in agent aid and is not wired into CI.
+
+## Design references
+
+This repository adapts two patterns without copying their product-specific
+rules:
+
+- [Lark CLI's contributor guide](https://github.com/larksuite/cli/blob/5efaf65aec59c33899475bb90e6bff1bc3b5b65c/AGENTS.md): one primary goal, machine-consumable errors/output, and validation selected by behavior surface.
+- [WeCom CLI's root routing guide](https://github.com/WecomTeam/wecom-cli/blob/9eb7898b959861af879495e211e37431fa908f19/AGENTS.md) and [human helper template](https://github.com/WecomTeam/wecom-cli/blob/9eb7898b959861af879495e211e37431fa908f19/src/helpers/HUMANS.md): a thin root guide, scoped implementation guidance, and a copyable request format.
+
+## 5. Pre-handoff self-check
+
+Confirm every applicable item:
+
+- The diff implements the stated goal and no unrelated cleanup.
+- Pre-existing changes are still present and were not attributed to this task.
+- New behavior has a regression test; removed behavior has an explicit reason.
+- Public command paths, flags, output, exit behavior, and compatibility remain
+  intentional.
+- Destructive or mutating operations retain the required confirmation path.
+- Generated files came from their reviewed inputs and generation is clean.
+- Docs and examples use commands accepted by current help/Schema.
+- Errors preserve actionable context without leaking secrets.
+- `git diff --check` passes and the final diff has been read.
+- Every reported check is labeled passed, failed, or not run with a reason.
+
+Use this compact handoff shape:
+
+```text
+Outcome: what is now true
+Files: intentional files changed
+Validation: exact commands and results
+Limits: unrun checks, environment constraints, follow-ups
+```

docs/coding-agent-task-template.md

@@ -0,0 +1,35 @@
+# Coding Agent Task Template
+
+Copy this block into an issue or coding-agent request. One task should have one
+primary outcome; split unrelated outcomes instead of hiding them in acceptance
+criteria.
+
+```text
+Task kind: bug | feature | refactor | docs | policy | release
+Goal (one primary outcome):
+Current behavior and evidence:
+Acceptance criteria:
+In scope (packages/files/surfaces):
+Out of scope:
+Compatibility constraints:
+Interface impact (commands/flags/output/errors/exit codes/Schema):
+Safety or data-mutation constraints:
+Expected validation:
+Known environment limitations:
+```
+
+For a command or remote-interface task, add the smallest concrete invocation
+and contract evidence available:
+
+```text
+CLI path and example argv:
+Current --help or Schema excerpt:
+Remote method and request/response shape, if relevant:
+Expected stdout/stderr and exit behavior:
+Mutation preview/confirmation behavior:
+```
+
+Do not paste credentials, tokens, private endpoints, or production business
+data. Use redacted fixtures and say which evidence is unavailable. Save the
+filled block and run `make coding-agent-task TASK=path/to/task.md` to validate
+it before implementation.

docs/helpers-structure-guide.md

@@ -0,0 +1,214 @@
+# Helpers Package Structure Guide
+
+This is the coding-structure contract for product commands under
+`internal/helpers/`. Business teams and coding agents must follow it when
+adding or moving CLI leaves. Behavioral contracts (stdout, errors, confirmation,
+Schema) stay in [`internal/helpers/AGENTS.md`](../internal/helpers/AGENTS.md);
+this document only owns file layout and split rules.
+
+Reference implementations already in-tree:
+
+| Pattern | Use as |
+|---|---|
+| [`sheet.go`](../internal/helpers/sheet.go) + `sheet_*.go` | Preferred product layout: thin root wiring, resource files |
+| [`chat_media_upload.go`](../internal/helpers/chat_media_upload.go) | Incremental extract from a megafile without behavior change |
+| `connect_*.go` | Concern-based split inside the same `helpers` package |
+
+Anti-pattern to stop growing: single megafiles such as `chat.go` / `aitable.go`
+(thousands of lines). New work must not enlarge them.
+
+## 1. Package boundary
+
+Keep product handlers in the flat package `helpers`:
+
+```text
+internal/helpers/                 # package helpers (default)
+  register_products.go            # public product registration table
+  {product}.go                    # product root: new{Product}Command()
+  {product}_{resource}.go         # one resource / cohesive concern
+  {product}_{resource}_test.go    # focused tests beside the file
+  helpers.go / interfaces.go …    # cross-product shared machinery
+```
+
+Do **not** create `internal/helpers/{product}/` subpackages by default. The flat
+package exists so leaves can share unexported helpers (`callMCPToolOnServer`,
+flag validators, confirmation wrappers, transport adapters) without exporting a
+public API surface. Split into a subpackage only when the concern is a real
+library boundary with its own tests and almost no need for helpers-private
+symbols—and get an explicit review for that exception.
+
+Repository layers outside helpers stay unchanged:
+
+| Layer | Owns |
+|---|---|
+| `internal/app` | Root/static wiring, plugin load |
+| `internal/helpers` | Product Cobra trees and handler behavior |
+| `internal/cobracmd` | Shared Cobra construction primitives |
+| `internal/executor`, `internal/transport` | Invocation and transport |
+| `internal/output`, `internal/errors`, `internal/safety` | Projection, failures, confirmation |
+| `internal/cli` | Schema identity / Agent metadata |
+
+Ordinary product leaves belong in helpers. Do not push product business mapping
+into app, transport, or Schema generators.
+
+## 2. File roles inside a product
+
+### 2.1 Product root — `{product}.go`
+
+Owns exactly one entry constructor, registered from `register_products.go`:
+
+```go
+func newChatCommand() *cobra.Command { /* wire subgroups only */ }
+```
+
+The root file should:
+
+1. Define the product `cobra.Command` (`Use` / `Short` / `Long` / aliases).
+2. Call resource factories and `AddCommand` them.
+3. Register product-level aliases or hint stubs when needed.
+4. Avoid large inline `RunE` bodies for leaves.
+
+Target size: wiring-only, roughly under **400 lines** (see `sheet.go` ≈ 270).
+If the root grows past that because leaves are inlined, extract a resource file.
+
+### 2.2 Resource / concern file — `{product}_{resource}.go`
+
+Split on the **CLI path segment or cohesive concern**, not on “one function per
+file”:
+
+| CLI path | File |
+|---|---|
+| `dws chat group …` | `chat_group.go` |
+| `dws chat message …` | `chat_message.go` |
+| `dws chat media …` | `chat_media_upload.go` (or `chat_media.go`) |
+| `dws sheet filter-view …` | `sheet_filter_view.go` |
+| `dws sheet dimension …` | `sheet_dimension.go` |
+
+Each file exposes one or more unexported factories, for example:
+
+```go
+func newChatGroupCmd() *cobra.Command { … }
+func newChatMessageCmd() *cobra.Command { … }
+func newWorkbookCmds() []*cobra.Command { … }
+```
+
+The product root only wires those factories. Prefer keeping a leaf’s flags,
+`RunE`, and nearby request-mapping helpers in the same resource file until a
+helper is reused by multiple resources.
+
+Soft size guide per resource file:
+
+| Lines | Action |
+|---|---|
+| < 600 | Normal |
+| 600–1000 | Prefer splitting the next cohesive subgroup before adding more |
+| > 1000 | Required split before landing substantial new leaves |
+
+“One command per file” is **not** required. Tiny sibling leaves that share
+flags and mapping belong together. Split when the file holds multiple unrelated
+resources or becomes hard to review.
+
+### 2.3 Shared product helpers — `{product}_{concern}.go`
+
+Use a concern suffix when code is shared across resources and is not itself a
+command tree:
+
+- `sheet_validate.go` — shared validation
+- `chat_args.go` / similar — grant/arg builders used by several leaves
+- `connect_command.go` — slash-command parsing used by the daemon
+
+Do not park unrelated products’ utilities in these files. Cross-product
+machinery belongs in non-prefixed shared files (`helpers.go`, `interfaces.go`,
+output/error helpers), never copied per product.
+
+### 2.4 Tests
+
+- Name tests after the file or scenario: `chat_message_search_test.go`,
+  `sheet_filter_view_test.go`.
+- Prefer exercising public product construction (`newChatCommand().Find(…)`)
+  for command-surface regressions.
+- Pure helpers may be unit-tested directly in the same package.
+- A mechanical file split without behavior change should keep existing tests
+  green with no assertion edits; if tests must change, the split leaked a
+  behavior or visibility change and needs review.
+
+## 3. Registration and naming
+
+1. Public products enter the CLI only through the table in
+   `register_products.go` (`new{Product}Command` factories). Do not invent a
+   second registration path for ordinary product leaves.
+2. Constructor names stay unexported (`new…`) unless a deliberate test helper
+   requires otherwise.
+3. File names are lowercase snake_case. Match the CLI resource token when
+   practical (`filter-view` → `filter_view`).
+4. Do not hand-edit generated sync markers in `register_products.go` outside
+   the product-registration workflow that owns that file.
+
+## 4. How to add a new command (business-team checklist)
+
+1. Confirm the owning product and CLI path with `dws <product> --help`.
+2. Open `{product}.go` only to wire `AddCommand`; put the leaf in
+   `{product}_{resource}.go`.
+3. If the product is still a megafile (`chat.go`, `aitable.go`, …) and your
+   resource file does not exist yet, **create the resource file and move only
+   the subgroup you need** (or add the new leaf there and wire it from the
+   root). Do not append another large leaf into the megafile.
+4. Keep stdout / stderr / error / confirmation contracts from
+   `internal/helpers/AGENTS.md`.
+5. If path, flags, safety, or Agent selection change, follow
+   [`schema-contributor-guide.md`](schema-contributor-guide.md).
+6. Add or extend the closest test; run
+   `go test ./internal/helpers -count=1` (or a tighter `-run`) plus the checks
+   selected by [`coding-agent-guide.md`](coding-agent-guide.md).
+
+## 5. Migrating an existing megafile
+
+Mechanical splits are welcome and preferred over “big-bang” rewrites.
+
+Rules for a split PR:
+
+1. **Behavior-neutral**: same command paths, flags, help text, request mapping,
+   confirmation, and output.
+2. **Stable entrypoint**: keep `new{Product}Command()` as the registration
+   symbol; only its body becomes wiring.
+3. **Move by resource**: extract one CLI subgroup per commit/PR when possible
+   (`group`, `message`, `category`, …).
+4. **No drive-by cleanups** in the same PR (renames, flag redesign, Schema
+   edits) unless the task explicitly includes them.
+5. **Stop growing the megafile**: after the first extract, new leaves for that
+   resource go to the new file only.
+
+Suggested first-wave split for `chat` (illustrative, not mandatory order):
+
+| Extract to | Contents |
+|---|---|
+| `chat.go` | `newChatCommand()` wiring only |
+| `chat_permission.go` | `chmod`, `data-auth` |
+| `chat_group.go` | `group` tree |
+| `chat_message.go` | `message` tree |
+| `chat_category.go` | `category` tree |
+| `chat_bot.go` | `bot` tree |
+| `chat_conversation.go` | top-level conversation ops (`set-top`, mute, red-point, …) |
+| existing `chat_media_upload.go` | keep; optionally rename to `chat_media.go` only in a dedicated rename PR |
+
+Apply the same pattern to `aitable`, `attendance`, `mail`, and `doc` when those
+products take new work.
+
+## 6. What this guide does not change
+
+- Wire format, MCP method names, or Schema identity rules.
+- The choice to keep `package helpers` flat.
+- Runtime confirmation / dry-run policy (still owned by safety + Schema metadata).
+- Permission to skip tests because a change was “only a move”—moves still need
+  the product’s focused tests green.
+
+## 7. Harness check
+
+After editing this guide or its links from `AGENTS.md` /
+`internal/helpers/AGENTS.md`, run:
+
+```bash
+make coding-agent-harness
+```
+
+This check is a local, opt-in agent aid; it is not part of `make policy` or CI.

docs/releasing.md

@@ -1,10 +1,54 @@
 # 发布手册（预发 / 正式）
 
-发布只走一条链路：本地脚本负责封板、验证并推送 annotated tag；GitHub Actions 负责构建和发布最终产物。不要直接运行 `goreleaser release`，也不要手工补打或移动 tag。
+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 以 workflow 自动创建的 Formula PR 经独立审核合入为交付边界。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` 要求 `CI Gate`，操作机已安装并登录 `gh`。本地脚本会在封 tag 前通过 API 检查 immutable releases、当前 SHA 的 `CI Gate` 和在途 Release；`v*` tag ruleset 仍需仓库管理员预先配置并由操作人确认。
+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`。云端和本地入口都会在封 tag 前检查 immutable releases、当前 SHA 的全部九个 context 和在途 Release；`v*` tag ruleset 仍需仓库管理员预先配置。
 
-## 日常只用一个入口
+## 推荐入口：GitHub 云端发布
+
+任何具有仓库写权限、因而可以手动运行 Actions workflow 的成员，都可以基于当时最新的 `main` 发起发布：
+
+1. 在 GitHub Actions 打开 `Release`，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`，并输入 `PUBLISH beta` 或 `PUBLISH stable`。
+
+`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 PR DAG。内置 `GITHUB_TOKEN` 创建的 tag 不依赖第二条 workflow 被再次触发。
+
+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：
 
@@ -25,11 +69,11 @@ dws-release config --remote origin
 ```text
 main 上的候选代码 + beta CHANGELOG
   → vX.Y.Z-beta.N（预发验证）
-  → 只允许补正式 CHANGELOG，源码不得再变化
-  → vX.Y.Z（正式发布）
+  → 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）
 ```
 
-正式版必须显式指定本次验证过的 beta。脚本会比较两者：除 `CHANGELOG.md` 外只要有任何文件变化，就拒绝正式发布。这样预发测过的代码、命令树和正式发布的代码是同一份。
+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
 
@@ -45,7 +89,7 @@ dws-release v1.2.3-beta.1
 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 `CI Gate` 和 immutable releases。通过后会在当前 Git worktree 的私有 Git 状态目录写入一个有效期六小时的证明，绑定版本、精确 commit、发布仓库、beta/stable 基线和远端 `main`：
+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。通过后会在当前 Git worktree 的私有 Git 状态目录写入一个有效期六小时的证明，绑定版本、精确 commit、发布仓库、beta/stable 基线和远端 `main`：
 
 ```bash
 dws-release v1.2.3-beta.1 --publish
@@ -68,7 +112,7 @@ dws-release v1.2.3 --from-beta v1.2.3-beta.1
 dws-release v1.2.3 --from-beta v1.2.3-beta.1 --publish
 ```
 
-`FROM_BETA` 不会自动推断，并会写入 stable annotated tag 的 `From-Beta` 元数据，CI 会再次读取和验证。
+本地入口的 `FROM_BETA` 不会自动推断；云端入口会按上述规则唯一选择。两种入口都会把它写入 stable annotated tag 的 `From-Beta` 元数据，CI 会再次读取和验证。
 
 ## CHANGELOG 契约
 
@@ -86,49 +130,66 @@ dws-release v1.2.3 --from-beta v1.2.3-beta.1 --publish
 
 ## 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，要求先重跑补齐。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；验证仍要求 release、Darwin 签名和最终发布三个 job 全部成功，不能接受任意 workflow_dispatch。
-- tag 必须是 annotated tag；本地脚本在推送前重新确认 HEAD 与远端 `main` 完全一致，CI 允许其后 `main` 前进，但要求封板提交仍位于 `main` 历史中。
+- 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，要求走受保护恢复补齐。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 必须是 annotated tag；本地脚本要求封板提交已通过 PR 合入并包含在远端 `main` 历史中，发布只推送 tag。CI 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`，更新 OSS `latest.txt` 和共享安装脚本；prerelease 发布到 npm `beta`，只更新 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；本地入口仍要求上一条 Release 完成后才能封下一个 tag。
-- 本地 tag push 失败时会删除本次新建的本地 tag。tag 一旦成功推送，后续发布归 CI 所有，禁止改 tag 指向或复用版本号。
+- 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。
+- 本地 tag push 失败时会删除本次新建的本地 tag。远端 tag 一旦创建，后续发布归 CI 所有；发布中途失败时走受保护恢复，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。
 
-OSS/Gitee 分发失败时直接重跑该 tag 的 `Publish npm and mirrors` failed job；各步会复用 immutable GitHub 资产并保持 channel 单调。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。
+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
 
-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：
+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 和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：
+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 exact-tag `Release` run；commit 必须仍在 `main` 历史中。
-- 目标只允许不存在 GitHub Release 或仍为 Draft；已经公开的版本只能走对应的 channel repair，不能全量重建。
+- 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。
 - `release-recovery` environment 必须限制为受保护分支、配置至少一名 required reviewer，并禁止自审；workflow 会通过 API 复核这些设置，未配置时 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 和 OSS jobs，不存在 recovery 专用 publisher 或门禁跳过。
+- 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。
 
-OSS 的 `latest.txt` / `beta.txt` 当前是镜像频道元数据；仓库内安装器仍从 GitHub/Gitee 解析版本，不能把 OSS pointer 当成已接入的安装通道。
+云端 seal 后不要使用 GitHub 的 “Re-run failed jobs” 作为交付修复：annotated tag 永久绑定最初的 run attempt，普通 rerun 不会成为可接受的交付证据。GitHub Release 尚未公开时走上述 protected 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。
 
-Homebrew 当前只属于本机预检/手工公式通道：预检会在当前 macOS 架构真实安装，但 Release workflow 不发布 tap，CI 生成的单主机公式也不应当作 Darwin 双架构正式交付。正式自动交付范围是 GitHub Release、npm、OSS，以及显式开启时的 Gitee fallback；Homebrew 双架构 tap 发布需另立需求。
+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，并分别为 stable/beta 打开 Homebrew PR；tap 的默认分支仍以独立审核合入为交付边界。撤回 workflow 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` 必须要求精确的 `CI Gate`；tag workflow 也会通过 Checks API 再确认该封板 SHA 已通过。
+- `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；tag 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 tag workflow 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。
 - 单独配置 `HOMEBREW_PR_TOKEN`，优先使用仅授权本仓库且具备 `Contents: write`、`Pull requests: write` 的 fine-grained PAT；若组织策略不允许该账号使用 fine-grained PAT，则回退到仅带 `public_repo` scope 的专用 classic PAT。治理预检和 tag contract 会验证 token 身份、classic scope，并用 `[skip ci]` 临时分支和 draft PR 完成真实写权限 canary，随后立即关闭 PR、删除分支；任何清理失败都会 fail closed。门禁也会拒绝与治理 token 复用。
 - 创建 `release-recovery` environment，只允许受保护分支，设置 required reviewer、禁止自审并关闭管理员绕过。workflow 会读取 environment 的 required-reviewer、prevent-self-review 和 protected-branch 规则；规则缺失时紧急恢复会失败，正常 beta/stable tag 发布不受影响。
+- 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 `contents: write`。若上述发布凭证采用 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 `CI Gate` 缺失时，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。
+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/schema-contributor-guide.md

@@ -0,0 +1,150 @@
+# Schema and Agent Contract Contributor Guide
+
+Read this guide when changing public CLI commands, Schema identity or
+parameters, Agent selection/safety metadata, or generated Schema assets.
+
+## Ownership and data flow
+
+The publication graph is one way:
+
+```text
+Cobra command tree
+  + reviewed CommandRegistry identity/navigation
+  + reviewed metadata parameter overlays
+  -> EffectiveCommandRegistry and executable binding
+  + parameter bindings
+  + reviewed selection and safety/interface metadata
+  + pinned MCP metadata
+  -> one resolved ToolSpec registry/index
+  -> generated Agent metadata and embedded Schema Catalog
+  -> dws schema projections and runtime metadata lookup
+```
+
+The owning sources are:
+
+| Concern | Authoritative input |
+|---|---|
+| Executable paths and accepted flags | Cobra tree built by `app.NewRootCommand()` |
+| Stable canonical identity, primary path, aliases, navigation | `internal/cli/schema_command_registry.json` |
+| Registry editing contract | `internal/cli/schema_command_registry.schema.json` |
+| Safety, interface, runtime gates, parameter overlays | `internal/cli/schema_hints/metadata/<product>.json` |
+| Agent selection prose and examples | `internal/cli/schema_hints/selection/<product>.json` |
+| Product-to-hint-file routing | `internal/cli/schema_hints/index.json` |
+| Flag/property bindings | `internal/cli/schema_parameter_bindings.json` |
+| Sanitized interface fallback | `internal/cli/schema_mcp_metadata.json` |
+| Exact reviewed omissions from Schema | `internal/cli/schema_command_exclusions.json` |
+
+Generated files under `internal/cli/schema_agent_metadata/` and
+`internal/cli/schema_catalog.json` are output only. Runtime loading is a delivery
+boundary: it must not create or repair commands, flags, registry entries, or
+generation inputs.
+
+## Invariants
+
+1. Every delivered tool resolves to a public runnable Cobra leaf.
+2. Every public runnable Cobra leaf resolves to Schema or has one exact,
+   reviewed exclusion with a non-empty reason. Wildcard/prefix exclusions are
+   forbidden.
+3. The reviewed CommandRegistry is the only stable identity/navigation source.
+   Native annotations, when present, are consistency assertions and must agree.
+4. Metadata overlays may describe or constrain real flags; they cannot create
+   commands, flags, interfaces, or unknown RPCs.
+5. Cobra-required flags are a hard floor. An overlay may make an optional flag
+   required but cannot make a Cobra-required flag optional.
+6. Each tool is resolved once into one typed `ToolSpec`; all Catalog, `schema
+   --all`, leaf, summary, safety, and runtime projections derive from it.
+7. Provenance winner values must equal delivered values. Same-precedence
+   conflicts fail instead of being merged silently.
+8. `confirmation=user_required` requires user confirmation before `--yes`.
+   Do not infer confirmation mechanically from risk/effect; keep runtime gates
+   and published metadata consistent.
+9. Stored examples use real primary/alias paths and accepted flags, satisfy all
+   required/constraint rules, contain no shell comments, and never add `--yes`.
+10. `schema --all` remains the complete compatibility export. Routine discovery
+    should use overview, product/group, then leaf queries.
+
+When Help and shipped Schema disagree, treat it as contract drift. Cobra still
+defines executable flags; use the safer interpretation for confirmation or
+stop rather than guessing.
+
+Parameter and safety resolution is source-precedence based and otherwise
+value-neutral: a value must not win merely because it looks stricter. Preserve
+all candidates and the selected source, and fail same-precedence conflicts.
+Command text resolves from reviewed tool hints, then command-specific Cobra
+help, then MCP metadata; generic RPC prose must not replace a specialized
+leaf's description. An alias lookup may change only view fields such as
+`cli_path` and `is_alias`, never the resolved command contract.
+
+## Editing workflow
+
+1. Confirm the live path and flags in the Cobra tree and current `--help`.
+2. Change only the owning reviewed block. Do not copy generated Catalog fields
+   into inputs or mix selection fields into metadata files.
+3. Keep registry edits limited to intentional identity/navigation changes.
+4. For selection prose, write decision-oriented routing: when to choose the
+   command, when a sibling is better, and the result shape. Do not restate help.
+5. For parameter overlays, use an exact runnable leaf and real flags; set
+   `reviewed: true` with a concrete review reason.
+6. Regenerate the complete snapshot; publication is deterministic even when
+   only one product input changed.
+7. Inspect authored and generated diffs separately, then run the gates below.
+
+Pinned MCP metadata is a sanitized fallback. When a task requires refreshing
+it and a personal session is available, inspect live metadata with `dws auth
+status`, `dws cache refresh`, and `dws schema <canonical> -f json`. Never print
+or commit tokens. Evidence precedence is Runtime/Cobra, live MCP, pinned MCP,
+then Skill prose as evidence only.
+
+## Required checks
+
+```bash
+make generate-schema
+./scripts/policy/check-runtime-confirmation-truth.sh
+./scripts/policy/check-generated-drift.sh
+./scripts/policy/check-schema-catalog.sh
+./scripts/policy/check-command-surface.sh --strict
+make test-schema-agent-examples
+```
+
+Also run focused tests for the changed binder, generator, command, or runtime
+consumer. Run reverse-completeness tests whenever the Cobra tree changes.
+
+Agent examples are contract-checked by default. Eligible reviewed dry-run
+examples are additionally exercised by `make test-schema-agent-examples` with
+isolated state; a runtime failure must not be converted into an ad hoc skip.
+Live-model selection evaluation is optional and never a normal CI dependency.
+
+An example enters runtime dry-run only when its final typed contract publishes
+an explicit reviewed dry-run capability. Risk or confirmation metadata does
+not manufacture preview support, and the harness never injects `--yes`. A
+narrow precondition that cannot be derived from the contract may use an exact,
+reviewed `example_dispositions` entry to narrow dry-run to contract-only; it
+must not become a general skip or a fallback applied after execution fails.
+
+Every `use_when` entry is a positive selection fixture and every `avoid_when`
+entry is a negative fixture for that tool. The deterministic gate checks
+coverage and contradictions; it does not claim to prove natural-language
+understanding. When explicitly requested, the optional live-model smoke test
+can be run with:
+
+```bash
+DWS_AGENT_SELECTION_LIVE=1 \
+ARK_API_KEY=... ARK_BASE_URL=... ARK_MODEL=... \
+go test ./internal/app -run TestManualAgentSelectionArkLive -count=1
+```
+
+Use `DWS_AGENT_SELECTION_FULL=1` for the full fixture or
+`DWS_AGENT_SELECTION_CASES=<comma-separated-ids>` for selected cases. A custom
+HTTPS provider must be explicitly allowlisted; plaintext is accepted only for
+a loopback test server so credentials are not sent over arbitrary clear text.
+
+## Runtime boundaries
+
+- `schema list` is a progressive overview; `schema --all` is the complete,
+  non-compact compatibility baseline with full parameters, constraints, and
+  safety semantics.
+- `--compact` saves discovery context but is not a full compatibility export.
+- `dws <path> --help` decides whether a path and its flags are executable. Leaf
+  Schema owns Agent selection, mapping, constraints, and safety semantics.
+- Help and Schema describe commands; they do not return DingTalk business
+  data. Execute the real read/list/search command after discovery.

docs/skill-authoring-guide.md

@@ -0,0 +1,75 @@
+# Skill Authoring Guide
+
+Contract for the bundled agent skills under `skills/`. Skills are embedded via
+`skills/embed.go` and installed by `dws skill setup`, so every edit ships with
+the binary. Keep skill prose concise: long command references belong in
+`references/`, not in `SKILL.md`.
+
+## Layout
+
+| Path | Role |
+|---|---|
+| `skills/mono/` | Single bundled skill (stable mode) with shared `references/` and `scripts/` |
+| `skills/multi/dingtalk-<product>/` | One skill per product (experimental mode) |
+| `skills/multi/dws-shared/` | Shared prerequisite: auth, global flags, routing, safety |
+| `skills/embed.go` | Embeds `mono` + `multi`; do not add new roots |
+
+A product skill directory contains:
+
+```text
+dingtalk-<product>/
+  SKILL.md          # frontmatter + concise routing/usage prose
+  references/       # long command references, playbooks
+  scripts/          # executable recipes (python), kept minimal
+```
+
+## SKILL.md contract
+
+Frontmatter:
+
+```yaml
+---
+name: dingtalk-<product>
+description: <触发场景>. Use when … Distinct from <相邻skill>(…). 命令前缀：dws <product>。
+cli_version: ">=<minimum dws version>"
+metadata:
+  category: product
+  stability: experimental
+  requires:
+    bins:
+      - dws
+---
+```
+
+Body rules:
+
+- State the safety rules directly in concise prose; there is no injected
+  preamble mechanism in this repository.
+- State the `dws-shared` prerequisite for multi skills.
+- Route by intent: shortcuts table first when one covers the scenario, then
+  scripts/recipes, then atomic commands with `dws schema` / `--help`.
+- Every referenced `dws` command must exist in the current binary; verify with
+  `dws <cmd> --help` and keep prose version-agnostic ("以当前 dws 二进制为准").
+- Mutating commands must point at the leaf Schema `confirmation` contract; do
+  not invent confirmation rules in prose.
+
+## Dual-write rule
+
+Skill behavior described in prose must match the CLI it references. When a
+command, flag, or confirmation contract changes, update the affected `SKILL.md`
+/ `references/` in the same change; when skill routing changes, check whether
+Schema selection hints (`internal/cli/schema_hints/`) need a reviewed update
+per [`schema-contributor-guide.md`](schema-contributor-guide.md).
+
+## Validation
+
+| Change | Check |
+|---|---|
+| Any skill edit | `make skill-command-integrity` |
+| Referenced CLI surface changed | `./scripts/policy/check-command-surface.sh --strict` |
+| Schema hints touched | `make generate-schema` + schema gates |
+| Recipe scripts | run the script's own smoke path or `test/skill_e2e` when applicable |
+
+`make skill-command-integrity` builds `scripts/policy/skill-command-check` and
+verifies every `dws` command referenced by skills resolves against the current
+binary. Run it before handoff; do not claim a command exists without it.

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

@@ -21,19 +21,61 @@ 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"
 )
 
+const accessTokenRefreshWindow = 5 * time.Minute
+
 type legacyTokenGetter interface {
 	GetToken() (string, string, error)
 }
 
+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 {
-		disc := slog.New(slog.NewTextHandler(io.Discard, nil))
-		provider := authpkg.NewOAuthProvider(configDir, disc)
+		discard := slog.New(slog.NewTextHandler(io.Discard, nil))
+		provider := authpkg.NewOAuthProvider(configDir, discard)
 		configureOAuthProviderCompatibility(provider, configDir)
 		return provider
 	}
@@ -44,64 +86,218 @@ var (
 	}
 )
 
-// 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) {
-	provider := newAccessTokenProvider(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
-	}
-	if strings.TrimSpace(authpkg.RuntimeProfile()) != "" {
-		if tokenErr != nil {
-			return "", tokenErr
-		}
-		return "", nil
-	}
-	manager := newLegacyTokenManager(configDir)
-	if leg, _, err := manager.GetToken(); err == nil && strings.TrimSpace(leg) != "" {
-		return strings.TrimSpace(leg), nil
-	}
-	if tokenErr != nil {
-		return "", tokenErr
-	}
-	return "", nil
-}
-
-// 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
+// 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/app_small_coverage_test.go

@@ -34,6 +34,10 @@ 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
@@ -212,14 +216,16 @@ func TestCrossPlatformCoverageConfigAndTokenSeamsCoverage(t *testing.T) {
 	if _, err := resolveAccessTokenFromDir(context.Background(), "unused"); !errors.Is(err, authpkg.ErrTokenDecryption) {
 		t.Fatalf("decryption error = %v", err)
 	}
-	newAccessTokenProvider = func(string) accessTokenGetter { return fakeAccessTokenGetter{err: errors.New("missing")} }
+	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 != "" || err == nil || err.Error() != "missing" {
+	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("")
@@ -246,10 +252,10 @@ func TestCrossPlatformCoverageConfigAndTokenSeamsCoverage(t *testing.T) {
 }
 
 func TestCrossPlatformCoverageForceRefreshAndStdioFailureCoverage(t *testing.T) {
-	oldMark, oldFactory := markAccessTokenStale, newRefreshProvider
+	oldLoad, oldFactory := loadRefreshTokenData, newRefreshProvider
 	oldStop := stopStdio
 	t.Cleanup(func() {
-		markAccessTokenStale, newRefreshProvider = oldMark, oldFactory
+		loadRefreshTokenData, newRefreshProvider = oldLoad, oldFactory
 		stopStdio = oldStop
 		stdioMu.Lock()
 		stdioClients = make(map[string]*transport.StdioClient)
@@ -257,11 +263,13 @@ func TestCrossPlatformCoverageForceRefreshAndStdioFailureCoverage(t *testing.T)
 	})
 	fail := errors.New("failure")
 	_ = oldFactory(t.TempDir())
-	markAccessTokenStale = func(string) error { return fail }
+	loadRefreshTokenData = func(string) (*authpkg.TokenData, error) { return nil, fail }
 	if _, err := ForceRefreshAccessToken(context.Background(), "config"); !errors.Is(err, fail) {
-		t.Fatalf("mark stale error = %v", err)
+		t.Fatalf("load rejected token error = %v", err)
+	}
+	loadRefreshTokenData = func(string) (*authpkg.TokenData, error) {
+		return &authpkg.TokenData{AccessToken: "rejected"}, nil
 	}
-	markAccessTokenStale = func(string) error { return nil }
 	for _, tc := range []struct {
 		getter fakeAccessTokenGetter
 		want   string
@@ -270,7 +278,7 @@ func TestCrossPlatformCoverageForceRefreshAndStdioFailureCoverage(t *testing.T)
 		{getter: fakeAccessTokenGetter{token: "  "}, want: "empty"},
 		{getter: fakeAccessTokenGetter{token: " refreshed "}},
 	} {
-		newRefreshProvider = func(string) accessTokenGetter { return tc.getter }
+		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)

internal/app/auth_refresh_retry.go

@@ -15,6 +15,15 @@ package app
 
 import (
 	"context"
+	"errors"
+	"fmt"
+	"log/slog"
+	"strings"
+
+	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/pkg/authretry"
 )
 
 // authRetryingKey marks a context that has already attempted one
@@ -23,8 +32,35 @@ import (
 // to the user instead.
 type authRetryingKeyType struct{}
 
+type authRefreshFailureError struct {
+	rejection error
+	refresh   error
+}
+
+func (e *authRefreshFailureError) Error() string {
+	return "automatic access token refresh failed"
+}
+
+func (e *authRefreshFailureError) Unwrap() []error {
+	if e == nil {
+		return nil
+	}
+	return []error{e.rejection, e.refresh}
+}
+
 var authRetryingKey = authRetryingKeyType{}
 
+var (
+	runnerForceRefreshRejectedAccessToken = forceRefreshRejectedAccessToken
+	runnerExecuteAuthRetry                func(*runtimeRunner, context.Context, string, executor.Invocation) (executor.Result, error)
+)
+
+func init() {
+	runnerExecuteAuthRetry = func(r *runtimeRunner, ctx context.Context, endpoint string, invocation executor.Invocation) (executor.Result, error) {
+		return r.executeInvocation(ctx, endpoint, invocation)
+	}
+}
+
 // IsAuthRetrying reports whether the current context is already inside an
 // AuthRefreshRequired retry. Mirrors IsPatRetrying.
 func IsAuthRetrying(ctx context.Context) bool {
@@ -34,3 +70,103 @@ func IsAuthRetrying(ctx context.Context) bool {
 	v, _ := ctx.Value(authRetryingKey).(bool)
 	return v
 }
+
+func withAuthRetrying(ctx context.Context) context.Context {
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	return context.WithValue(ctx, authRetryingKey, true)
+}
+
+func authRefreshLogger() *slog.Logger {
+	if logger := FileLoggerInstance(); logger != nil {
+		return logger
+	}
+	return slog.Default()
+}
+
+func (r *runtimeRunner) managesRuntimeOAuth(hasPluginAuth bool) bool {
+	if r == nil || hasPluginAuth {
+		return false
+	}
+	return r.globalFlags == nil || strings.TrimSpace(r.globalFlags.Token) == ""
+}
+
+// retryAuthRefreshRequired consumes only the explicit edition marker. It does
+// not infer retryability from free text, generic auth categories, HTTP 403, or
+// ordinary business errors.
+func (r *runtimeRunner) retryAuthRefreshRequired(
+	ctx context.Context,
+	endpoint string,
+	invocation executor.Invocation,
+	rejectedAccessToken string,
+	markerErr error,
+	hasPluginAuth bool,
+) (executor.Result, error, bool) {
+	marker, marked := authretry.As(markerErr)
+	if !marked {
+		return executor.Result{}, nil, false
+	}
+	cause := marker.Cause
+	if cause == nil {
+		cause = markerErr
+	}
+
+	// Explicit --token and plugin credentials are not backed by the default
+	// OAuth refresh store. Preserve the overlay cause without mutating an
+	// unrelated persisted login.
+	if !r.managesRuntimeOAuth(hasPluginAuth) {
+		return executor.Result{}, cause, true
+	}
+	if IsAuthRetrying(ctx) {
+		authRefreshLogger().Warn("auth.runtime.refresh.retry_exhausted",
+			"product", invocation.CanonicalProduct,
+			"tool", invocation.Tool,
+		)
+		return executor.Result{}, cause, true
+	}
+
+	if _, err := runnerForceRefreshRejectedAccessToken(ctx, defaultConfigDir(), rejectedAccessToken); err != nil {
+		// Keep every log credential-safe. The returned error chain retains the
+		// complete cause for in-process diagnosis; even DWS_DEBUG_AUTH must not
+		// serialize an OAuth response body or other attacker-controlled text.
+		authRefreshLogger().Warn("auth.runtime.refresh.failed",
+			"product", invocation.CanonicalProduct,
+			"tool", invocation.Tool,
+			"stage", "force_refresh_rejected_token",
+			"error_type", fmt.Sprintf("%T", err),
+		)
+		logging.AuthDebug("auth.runtime.refresh.failed.detail",
+			"product", invocation.CanonicalProduct,
+			"tool", invocation.Tool,
+			"stage", "force_refresh_rejected_token",
+			"error_type", fmt.Sprintf("%T", err),
+		)
+		combined := &authRefreshFailureError{rejection: cause, refresh: err}
+		return executor.Result{}, apperrors.NewAuth(
+			"automatic access token refresh failed",
+			apperrors.WithOperation("auth/token/refresh"),
+			apperrors.WithReason("auth_refresh_failed"),
+			apperrors.WithHint("本地凭证已保留；可稍后重试，若持续失败请查看认证诊断日志。"),
+			apperrors.WithCause(combined),
+		), true
+	}
+
+	logging.AuthDebug("auth.runtime.refresh.succeeded",
+		"product", invocation.CanonicalProduct,
+		"tool", invocation.Tool,
+	)
+	result, err := runnerExecuteAuthRetry(r, withAuthRetrying(ctx), endpoint, invocation)
+	return result, err, true
+}
+
+// isRefreshableTransportAuthError deliberately excludes HTTP/RPC 403 and
+// generic CategoryAuth values. OnAuthError may request a refresh only for an
+// exact transport-level unauthorized signal.
+func isRefreshableTransportAuthError(err error) bool {
+	var typed *apperrors.Error
+	if !errors.As(err, &typed) || typed.Category != apperrors.CategoryAuth {
+		return false
+	}
+	return typed.Reason == "http_401" || typed.RPCCode == 401
+}

internal/app/auth_refresh_retry_runner_test.go

@@ -0,0 +1,354 @@
+// 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 (
+	"bytes"
+	"context"
+	"errors"
+	"log/slog"
+	"strings"
+	"testing"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/audit"
+	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/internal/executor"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/transport"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/authretry"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
+)
+
+func installAuthRefreshRunnerSeams(t *testing.T) {
+	t.Helper()
+	previousHooks := edition.Get()
+	previousCall := runnerCallTool
+	previousPreflight := runnerPreflightDocDownload
+	previousRefresh := runnerForceRefreshRejectedAccessToken
+	previousRetry := runnerExecuteAuthRetry
+	previousCapture := runnerCaptureRuntimeFailure
+	previousProfile := authpkg.RuntimeProfile()
+
+	pluginAuthMu.Lock()
+	previousPlugins := pluginAuthRegistry
+	pluginAuthRegistry = make(map[string]*PluginAuth)
+	pluginAuthMu.Unlock()
+
+	runnerPreflightDocDownload = func(*runtimeRunner, context.Context, *transport.Client, string, executor.Invocation) error {
+		return nil
+	}
+	runnerCaptureRuntimeFailure = func(executor.Invocation, error, error) {}
+	authpkg.SetRuntimeProfile("")
+	runtimeTokenManager.Invalidate()
+	t.Setenv("DWS_CONFIG_DIR", "")
+	t.Setenv("DWS_DEBUG_AUTH", "0")
+
+	t.Cleanup(func() {
+		edition.Override(previousHooks)
+		runnerCallTool = previousCall
+		runnerPreflightDocDownload = previousPreflight
+		runnerForceRefreshRejectedAccessToken = previousRefresh
+		runnerExecuteAuthRetry = previousRetry
+		runnerCaptureRuntimeFailure = previousCapture
+		authpkg.SetRuntimeProfile(previousProfile)
+		runtimeTokenManager.Invalidate()
+		pluginAuthMu.Lock()
+		pluginAuthRegistry = previousPlugins
+		pluginAuthMu.Unlock()
+	})
+}
+
+func authRefreshTestRunner(flags *GlobalFlags) *runtimeRunner {
+	return &runtimeRunner{
+		transport:   transport.NewClient(nil),
+		globalFlags: flags,
+		auditSink:   audit.NopSink{},
+	}
+}
+
+func authRefreshTestInvocation() executor.Invocation {
+	return executor.Invocation{
+		CanonicalProduct: "auth-retry-test-product",
+		Tool:             "test_tool",
+		Params:           map[string]any{"value": "safe"},
+	}
+}
+
+func authRefreshTokenHooks(configDir string, token *string, classify func(map[string]any) error) *edition.Hooks {
+	return &edition.Hooks{
+		ConfigDir: func() string { return configDir },
+		TokenProvider: func(context.Context, func() (string, error)) (string, error) {
+			return *token, nil
+		},
+		ClassifyToolResult: classify,
+	}
+}
+
+func TestCrossPlatformCoverageRunnerRetriesEditionAuthMarkerOnce(t *testing.T) {
+	installAuthRefreshRunnerSeams(t)
+	configDir := t.TempDir()
+	token := "old-access"
+	rejection := apperrors.NewAuth("server rejected access token", apperrors.WithReason("access_token_rejected"))
+	edition.Override(authRefreshTokenHooks(configDir, &token, func(content map[string]any) error {
+		if expired, _ := content["expired"].(bool); expired {
+			return &authretry.AuthRefreshRequired{Cause: rejection}
+		}
+		return nil
+	}))
+
+	var callTokens []string
+	runnerCallTool = func(client *transport.Client, _ context.Context, _, _ string, _ map[string]any) (transport.ToolCallResult, error) {
+		callTokens = append(callTokens, client.AuthToken)
+		if len(callTokens) == 1 {
+			return transport.ToolCallResult{Content: map[string]any{"expired": true}}, nil
+		}
+		return transport.ToolCallResult{Content: map[string]any{"value": "ok"}}, nil
+	}
+	refreshCalls := 0
+	runnerForceRefreshRejectedAccessToken = func(_ context.Context, gotDir, rejected string) (string, error) {
+		refreshCalls++
+		if gotDir != configDir || rejected != "old-access" {
+			t.Fatalf("refresh input = dir %q token %q", gotDir, rejected)
+		}
+		token = "new-access"
+		return token, nil
+	}
+
+	result, err := authRefreshTestRunner(nil).executeInvocation(context.Background(), "https://example.test", authRefreshTestInvocation())
+	if err != nil {
+		t.Fatal(err)
+	}
+	if refreshCalls != 1 || len(callTokens) != 2 || callTokens[0] != "old-access" || callTokens[1] != "new-access" {
+		t.Fatalf("refreshes=%d call tokens=%v", refreshCalls, callTokens)
+	}
+	content, _ := result.Response["content"].(map[string]any)
+	if content["value"] != "ok" || content["success"] != true {
+		t.Fatalf("result content = %#v", content)
+	}
+}
+
+func TestCrossPlatformCoverageRunnerRefreshFailurePreservesBothCausesAndSafeLog(t *testing.T) {
+	installAuthRefreshRunnerSeams(t)
+	t.Setenv("DWS_DEBUG_AUTH", "1")
+	configDir := t.TempDir()
+	token := "old-access"
+	rejection := apperrors.NewAuth("server rejected access token", apperrors.WithReason("access_token_rejected"))
+	edition.Override(authRefreshTokenHooks(configDir, &token, func(map[string]any) error {
+		return &authretry.AuthRefreshRequired{Cause: rejection}
+	}))
+	runnerCallTool = func(*transport.Client, context.Context, string, string, map[string]any) (transport.ToolCallResult, error) {
+		return transport.ToolCallResult{Content: map[string]any{"expired": true}}, nil
+	}
+	refreshErr := errors.New(`oauth refresh response parse failed: body={"access_token":"access-token-secret","refresh_token":"refresh-token-secret","uid":"uid-secret-value"}`)
+	runnerForceRefreshRejectedAccessToken = func(context.Context, string, string) (string, error) {
+		return "", refreshErr
+	}
+
+	var logs bytes.Buffer
+	previousLogger := slog.Default()
+	slog.SetDefault(slog.New(slog.NewJSONHandler(&logs, &slog.HandlerOptions{Level: slog.LevelDebug})))
+	t.Cleanup(func() { slog.SetDefault(previousLogger) })
+
+	_, err := authRefreshTestRunner(nil).executeInvocation(context.Background(), "https://example.test", authRefreshTestInvocation())
+	if !errors.Is(err, rejection) || !errors.Is(err, refreshErr) {
+		t.Fatalf("error = %v, want rejection and refresh causes", err)
+	}
+	var typed *apperrors.Error
+	if !errors.As(err, &typed) || typed.Category != apperrors.CategoryAuth || typed.Reason != "auth_refresh_failed" || typed.Operation != "auth/token/refresh" {
+		t.Fatalf("refresh envelope = %#v", typed)
+	}
+	var rendered bytes.Buffer
+	if printErr := apperrors.PrintJSON(&rendered, err); printErr != nil {
+		t.Fatal(printErr)
+	}
+	for _, want := range []string{`"category": "auth"`, `"reason": "auth_refresh_failed"`, `"operation": "auth/token/refresh"`} {
+		if !strings.Contains(rendered.String(), want) {
+			t.Fatalf("structured stderr missing %s: %s", want, rendered.String())
+		}
+	}
+	for _, secret := range []string{"access-token-secret", "refresh-token-secret", "uid-secret-value"} {
+		if strings.Contains(err.Error(), secret) || strings.Contains(logs.String(), secret) || strings.Contains(rendered.String(), secret) {
+			t.Fatalf("auth output leaked %q: error=%q logs=%s stderr=%s", secret, err, logs.String(), rendered.String())
+		}
+	}
+	for _, want := range []string{"auth.runtime.refresh.failed", "auth.runtime.refresh.failed.detail", "force_refresh_rejected_token", "error_type"} {
+		if !strings.Contains(logs.String(), want) {
+			t.Fatalf("safe refresh log missing %q: %s", want, logs.String())
+		}
+	}
+}
+
+func TestCrossPlatformCoverageRunnerSecondEditionMarkerReturnsSecondCause(t *testing.T) {
+	installAuthRefreshRunnerSeams(t)
+	configDir := t.TempDir()
+	token := "old-access"
+	firstCause := errors.New("first rejection")
+	secondCause := errors.New("second rejection")
+	edition.Override(authRefreshTokenHooks(configDir, &token, func(content map[string]any) error {
+		attempt, _ := content["attempt"].(int)
+		if attempt == 1 {
+			return &authretry.AuthRefreshRequired{Cause: firstCause}
+		}
+		return &authretry.AuthRefreshRequired{Cause: secondCause}
+	}))
+	calls := 0
+	runnerCallTool = func(*transport.Client, context.Context, string, string, map[string]any) (transport.ToolCallResult, error) {
+		calls++
+		return transport.ToolCallResult{Content: map[string]any{"attempt": calls}}, nil
+	}
+	refreshCalls := 0
+	runnerForceRefreshRejectedAccessToken = func(context.Context, string, string) (string, error) {
+		refreshCalls++
+		token = "new-access"
+		return token, nil
+	}
+
+	_, err := authRefreshTestRunner(nil).executeInvocation(context.Background(), "https://example.test", authRefreshTestInvocation())
+	if !errors.Is(err, secondCause) || errors.Is(err, firstCause) {
+		t.Fatalf("error = %v, want only second rejection cause", err)
+	}
+	if calls != 2 || refreshCalls != 1 {
+		t.Fatalf("calls=%d refreshes=%d", calls, refreshCalls)
+	}
+}
+
+func TestCrossPlatformCoverageRunnerOnAuthErrorOnlyRetriesExactUnauthorized(t *testing.T) {
+	t.Run("http 401 marker retries once", func(t *testing.T) {
+		installAuthRefreshRunnerSeams(t)
+		configDir := t.TempDir()
+		token := "old-access"
+		rejection := errors.New("transport rejected token")
+		hookCalls := 0
+		hooks := authRefreshTokenHooks(configDir, &token, nil)
+		hooks.OnAuthError = func(string, error) error {
+			hookCalls++
+			return &authretry.AuthRefreshRequired{Cause: rejection}
+		}
+		edition.Override(hooks)
+		calls := 0
+		var callTokens []string
+		runnerCallTool = func(client *transport.Client, _ context.Context, _, _ string, _ map[string]any) (transport.ToolCallResult, error) {
+			calls++
+			callTokens = append(callTokens, client.AuthToken)
+			if calls == 1 {
+				return transport.ToolCallResult{}, apperrors.NewAuth("unauthorized", apperrors.WithReason("http_401"))
+			}
+			return transport.ToolCallResult{Content: map[string]any{"value": "ok"}}, nil
+		}
+		refreshCalls := 0
+		runnerForceRefreshRejectedAccessToken = func(context.Context, string, string) (string, error) {
+			refreshCalls++
+			token = "new-access"
+			return token, nil
+		}
+
+		if _, err := authRefreshTestRunner(nil).executeInvocation(context.Background(), "https://example.test", authRefreshTestInvocation()); err != nil {
+			t.Fatal(err)
+		}
+		if hookCalls != 1 || refreshCalls != 1 || calls != 2 || strings.Join(callTokens, ",") != "old-access,new-access" {
+			t.Fatalf("hook=%d refresh=%d calls=%d tokens=%v", hookCalls, refreshCalls, calls, callTokens)
+		}
+	})
+
+	for _, tc := range []struct {
+		name string
+		err  error
+	}{
+		{name: "http 403", err: apperrors.NewAuth("forbidden", apperrors.WithReason("http_403"))},
+		{name: "ordinary auth", err: apperrors.NewAuth("load failed", apperrors.WithReason("auth_load_failed"))},
+	} {
+		t.Run(tc.name+" does not enter hook", func(t *testing.T) {
+			installAuthRefreshRunnerSeams(t)
+			configDir := t.TempDir()
+			token := "old-access"
+			hookCalls := 0
+			hooks := authRefreshTokenHooks(configDir, &token, nil)
+			hooks.OnAuthError = func(string, error) error {
+				hookCalls++
+				return &authretry.AuthRefreshRequired{Cause: errors.New("must not run")}
+			}
+			edition.Override(hooks)
+			runnerCallTool = func(*transport.Client, context.Context, string, string, map[string]any) (transport.ToolCallResult, error) {
+				return transport.ToolCallResult{}, tc.err
+			}
+			refreshCalls := 0
+			runnerForceRefreshRejectedAccessToken = func(context.Context, string, string) (string, error) {
+				refreshCalls++
+				return "", nil
+			}
+
+			_, err := authRefreshTestRunner(nil).executeInvocation(context.Background(), "https://example.test", authRefreshTestInvocation())
+			if !errors.Is(err, tc.err) || hookCalls != 0 || refreshCalls != 0 {
+				t.Fatalf("error=%v hook=%d refresh=%d", err, hookCalls, refreshCalls)
+			}
+		})
+	}
+}
+
+func TestCrossPlatformCoverageRunnerDoesNotRefreshExplicitTokenMarker(t *testing.T) {
+	installAuthRefreshRunnerSeams(t)
+	rejection := errors.New("explicit token rejected")
+	edition.Override(&edition.Hooks{ClassifyToolResult: func(map[string]any) error {
+		return &authretry.AuthRefreshRequired{Cause: rejection}
+	}})
+	calls := 0
+	runnerCallTool = func(*transport.Client, context.Context, string, string, map[string]any) (transport.ToolCallResult, error) {
+		calls++
+		return transport.ToolCallResult{Content: map[string]any{"expired": true}}, nil
+	}
+	refreshCalls := 0
+	runnerForceRefreshRejectedAccessToken = func(context.Context, string, string) (string, error) {
+		refreshCalls++
+		return "", nil
+	}
+
+	_, err := authRefreshTestRunner(&GlobalFlags{Token: "explicit-token"}).executeInvocation(context.Background(), "https://example.test", authRefreshTestInvocation())
+	if !errors.Is(err, rejection) || calls != 1 || refreshCalls != 0 {
+		t.Fatalf("error=%v calls=%d refresh=%d", err, calls, refreshCalls)
+	}
+}
+
+func TestCrossPlatformCoverageRunnerRetriesPreflightEditionMarkerOnce(t *testing.T) {
+	installAuthRefreshRunnerSeams(t)
+	configDir := t.TempDir()
+	token := "old-access"
+	rejection := errors.New("preflight token rejected")
+	edition.Override(authRefreshTokenHooks(configDir, &token, nil))
+	preflightCalls := 0
+	runnerPreflightDocDownload = func(*runtimeRunner, context.Context, *transport.Client, string, executor.Invocation) error {
+		preflightCalls++
+		if preflightCalls == 1 {
+			return &authretry.AuthRefreshRequired{Cause: rejection}
+		}
+		return nil
+	}
+	toolCalls := 0
+	runnerCallTool = func(*transport.Client, context.Context, string, string, map[string]any) (transport.ToolCallResult, error) {
+		toolCalls++
+		return transport.ToolCallResult{Content: map[string]any{"value": "ok"}}, nil
+	}
+	refreshCalls := 0
+	runnerForceRefreshRejectedAccessToken = func(context.Context, string, string) (string, error) {
+		refreshCalls++
+		token = "new-access"
+		return token, nil
+	}
+
+	if _, err := authRefreshTestRunner(nil).executeInvocation(context.Background(), "https://example.test", authRefreshTestInvocation()); err != nil {
+		t.Fatal(err)
+	}
+	if preflightCalls != 2 || toolCalls != 1 || refreshCalls != 1 {
+		t.Fatalf("preflights=%d tools=%d refreshes=%d", preflightCalls, toolCalls, refreshCalls)
+	}
+}

internal/app/auth_registry.go

@@ -49,6 +49,15 @@ func RegisterPluginAuth(productID string, auth *PluginAuth) {
 	pluginAuthRegistry[productID] = auth
 }
 
+// ClearPluginAuth removes credentials for a plugin product. Registration uses
+// this before applying an accepted descriptor so a descriptor without custom
+// auth cannot inherit stale credentials from an earlier root construction.
+func ClearPluginAuth(productID string) {
+	pluginAuthMu.Lock()
+	defer pluginAuthMu.Unlock()
+	delete(pluginAuthRegistry, productID)
+}
+
 // LookupPluginAuth returns the authentication credentials registered
 // for the given product ID, or nil if none exists.
 func LookupPluginAuth(productID string) (*PluginAuth, bool) {

internal/app/coverage_edges_test.go

@@ -554,7 +554,7 @@ func TestCrossPlatformCoverageRecoveryRuntimeHTTP(t *testing.T) {
 	defer server.Close()
 	SetDynamicServers([]mcptypes.ServerDescriptor{{Endpoint: server.URL, CLI: mcptypes.CLIOverlay{ID: "devdoc", Tools: []mcptypes.CLITool{{Name: "search_open_platform_docs_rag"}}}}})
 	t.Cleanup(func() { SetDynamicServers(nil) })
-	runtime := &recoveryRuntime{transport: transport.NewClient(server.Client())}
+	runtime := &recoveryRuntime{transport: transport.NewClient(server.Client()), flags: &GlobalFlags{Token: "token"}}
 	got, err := runtime.Search(context.Background(), "query", recovery.RecoveryContext{ToolName: "search"})
 	if err != nil || got.DocSearch.Status != "success" || len(got.KBHits) == 0 {
 		t.Fatalf("recovery search = %#v %v", got, err)
@@ -1469,10 +1469,10 @@ func TestCrossPlatformCoveragePersonalEventPureCoverage(t *testing.T) {
 	if !ok {
 		t.Fatal("mention definition missing")
 	}
-	if err := renderPersonalSchema(io.Discard, def, ""); err != nil {
+	if err := renderPersonalSchema(io.Discard, def, "", false); err != nil {
 		t.Fatal(err)
 	}
-	if err := renderPersonalSchema(io.Discard, def, "yaml"); err == nil {
+	if err := renderPersonalSchema(io.Discard, def, "yaml", true); err == nil {
 		t.Fatal("unsupported schema format succeeded")
 	}
 	for _, key := range []string{"", "unknown", personal.EventMention, personal.EventFromUser} {
@@ -1650,6 +1650,8 @@ func TestCrossPlatformCoveragePersonalSubscriptionAndSourceCoverage(t *testing.T
 }
 
 func TestCrossPlatformCoveragePersonalEventCommandRuntimeCoverage(t *testing.T) {
+	authpkg.SetRuntimeProfile("")
+	t.Cleanup(func() { authpkg.SetRuntimeProfile("") })
 	configDir := setupPersonalIdentityToken(t, &authpkg.TokenData{
 		AccessToken: "access", RefreshToken: "refresh", ExpiresAt: time.Now().Add(time.Hour),
 		CorpID: "corp", UserID: "user", ClientID: "client",

internal/app/coverage_more_test.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
+	"errors"
 	"io"
 	"net/http"
 	"net/http/httptest"
@@ -164,7 +165,7 @@ func TestCrossPlatformCoverageRawAPIAndTokenCoverage(t *testing.T) {
 	}
 	newAccessTokenProvider = func(string) accessTokenGetter { return fakeAccessTokenGetter{} }
 	missing := t.TempDir()
-	if got, err := resolveAccessTokenFromDir(context.Background(), missing); err != nil || got != "" {
+	if got, err := resolveAccessTokenFromDir(context.Background(), missing); got != "" || !errors.Is(err, authpkg.ErrTokenDataNotFound) {
 		t.Fatalf("missing access token = %q, %v", got, err)
 	}
 	if _, err := ResolveAuxiliaryAccessToken(context.Background(), missing, ""); err == nil {

internal/app/event_command.go

@@ -56,6 +56,7 @@ var (
 	eventNewEventSource        = newEventSource
 	eventNewDingtalkSource     = source.New
 	eventResolveAccessToken    = ResolveAuxiliaryAccessToken
+	eventForceRefreshRejected  = forceRefreshRejectedAccessToken
 	eventBusRun                = bus.Run
 	eventReadyFDFromEnv        = busctl.ReadyFDFromEnv
 	eventResolvePersonal       = resolvePersonalEventIdentity
@@ -111,6 +112,7 @@ func newEventConsumeCommand() *cobra.Command {
 		dryRun       bool
 		foreground   bool
 		asIdentity   string
+		flatten      bool
 		personalOpts personalConsumeOptions
 		streamOpts   eventStreamTicketOptions
 	)
@@ -126,7 +128,11 @@ func newEventConsumeCommand() *cobra.Command {
   json            每事件多行美化 JSON（必须配 --max-events 或 --duration）
   pretty          同 json，未来加颜色
   raw             仅 SDK 原始 payload，无外层封装
-  compact         扁平化 + 解析嵌套 + 抽取语义字段（Agent 友好）
+  compact         单行紧凑 JSON；不传 --flatten 时沿用原 compact processor
+
+数据结构：
+  ndjson/json/pretty 默认保持 transport envelope（type/event_type/data/headers）
+  --flatten          结构化格式输出稳定的顶层业务字段，适合 Agent / 脚本直接消费
 
 默认使用当前 OAuth 登录态自动创建/复用个人订阅并建立个人长连接；非默认组织加
 --profile。连上后 stderr 打就绪行 [event] ready，等它出现再读 stdout；停机用
@@ -143,6 +149,7 @@ SIGTERM、关 stdin，或先用 dws event stop <subscribe_id> --dry-run 预览
 			}
 			if as == "user" {
 				personalOpts.EventKey = firstArg(args)
+				personalOpts.Flatten = flatten
 				personalOpts.Common = commonConsumeOptions{
 					EventTypes: eventTypes,
 					Filter:     filter,
@@ -166,6 +173,7 @@ SIGTERM、关 stdin，或先用 dws event stop <subscribe_id> --dry-run 预览
 				return fmt.Errorf("event consume: --debug-raw-events is only supported with --as user")
 			}
 			if err := rejectChangedFlags(c, "user",
+				"flatten",
 				"subscribe-id",
 				"rule",
 				"name",
@@ -279,6 +287,8 @@ SIGTERM、关 stdin，或先用 dws event stop <subscribe_id> --dry-run 预览
 		"提示 bus 客户端期望 compact 渲染（语义透传，bus 仍按原 payload 投递）")
 	f.StringVarP(&formatRaw, "format", "f", "ndjson",
 		"输出格式 (ndjson/json/pretty/raw/compact)；事件流默认 ndjson")
+	f.BoolVar(&flatten, "flatten", false,
+		"将个人事件 transport envelope 投影为稳定的顶层业务字段")
 	f.StringVar(&outputDir, "output-dir", "",
 		"每事件写一个文件到该目录 ({type}_{id}_{ts}.json)；与 stdout 互斥")
 	f.StringArrayVar(&routesRaw, "route", nil,
@@ -413,7 +423,7 @@ func eventStreamBusID(streamOpts eventStreamTicketOptions) string {
 	return "portal-ticket-normal:" + sourceID
 }
 
-func newEventSource(ctx context.Context, configDir, clientID, clientSecret string, streamOpts eventStreamTicketOptions) (*source.DingtalkSource, error) {
+func newEventSource(_ context.Context, configDir, clientID, clientSecret string, streamOpts eventStreamTicketOptions) (*source.DingtalkSource, error) {
 	if !streamOpts.enabled() {
 		return eventNewDingtalkSource(source.Config{
 			ClientID:     clientID,
@@ -421,14 +431,6 @@ func newEventSource(ctx context.Context, configDir, clientID, clientSecret strin
 		})
 	}
 
-	token, err := eventResolveAccessToken(ctx, configDir, "")
-	if err != nil {
-		return nil, fmt.Errorf("event stream ticket: resolve user token: %w", err)
-	}
-	if strings.TrimSpace(token) == "" {
-		return nil, errors.New("event stream ticket: empty user token")
-	}
-
 	portalClientID := clientID
 	portalClientSecret := clientSecret
 	if streamOpts.usesPortalNormalMode() {
@@ -440,8 +442,13 @@ func newEventSource(ctx context.Context, configDir, clientID, clientSecret strin
 		ClientID:     portalClientID,
 		ClientSecret: portalClientSecret,
 		PortalTicket: &source.PortalTicketConfig{
-			TicketURL:    eventStreamTicketURL(streamOpts.TicketURL),
-			AccessToken:  token,
+			TicketURL: eventStreamTicketURL(streamOpts.TicketURL),
+			AccessTokenProvider: func(ctx context.Context) (string, error) {
+				return eventResolveAccessToken(ctx, configDir, "")
+			},
+			ForceRefreshToken: func(ctx context.Context, rejectedToken string) (string, error) {
+				return eventForceRefreshRejected(ctx, configDir, rejectedToken)
+			},
 			SourceID:     eventStreamSourceID(streamOpts.SourceID),
 			Mode:         streamOpts.Mode,
 			ClientID:     portalClientID,

internal/app/event_command_full_coverage_test.go

@@ -132,14 +132,18 @@ func TestCrossPlatformCoverageEventSourcesAndForegroundCoverage(t *testing.T) {
 	if _, err := newEventSource(context.Background(), "config", "client", "secret", eventStreamTicketOptions{}); err != nil {
 		t.Fatal(err)
 	}
-	eventResolveAccessToken = func(context.Context, string, string) (string, error) { return "", fail }
 	stream := eventStreamTicketOptions{Mode: "custom"}
-	if _, err := newEventSource(context.Background(), "config", "client", "secret", stream); !errors.Is(err, fail) {
-		t.Fatalf("stream token error = %v", err)
+	var captured source.Config
+	eventNewDingtalkSource = func(cfg source.Config, _ ...source.SourceOption) (*source.DingtalkSource, error) {
+		captured = cfg
+		return &source.DingtalkSource{}, nil
 	}
-	eventResolveAccessToken = func(context.Context, string, string) (string, error) { return " ", nil }
-	if _, err := newEventSource(context.Background(), "config", "client", "secret", stream); err == nil {
-		t.Fatal("empty stream token succeeded")
+	eventResolveAccessToken = func(context.Context, string, string) (string, error) { return "", fail }
+	if _, err := newEventSource(context.Background(), "config", "client", "secret", stream); err != nil {
+		t.Fatalf("stream source construction = %v", err)
+	}
+	if _, err := captured.PortalTicket.AccessTokenProvider(context.Background()); !errors.Is(err, fail) {
+		t.Fatalf("stream token provider error = %v", err)
 	}
 	eventResolveAccessToken = func(context.Context, string, string) (string, error) { return "token", nil }
 	for _, mode := range []string{"custom", "normal"} {

internal/app/event_force_refresh_wire_test.go

@@ -0,0 +1,47 @@
+package app
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/event/source"
+)
+
+// TestCrossPlatformCoverageNewEventSourceWiresForceRefreshRejectedToken asserts the portal ticket
+// source receives a ForceRefreshToken callback that forwards the actual
+// rejected token into the app-level compare-and-refresh chain.
+func TestCrossPlatformCoverageNewEventSourceWiresForceRefreshRejectedToken(t *testing.T) {
+	oldNew, oldRefresh := eventNewDingtalkSource, eventForceRefreshRejected
+	t.Cleanup(func() { eventNewDingtalkSource, eventForceRefreshRejected = oldNew, oldRefresh })
+
+	var captured source.Config
+	eventNewDingtalkSource = func(cfg source.Config, _ ...source.SourceOption) (*source.DingtalkSource, error) {
+		captured = cfg
+		return &source.DingtalkSource{}, nil
+	}
+	var gotDir, gotRejected string
+	eventForceRefreshRejected = func(_ context.Context, configDir, rejectedToken string) (string, error) {
+		gotDir, gotRejected = configDir, rejectedToken
+		return "fresh", nil
+	}
+	if _, err := newEventSource(context.Background(), "config-dir", "client", "secret", eventStreamTicketOptions{Mode: "custom"}); err != nil {
+		t.Fatal(err)
+	}
+	if captured.PortalTicket == nil || captured.PortalTicket.ForceRefreshToken == nil {
+		t.Fatal("ForceRefreshToken not wired into portal ticket config")
+	}
+	tok, err := captured.PortalTicket.ForceRefreshToken(context.Background(), "rejected-token")
+	if err != nil || tok != "fresh" {
+		t.Fatalf("force refresh = %q, %v", tok, err)
+	}
+	if gotDir != "config-dir" || gotRejected != "rejected-token" {
+		t.Fatalf("wiring passed dir %q rejected %q", gotDir, gotRejected)
+	}
+
+	fail := errors.New("refresh failed")
+	eventForceRefreshRejected = func(context.Context, string, string) (string, error) { return "", fail }
+	if _, err := captured.PortalTicket.ForceRefreshToken(context.Background(), "x"); !errors.Is(err, fail) {
+		t.Fatalf("refresh error = %v", err)
+	}
+}

internal/app/event_personal_command.go

@@ -61,6 +61,7 @@ type commonConsumeOptions struct {
 type personalConsumeOptions struct {
 	Common           commonConsumeOptions
 	EventKey         string
+	Flatten          bool
 	DebugRawEvents   bool
 	SubscribeID      string
 	Rule             string
@@ -131,6 +132,7 @@ var (
 	personalFindProcess                 = os.FindProcess
 	personalSignalProcess               = (*os.Process).Signal
 	personalResolveAuxiliaryAccessToken = ResolveAuxiliaryAccessToken
+	personalForceRefreshRejectedToken   = forceRefreshRejectedAccessToken
 	personalLoadTokenData               = authpkg.LoadTokenData
 	personalClientID                    = authpkg.ClientID
 	personalResolveAppCredentialsStrict = authpkg.ResolveAppCredentialsStrict
@@ -139,6 +141,7 @@ var (
 func newEventSchemaCommand() *cobra.Command {
 	var asIdentity string
 	var formatRaw string
+	var flatten bool
 	cmd := &cobra.Command{
 		Use:               "schema <event_key>",
 		Short:             "显示事件 schema",
@@ -156,11 +159,12 @@ func newEventSchemaCommand() *cobra.Command {
 			if !def.Public {
 				return personal.PublicAvailabilityError(args[0])
 			}
-			return renderPersonalSchema(c.OutOrStdout(), def, formatRaw)
+			return renderPersonalSchema(c.OutOrStdout(), def, formatRaw, flatten)
 		},
 	}
 	cmd.Flags().StringVar(&asIdentity, "as", "user", "事件身份: user")
 	cmd.Flags().StringVarP(&formatRaw, "format", "f", "json", "输出格式: json")
+	cmd.Flags().BoolVar(&flatten, "flatten", false, "显示 --flatten 消费模式对应的顶层业务字段 schema")
 	hideEventInternalFlags(cmd, "as")
 	cli.AnnotateRuntimePositionals(cmd, cli.RuntimeSchemaPositional{
 		Name:        "event_key",
@@ -188,7 +192,7 @@ func runPersonalEventList(c *cobra.Command, opts personalListOptions) error {
 	return tw.Flush()
 }
 
-func renderPersonalSchema(w io.Writer, def personal.Definition, format string) error {
+func renderPersonalSchema(w io.Writer, def personal.Definition, format string, flatten bool) error {
 	format = strings.ToLower(strings.TrimSpace(format))
 	if format == "" {
 		format = "json"
@@ -198,7 +202,7 @@ func renderPersonalSchema(w io.Writer, def personal.Definition, format string) e
 	}
 	enc := json.NewEncoder(w)
 	enc.SetIndent("", "  ")
-	return enc.Encode(personal.BuildSchemaDocument(def))
+	return enc.Encode(personal.BuildSchemaDocumentForMode(def, flatten))
 }
 
 func runPersonalEventConsume(c *cobra.Command, opts personalConsumeOptions) error {
@@ -206,6 +210,19 @@ func runPersonalEventConsume(c *cobra.Command, opts personalConsumeOptions) erro
 	if err := ensurePublicPersonalEvent(opts.EventKey); err != nil {
 		return err
 	}
+	rawFormat := ""
+	if f := c.Flags().Lookup("format"); f != nil && f.Changed {
+		rawFormat = opts.Common.FormatRaw
+	}
+	normalised, fellback := consume.NormalizeFormat(rawFormat)
+	if fellback && !opts.Common.Quiet {
+		fmt.Fprintf(c.ErrOrStderr(), "WARN: --format %q has no meaning for event stream; using ndjson\n", rawFormat)
+	}
+	if err := validatePersonalEventOutputMode(opts.Flatten, opts.DebugRawEvents, normalised); err != nil {
+		return fmt.Errorf("event consume --as user: %w", err)
+	}
+	projector := personalEventProjector(opts.DebugRawEvents, opts.Flatten)
+
 	configDir := defaultConfigDir()
 	identity, err := personalResolveEventIdentity(ctx, configDir, opts.StreamSourceID)
 	if err != nil {
@@ -220,16 +237,6 @@ func runPersonalEventConsume(c *cobra.Command, opts personalConsumeOptions) erro
 	if err != nil {
 		return fmt.Errorf("event consume --as user: %w", err)
 	}
-	rawFormat := ""
-	if f := c.Flags().Lookup("format"); f != nil && f.Changed {
-		rawFormat = opts.Common.FormatRaw
-	}
-	normalised, fellback := consume.NormalizeFormat(rawFormat)
-	if fellback && !opts.Common.Quiet {
-		fmt.Fprintf(c.ErrOrStderr(), "WARN: --format %q has no meaning for event stream; using ndjson\n", rawFormat)
-	}
-	projector := personalEventProjector(opts.DebugRawEvents)
-
 	if opts.Common.DryRun {
 		if strings.TrimSpace(opts.SubscribeID) == "" {
 			if err := validatePersonalSubscriptionOptions(opts); err != nil {
@@ -246,6 +253,7 @@ func runPersonalEventConsume(c *cobra.Command, opts personalConsumeOptions) erro
 			Duration:       opts.Common.Duration,
 			EventKey:       opts.EventKey,
 			Format:         normalised,
+			Flatten:        opts.Flatten,
 			OutputDir:      opts.Common.OutputDir,
 			Routes:         routes,
 			Projector:      projector,
@@ -259,7 +267,7 @@ func runPersonalEventConsume(c *cobra.Command, opts personalConsumeOptions) erro
 		return personalConsumeRun(ctx, cfg)
 	}
 
-	client := personal.NewClient(personalEventControlBaseURL(opts.ControlBaseURL, configDir), identity)
+	client := newPersonalEventControlClient(configDir, personalEventControlBaseURL(opts.ControlBaseURL, configDir), identity)
 	sub, eventKey, ruleType, err := personalEnsureSubscription(ctx, client, identity, opts)
 	if err != nil {
 		return fmt.Errorf("event consume --as user: %w", err)
@@ -302,6 +310,7 @@ func runPersonalEventConsume(c *cobra.Command, opts personalConsumeOptions) erro
 		Duration:         opts.Common.Duration,
 		EventKey:         eventKey,
 		Format:           normalised,
+		Flatten:          opts.Flatten,
 		OutputDir:        opts.Common.OutputDir,
 		Routes:           routes,
 		Projector:        projector,
@@ -365,11 +374,27 @@ func runPersonalEventConsume(c *cobra.Command, opts personalConsumeOptions) erro
 	return err
 }
 
-func personalEventProjector(debugRawEvents bool) consume.Projector {
+func personalEventProjector(debugRawEvents, flatten bool) consume.Projector {
 	if debugRawEvents {
 		return func(ev transport.Event) (any, error) { return ev, nil }
 	}
-	return personal.ProjectOutput
+	if flatten {
+		return personal.ProjectOutput
+	}
+	return nil
+}
+
+func validatePersonalEventOutputMode(flatten, debugRawEvents bool, format consume.Format) error {
+	if !flatten {
+		return nil
+	}
+	if debugRawEvents {
+		return fmt.Errorf("--flatten and --debug-raw-events are mutually exclusive")
+	}
+	if format == consume.FormatRaw {
+		return fmt.Errorf("--flatten and --format raw are mutually exclusive")
+	}
+	return nil
 }
 
 func applyPersonalConsumeFilters(cfg *consume.Config, opts personalConsumeOptions, subscribeID, eventKey string) {
@@ -498,7 +523,7 @@ func runPersonalEventStatus(c *cobra.Command, opts personalStatusOptions) error
 	if status == "" || status == "all" {
 		status = ""
 	}
-	subs, err := personalListSubscriptions(personal.NewClient(personalEventControlBaseURL(opts.ControlBaseURL, configDir), identity), ctx, personal.ListOptions{
+	subs, err := personalListSubscriptions(newPersonalEventControlClient(configDir, personalEventControlBaseURL(opts.ControlBaseURL, configDir), identity), ctx, personal.ListOptions{
 		Status:      status,
 		EventKey:    opts.EventKey,
 		SubscribeID: opts.SubscribeID,
@@ -613,7 +638,7 @@ func runPersonalEventStop(c *cobra.Command, opts personalStopOptions) error {
 	if err != nil {
 		return fmt.Errorf("event stop --as user: %w", err)
 	}
-	client := personal.NewClient(personalEventControlBaseURL(opts.ControlBaseURL, configDir), identity)
+	client := newPersonalEventControlClient(configDir, personalEventControlBaseURL(opts.ControlBaseURL, configDir), identity)
 	for _, id := range subscribeIDs {
 		if err := personalDeleteSubscription(client, ctx, id); err != nil {
 			return fmt.Errorf("event stop --as user: cancel subscription %s: %w", id, err)
@@ -723,7 +748,10 @@ func resolvePersonalEventIdentity(ctx context.Context, configDir string, sourceI
 	if err != nil {
 		return personal.Identity{}, err
 	}
-	tokenData, _ := personalLoadTokenData(configDir)
+	tokenData, err := personalLoadTokenData(configDir)
+	if err != nil && !errors.Is(err, authpkg.ErrTokenDataNotFound) {
+		return personal.Identity{}, fmt.Errorf("load OAuth identity metadata: %w", err)
+	}
 	var corpID, userID, clientID, refreshToken string
 	if tokenData != nil {
 		corpID = tokenData.CorpID
@@ -769,6 +797,15 @@ func resolvePersonalEventIdentity(ctx context.Context, configDir string, sourceI
 	}, nil
 }
 
+func newPersonalEventControlClient(configDir, baseURL string, identity personal.Identity) *personal.Client {
+	identity.AccessToken = ""
+	client := personal.NewClient(baseURL, identity)
+	client.AccessTokenProvider = func(ctx context.Context) (string, error) {
+		return personalResolveAuxiliaryAccessToken(ctx, configDir, "")
+	}
+	return client
+}
+
 func personalTokenSubject(kind, token string) string {
 	token = strings.TrimSpace(token)
 	if token == "" {
@@ -817,7 +854,12 @@ func newPersonalStreamSource(ctx context.Context, opts personalStreamSourceOptio
 	}
 	_ = ctx
 	return source.NewPersonal(source.PersonalConfig{
-		AccessToken:  opts.Identity.AccessToken,
+		AccessTokenProvider: func(ctx context.Context) (string, error) {
+			return personalResolveAuxiliaryAccessToken(ctx, opts.ConfigDir, "")
+		},
+		ForceRefreshToken: func(ctx context.Context, rejectedToken string) (string, error) {
+			return personalForceRefreshRejectedToken(ctx, opts.ConfigDir, rejectedToken)
+		},
 		ClientID:     clientID,
 		ClientSecret: clientSecret,
 		SourceID:     opts.Identity.SourceID,

internal/app/event_personal_debug_test.go

@@ -20,6 +20,7 @@ import (
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/event/consume"
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/event/personal"
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/event/transport"
+	"github.com/spf13/cobra"
 )
 
 func TestApplyPersonalConsumeFiltersDebugRawEvents(t *testing.T) {
@@ -52,11 +53,14 @@ func TestApplyPersonalConsumeFiltersDefault(t *testing.T) {
 	}
 }
 
-func TestPersonalEventProjectorUsesRawEnvelopeForDebug(t *testing.T) {
-	if personalEventProjector(false) == nil {
-		t.Fatal("normal personal consume projector = nil")
+func TestPersonalEventProjectorSelectsExplicitModes(t *testing.T) {
+	if personalEventProjector(false, false) != nil {
+		t.Fatal("default personal consume should preserve transport envelope")
 	}
-	projector := personalEventProjector(true)
+	if personalEventProjector(false, true) == nil {
+		t.Fatal("flatten personal consume projector = nil")
+	}
+	projector := personalEventProjector(true, false)
 	if projector == nil {
 		t.Fatal("debug raw personal consume projector = nil")
 	}
@@ -74,6 +78,69 @@ func TestPersonalEventProjectorUsesRawEnvelopeForDebug(t *testing.T) {
 	}
 }
 
+func TestEventConsumeFlattenRejectsRawModesBeforeIdentityResolution(t *testing.T) {
+	for _, tc := range []struct {
+		name string
+		args []string
+		want string
+	}{
+		{
+			name: "raw format",
+			args: []string{personal.EventMention, "--flatten", "--format", "raw"},
+			want: "--flatten and --format raw are mutually exclusive",
+		},
+		{
+			name: "raw debug",
+			args: []string{personal.EventMention, "--flatten", "--debug-raw-events"},
+			want: "--flatten and --debug-raw-events are mutually exclusive",
+		},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Setenv("DWS_CONFIG_DIR", t.TempDir())
+			cmd := newEventConsumeCommand()
+			cmd.SilenceUsage = true
+			cmd.SilenceErrors = true
+			cmd.SetArgs(tc.args)
+			err := cmd.Execute()
+			if err == nil || !strings.Contains(err.Error(), tc.want) {
+				t.Fatalf("Execute() error = %v, want %q", err, tc.want)
+			}
+			if strings.Contains(err.Error(), "login") || strings.Contains(err.Error(), "token") {
+				t.Fatalf("output-mode validation ran after identity resolution: %v", err)
+			}
+		})
+	}
+}
+
+func TestValidatePersonalEventOutputModeAllowsFlattenStructuredFormats(t *testing.T) {
+	for _, format := range []consume.Format{consume.FormatNDJSON, consume.FormatJSON, consume.FormatPretty, consume.FormatCompact} {
+		if err := validatePersonalEventOutputMode(true, false, format); err != nil {
+			t.Fatalf("validatePersonalEventOutputMode(true, false, %q) error = %v", format, err)
+		}
+	}
+}
+
+func TestEventConsumeFlattenFlagIsForwarded(t *testing.T) {
+	oldRun := eventRunPersonalConsume
+	t.Cleanup(func() { eventRunPersonalConsume = oldRun })
+
+	var got personalConsumeOptions
+	eventRunPersonalConsume = func(_ *cobra.Command, opts personalConsumeOptions) error {
+		got = opts
+		return nil
+	}
+	cmd := newEventConsumeCommand()
+	cmd.SilenceUsage = true
+	cmd.SilenceErrors = true
+	cmd.SetArgs([]string{personal.EventMention, "--flatten", "--format", "compact"})
+	if err := cmd.Execute(); err != nil {
+		t.Fatalf("Execute() error = %v", err)
+	}
+	if !got.Flatten || got.Common.FormatRaw != "compact" {
+		t.Fatalf("forwarded options = %#v", got)
+	}
+}
+
 func TestEventConsumeDebugRawEventsRequiresUserMode(t *testing.T) {
 	cmd := newEventConsumeCommand()
 	cmd.SilenceUsage = true

internal/app/event_personal_force_refresh_wire_test.go

@@ -0,0 +1,54 @@
+package app
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	dwsevent "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/event"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/event/personal"
+)
+
+// TestCrossPlatformCoverageNewPersonalStreamSourceWiresForceRefreshRejectedToken asserts the
+// personal stream source receives a ForceRefreshToken callback that forwards
+// the rejected token into the app-level compare-and-refresh chain.
+func TestCrossPlatformCoverageNewPersonalStreamSourceWiresForceRefreshRejectedToken(t *testing.T) {
+	oldAux := personalResolveAuxiliaryAccessToken
+	oldRefresh := personalForceRefreshRejectedToken
+	t.Cleanup(func() {
+		personalResolveAuxiliaryAccessToken = oldAux
+		personalForceRefreshRejectedToken = oldRefresh
+	})
+	personalResolveAuxiliaryAccessToken = func(context.Context, string, string) (string, error) {
+		return "old-token", nil
+	}
+	refreshErr := errors.New("refresh rejected")
+	var gotDir, gotRejected string
+	personalForceRefreshRejectedToken = func(_ context.Context, configDir, rejectedToken string) (string, error) {
+		gotDir, gotRejected = configDir, rejectedToken
+		return "", refreshErr
+	}
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusUnauthorized)
+	}))
+	defer srv.Close()
+
+	src, err := newPersonalStreamSource(context.Background(), personalStreamSourceOptions{
+		ConfigDir: "config-dir",
+		Identity:  personal.Identity{ClientID: "client", SourceID: "source"},
+		TicketURL: srv.URL,
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	// The 401 ticket response routes the rejected token through the wired
+	// ForceRefreshToken; the unknown refresh failure stays fatal.
+	if err := src.Start(context.Background(), func(*dwsevent.RawEvent) {}); !errors.Is(err, refreshErr) {
+		t.Fatalf("Start() error = %v, want wrapped refresh error", err)
+	}
+	if gotDir != "config-dir" || gotRejected != "old-token" {
+		t.Fatalf("refresh wiring got dir %q rejected %q", gotDir, gotRejected)
+	}
+}

internal/app/event_personal_output_test.go

@@ -188,7 +188,44 @@ func TestPersonalEventSchemaHidesSchemaIDs(t *testing.T) {
 	}
 }
 
-func TestPersonalEventSchemaUsesSingleJSONSchema(t *testing.T) {
+func TestPersonalEventSchemaDefaultsToTransportEnvelope(t *testing.T) {
+	cmd := newEventSchemaCommand()
+	cmd.SilenceUsage = true
+	cmd.SilenceErrors = true
+	var out bytes.Buffer
+	cmd.SetOut(&out)
+	cmd.SetArgs([]string{personal.EventSingleChat})
+	if err := cmd.Execute(); err != nil {
+		t.Fatalf("Execute() error = %v", err)
+	}
+	var doc map[string]any
+	if err := json.Unmarshal(out.Bytes(), &doc); err != nil {
+		t.Fatalf("schema output is not JSON: %v\n%s", err, out.String())
+	}
+	if doc["jq_root_path"] != ".data | fromjson" {
+		t.Fatalf("jq_root_path = %#v, want .data | fromjson", doc["jq_root_path"])
+	}
+	schema, ok := doc["schema"].(map[string]any)
+	if !ok {
+		t.Fatalf("schema = %#v, want object", doc["schema"])
+	}
+	props, ok := schema["properties"].(map[string]any)
+	if !ok {
+		t.Fatalf("schema.properties = %#v, want object", schema["properties"])
+	}
+	for _, field := range []string{"type", "seq", "event_type", "data", "headers", "subscribe_id"} {
+		if _, ok := props[field]; !ok {
+			t.Fatalf("default envelope schema missing %q: %#v", field, props)
+		}
+	}
+	for _, field := range []string{"content", "sender", "conversation_id", "timestamp"} {
+		if _, ok := props[field]; ok {
+			t.Fatalf("default envelope schema unexpectedly contains flat field %q", field)
+		}
+	}
+}
+
+func TestPersonalEventFlattenedSchemaUsesSingleJSONSchema(t *testing.T) {
 	for _, eventKey := range []string{
 		personal.EventMention,
 		personal.EventSingleChat,
@@ -200,7 +237,7 @@ func TestPersonalEventSchemaUsesSingleJSONSchema(t *testing.T) {
 			cmd.SilenceErrors = true
 			var out bytes.Buffer
 			cmd.SetOut(&out)
-			cmd.SetArgs([]string{eventKey})
+			cmd.SetArgs([]string{eventKey, "--flatten"})
 			if err := cmd.Execute(); err != nil {
 				t.Fatalf("Execute() error = %v", err)
 			}
@@ -316,7 +353,7 @@ func TestPersonalActionEventSchemaMatchesFlatOutput(t *testing.T) {
 				cmd.SilenceErrors = true
 				var out bytes.Buffer
 				cmd.SetOut(&out)
-				cmd.SetArgs([]string{eventKey})
+				cmd.SetArgs([]string{eventKey, "--flatten"})
 				if err := cmd.Execute(); err != nil {
 					t.Fatal(err)
 				}
@@ -367,6 +404,9 @@ func TestEventSchemaDefaultsToUser(t *testing.T) {
 	if doc["event_key"] != personal.EventSingleChat {
 		t.Fatalf("event_key = %#v, want %s", doc["event_key"], personal.EventSingleChat)
 	}
+	if doc["jq_root_path"] != ".data | fromjson" {
+		t.Fatalf("jq_root_path = %#v, want default envelope path", doc["jq_root_path"])
+	}
 }
 
 func TestPersonalEventFromUserIsPubliclyAvailable(t *testing.T) {
@@ -469,6 +509,9 @@ func TestEventConsumeCobraSchemaIncludesOpenDingTalkID(t *testing.T) {
 	if _, ok := params["odid"]; ok {
 		t.Fatalf("schema parameters unexpectedly include odid alias: %#v", params)
 	}
+	if _, ok := params["flatten"]; !ok {
+		t.Fatalf("schema parameters missing flatten: %#v", params)
+	}
 	for _, name := range []string{"user", "open-dingtalk-id", "group"} {
 		param, ok := params[name].(map[string]any)
 		if !ok {

internal/app/force_refresh.go

@@ -27,9 +27,13 @@ type accessTokenGetter interface {
 	GetAccessToken(context.Context) (string, error)
 }
 
+type rejectedAccessTokenRefresher interface {
+	ForceRefreshRejectedToken(context.Context, string) (string, error)
+}
+
 var (
-	markAccessTokenStale = authpkg.MarkAccessTokenStale
-	newRefreshProvider   = func(configDir string) accessTokenGetter {
+	loadRefreshTokenData = authpkg.LoadTokenData
+	newRefreshProvider   = func(configDir string) rejectedAccessTokenRefresher {
 		disc := slog.New(slog.NewTextHandler(io.Discard, nil))
 		provider := authpkg.NewOAuthProvider(configDir, disc)
 		configureOAuthProviderCompatibility(provider, configDir)
@@ -42,26 +46,33 @@ var (
 // server-side rejection (HTTP 401 or business code such as
 // TOKEN_VERIFIED_FAILED) on what locally appeared to be a still-valid token.
 //
-// Steps:
-//  1. MarkAccessTokenStale rewrites ExpiresAt to a past instant so
-//     OAuthProvider.GetAccessToken's fast-path will miss.
-//  2. NewOAuthProvider + GetAccessToken triggers lockedRefresh, which uses the
-//     existing dual-layer lock (process + file) to serialize concurrent
-//     refresh attempts across goroutines and processes.
-//  3. ResetRuntimeTokenCache clears the per-process sync.Once cache so the
-//     next resolveAuthToken call re-reads from disk.
-//
-// Existing OAuthProvider.GetAccessToken behaviour is unchanged; this helper
-// is the only entry point that orchestrates "force refresh" semantics.
+// It snapshots the current access token, then delegates to the OAuth
+// provider's dual-locked compare-and-refresh operation. If another caller has
+// already rotated the token, that newer token is reused without another
+// refresh request.
 func ForceRefreshAccessToken(ctx context.Context, configDir string) (string, error) {
 	if strings.TrimSpace(configDir) == "" {
 		return "", fmt.Errorf("config directory is empty")
 	}
-	if err := markAccessTokenStale(configDir); err != nil {
-		return "", fmt.Errorf("mark access token stale: %w", err)
+	data, err := loadRefreshTokenData(configDir)
+	if err != nil {
+		return "", err
+	}
+	if data == nil || strings.TrimSpace(data.AccessToken) == "" {
+		return "", fmt.Errorf("stored access token is empty")
+	}
+	return forceRefreshRejectedAccessToken(ctx, configDir, data.AccessToken)
+}
+
+func forceRefreshRejectedAccessToken(ctx context.Context, configDir, rejectedAccessToken string) (string, error) {
+	if strings.TrimSpace(configDir) == "" {
+		return "", fmt.Errorf("config directory is empty")
+	}
+	if strings.TrimSpace(rejectedAccessToken) == "" {
+		return "", fmt.Errorf("rejected access token is empty")
 	}
 	provider := newRefreshProvider(configDir)
-	tok, err := provider.GetAccessToken(ctx)
+	tok, err := provider.ForceRefreshRejectedToken(ctx, rejectedAccessToken)
 	if err != nil {
 		return "", err
 	}

internal/app/legacy.go

@@ -27,21 +27,27 @@ import (
 	"github.com/spf13/cobra"
 )
 
-func newLegacyPublicCommands(runner executor.Runner, caller edition.ToolCaller) []*cobra.Command {
+func newLegacyPublicCommands(runner executor.Runner, caller edition.ToolCaller, loadUserShortcuts bool) []*cobra.Command {
 	injectStaticServers()
 	helpers.InitDeps(caller)
 	commands := helpers.NewPublicCommands(runner)
 	// Load user-defined shortcuts (~/.dws/shortcuts/*.yaml) BEFORE compiling the
 	// command tree, so distilled high-frequency operations mount alongside the
 	// built-ins. Conflicts with built-ins are skipped inside Load.
-	if _, err := userdef.Load(); err != nil {
-		slog.Warn("shortcut: failed to load user-defined shortcuts", "error", err)
+	if loadUserShortcuts {
+		if _, err := userdef.Load(); err != nil {
+			slog.Warn("shortcut: failed to load user-defined shortcuts", "error", err)
+		}
 	}
 	// Built-in + user shortcuts (`dws <service> +<command>`) share the same
 	// command tree; mergeTopLevelCommands folds each shortcut's service parent
 	// into the matching helper command so the `+leaf` sits alongside existing
 	// subcommands.
-	commands = append(commands, builtin.Commands()...)
+	if loadUserShortcuts {
+		commands = append(commands, builtin.Commands()...)
+	} else {
+		commands = append(commands, builtin.BaseCommands()...)
+	}
 	return mergeTopLevelCommands(commands)
 }
 

internal/app/pat_auth_retry.go

@@ -56,7 +56,7 @@ var openBrowserFunc = tryOpenBrowser
 var (
 	patAuthorizationTimeout       = PatAuthRetryTimeout
 	patAuthorizationPollInterval  = PatAuthPollInterval
-	patLoadTokenData              = authpkg.LoadTokenData
+	patResolveAccessToken         = ResolveAuxiliaryAccessToken
 	patWaitForAuthorization       = WaitForPatAuthorization
 	patPollDeviceFlowWithInterval = pollPatDeviceFlowWithInterval
 	patSaveAppConfig              = authpkg.SaveAppConfig
@@ -272,7 +272,7 @@ func patAuthorizationURIFromData(data map[string]any) string {
 
 // WaitForPatAuthorization polls until the user completes authorization or timeout.
 // It returns true if authorization was completed, false if timed out or cancelled.
-func WaitForPatAuthorization(ctx context.Context, configDir string, output io.Writer) bool {
+func WaitForPatAuthorization(ctx context.Context, configDir string, output io.Writer) (bool, error) {
 	timeout := patAuthorizationTimeout
 	deadline := time.Now().Add(timeout)
 	pollTicker := time.NewTicker(patAuthorizationPollInterval)
@@ -290,27 +290,26 @@ func WaitForPatAuthorization(ctx context.Context, configDir string, output io.Wr
 		select {
 		case <-ctx.Done():
 			fmt.Fprintf(output, "%s 操作已取消\n", tui.StateMark("error"))
-			return false
+			return false, ctx.Err()
 
 		case <-time.After(time.Until(deadline)):
 			fmt.Fprintf(output, "%s 等待授权超时 (%s)\n", tui.StateMark("error"), timeout)
 			fmt.Fprintf(output, "  %s 请重新执行命令\n", tui.Dim("ℹ"))
-			return false
+			return false, nil
 
 		case <-pollTicker.C:
 			pollCount++
 			elapsed := time.Since(start).Truncate(time.Second)
 			remaining := time.Until(deadline).Truncate(time.Second)
 
-			// Check if token is now valid
-			tokenData, err := patLoadTokenData(configDir)
-			if err == nil && tokenData != nil {
-				if tokenData.IsAccessTokenValid() || tokenData.IsRefreshTokenValid() {
-					fmt.Fprintf(output, "\r%s %s (%s 已用, %s 剩余)          \n",
-						tui.StateMark("ok"), tui.Bold("授权成功!"), elapsed, remaining)
-					fmt.Fprintln(output)
-					return true
-				}
+			// Check the same resolver used by every outbound bearer request.
+			if _, err := patResolveAccessToken(ctx, configDir, ""); err == nil {
+				fmt.Fprintf(output, "\r%s %s (%s 已用, %s 剩余)          \n",
+					tui.StateMark("ok"), tui.Bold("授权成功!"), elapsed, remaining)
+				fmt.Fprintln(output)
+				return true, nil
+			} else if !stderrors.Is(err, authpkg.ErrTokenDataNotFound) {
+				return false, fmt.Errorf("check authorization token: %w", err)
 			}
 
 			// Show polling status
@@ -340,7 +339,10 @@ func retryWithPatAuthRetry(ctx context.Context, runner executor.Runner, invocati
 	PrintPatAuthError(output, scopeErr)
 
 	// Wait for user to complete authorization
-	authorized := patWaitForAuthorization(ctx, configDir, output)
+	authorized, waitErr := patWaitForAuthorization(ctx, configDir, output)
+	if waitErr != nil {
+		return executor.Result{}, waitErr
+	}
 	if !authorized {
 		return executor.Result{}, apperrors.NewAuth(
 			"等待用户授权超时",
@@ -794,12 +796,6 @@ func pollPatDeviceFlowWithInterval(ctx context.Context, flowID string, configDir
 	pollURL := fmt.Sprintf("%s%s?flowId=%s",
 		authpkg.GetMCPBaseURL(), authpkg.DevicePollPath, url.QueryEscape(flowID))
 
-	// Load user access token for the poll request header.
-	var accessToken string
-	if tokenData, err := authpkg.LoadTokenData(configDir); err == nil && tokenData != nil {
-		accessToken = tokenData.AccessToken
-	}
-
 	// Use a client that does NOT follow redirects, so we can detect SSO 302.
 	noRedirectClient := &http.Client{
 		CheckRedirect: func(req *http.Request, via []*http.Request) error {
@@ -828,6 +824,10 @@ func pollPatDeviceFlowWithInterval(ctx context.Context, flowID string, configDir
 				slog.Debug("PAT poll: failed to create request", "error", err)
 				continue
 			}
+			accessToken, tokenErr := patResolveAccessToken(ctx, configDir, "")
+			if tokenErr != nil && !stderrors.Is(tokenErr, authpkg.ErrTokenDataNotFound) {
+				return "", "", fmt.Errorf("resolve PAT poll access token: %w", tokenErr)
+			}
 			if accessToken != "" {
 				req.Header.Set("x-user-access-token", accessToken)
 			}

internal/app/pat_auth_retry_full_coverage_test.go

@@ -52,39 +52,41 @@ func TestCrossPlatformCoveragePATRetryRemainingPureAndWaitCoverage(t *testing.T)
 
 	oldTimeout := patAuthorizationTimeout
 	oldInterval := patAuthorizationPollInterval
-	oldLoad := patLoadTokenData
+	oldResolve := patResolveAccessToken
 	t.Cleanup(func() {
 		patAuthorizationTimeout = oldTimeout
 		patAuthorizationPollInterval = oldInterval
-		patLoadTokenData = oldLoad
+		patResolveAccessToken = oldResolve
 	})
 	patAuthorizationTimeout = 50 * time.Millisecond
 	patAuthorizationPollInterval = time.Millisecond
-	patLoadTokenData = func(string) (*authpkg.TokenData, error) {
-		return &authpkg.TokenData{AccessToken: "token", ExpiresAt: time.Now().Add(time.Hour)}, nil
+	patResolveAccessToken = func(context.Context, string, string) (string, error) {
+		return "token", nil
 	}
 	out.Reset()
-	if !WaitForPatAuthorization(context.Background(), "", &out) {
+	if ok, err := WaitForPatAuthorization(context.Background(), "", &out); err != nil || !ok {
 		t.Fatal("valid token did not authorize")
 	}
 	ctx, cancel := context.WithCancel(context.Background())
 	cancel()
 	out.Reset()
-	if WaitForPatAuthorization(ctx, "", &out) {
-		t.Fatal("cancelled authorization succeeded")
+	if ok, err := WaitForPatAuthorization(ctx, "", &out); ok || !errors.Is(err, context.Canceled) {
+		t.Fatalf("cancelled authorization = %v, %v", ok, err)
 	}
 	patAuthorizationTimeout = time.Millisecond
 	patAuthorizationPollInterval = time.Hour
 	out.Reset()
-	if WaitForPatAuthorization(context.Background(), "", &out) {
-		t.Fatal("timed out authorization succeeded")
+	if ok, err := WaitForPatAuthorization(context.Background(), "", &out); err != nil || ok {
+		t.Fatalf("timed out authorization = %v, %v", ok, err)
 	}
 	patAuthorizationTimeout = 5 * time.Millisecond
 	patAuthorizationPollInterval = time.Millisecond
-	patLoadTokenData = func(string) (*authpkg.TokenData, error) { return nil, nil }
+	patResolveAccessToken = func(context.Context, string, string) (string, error) {
+		return "", authpkg.ErrTokenDataNotFound
+	}
 	out.Reset()
-	if WaitForPatAuthorization(context.Background(), "", &out) || !strings.Contains(out.String(), "等待授权中") {
-		t.Fatalf("invalid-token polling output = %q", out.String())
+	if ok, err := WaitForPatAuthorization(context.Background(), "", &out); err != nil || ok || !strings.Contains(out.String(), "等待授权中") {
+		t.Fatalf("invalid-token polling = %v, %v, output %q", ok, err, out.String())
 	}
 }
 
@@ -109,12 +111,12 @@ func TestCrossPlatformCoveragePATRetryRemainingOrchestrationCoverage(t *testing.
 	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
 
 	scope := &PatScopeError{OriginalError: "missing", Identity: "user", ErrorType: "missing_scope", Message: "missing", Hint: "login", MissingScope: "calendar:read"}
-	patWaitForAuthorization = func(context.Context, string, io.Writer) bool { return false }
+	patWaitForAuthorization = func(context.Context, string, io.Writer) (bool, error) { return false, nil }
 	if _, err := retryWithPatAuthRetry(context.Background(), runnerCoverageFallback{}, executor.Invocation{}, scope, t.TempDir(), io.Discard); err == nil {
 		t.Fatal("PAT retry timeout succeeded")
 	}
 	wantErr := errors.New("runner failed")
-	patWaitForAuthorization = func(context.Context, string, io.Writer) bool { return true }
+	patWaitForAuthorization = func(context.Context, string, io.Writer) (bool, error) { return true, nil }
 	if _, err := retryWithPatAuthRetry(context.Background(), runnerCoverageFallback{err: wantErr}, executor.Invocation{}, scope, t.TempDir(), io.Discard); !errors.Is(err, wantErr) {
 		t.Fatalf("authorized retry = %v", err)
 	}
@@ -215,15 +217,15 @@ func patRaw(flowID, clientID, secret string) string {
 func TestCrossPlatformCoveragePATRetryRemainingPollAndBrowserCoverage(t *testing.T) {
 	oldDo := patPollHTTPDo
 	oldRequest := patPollNewRequest
-	oldLoad := patLoadTokenData
+	oldResolve := patResolveAccessToken
 	oldBrowser := patBrowserOpenCommand
 	t.Cleanup(func() {
 		patPollHTTPDo = oldDo
 		patPollNewRequest = oldRequest
-		patLoadTokenData = oldLoad
+		patResolveAccessToken = oldResolve
 		patBrowserOpenCommand = oldBrowser
 	})
-	patLoadTokenData = func(string) (*authpkg.TokenData, error) { return &authpkg.TokenData{AccessToken: "token"}, nil }
+	patResolveAccessToken = func(context.Context, string, string) (string, error) { return "token", nil }
 	cancelled, cancelNow := context.WithCancel(context.Background())
 	cancelNow()
 	if status, _, err := pollPatDeviceFlowWithInterval(cancelled, "flow", t.TempDir(), io.Discard, 0); err != nil || status != authpkg.StatusCancelled {

internal/app/plugin_commands_coverage_test.go

@@ -0,0 +1,675 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0 (the "License");
+
+package app
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"reflect"
+	"strings"
+	"testing"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/cobracmd"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/executor"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/cmdutil"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/mcptypes"
+	"github.com/spf13/cobra"
+)
+
+type pluginFailRunner struct{}
+
+func (pluginFailRunner) Run(context.Context, executor.Invocation) (executor.Result, error) {
+	return executor.Result{}, errors.New("runner failed")
+}
+
+type pluginWrongFlagValue struct{}
+
+func (pluginWrongFlagValue) String() string   { return "" }
+func (pluginWrongFlagValue) Set(string) error { return nil }
+func (pluginWrongFlagValue) Type() string     { return "wrong" }
+
+func TestPluginCompilerRejectsInvalidDuplicateAndEmptyDefinitions(t *testing.T) {
+	invalidRoot := conferencePluginDescriptor()
+	invalidRoot.CLI.Command = "Invalid Root"
+	if commands := buildPluginCommands([]mcptypes.ServerDescriptor{invalidRoot}, executor.EchoRunner{}, nil); len(commands) != 0 {
+		t.Fatalf("invalid root produced commands %#v", commands)
+	}
+
+	descriptor := conferencePluginDescriptor()
+	descriptor.CLI.Groups = map[string]mcptypes.CLIGroupDef{
+		"empty": {Description: "removed when no leaf survives"},
+	}
+	descriptor.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+		"":        {CLIName: "blank-tool"},
+		"hidden":  {CLIName: "hidden", Hidden: true},
+		"invalid": {CLIName: "Invalid Leaf"},
+		"first":   {CLIName: "same"},
+		"second":  {CLIName: "same"},
+	}
+	commands := buildPluginCommands([]mcptypes.ServerDescriptor{descriptor}, executor.EchoRunner{}, nil)
+	if len(commands) != 1 {
+		t.Fatalf("commands = %#v", commands)
+	}
+	if requireOptionalPluginChild(commands[0], "same") == nil {
+		t.Fatal("valid leaf was not retained")
+	}
+	if requireOptionalPluginChild(commands[0], "empty") != nil {
+		t.Fatal("empty group was not pruned")
+	}
+
+	empty := conferencePluginDescriptor()
+	empty.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+		"hidden": {CLIName: "hidden", Hidden: true},
+	}
+	if commands := buildPluginCommands([]mcptypes.ServerDescriptor{empty}, executor.EchoRunner{}, nil); len(commands) != 0 {
+		t.Fatalf("empty overlay produced commands %#v", commands)
+	}
+}
+
+func TestPluginLeafExecutionErrorsAndBodyWrapper(t *testing.T) {
+	base := conferencePluginDescriptor()
+	base.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+		"wrapped": {
+			CLIName:     "wrapped",
+			BodyWrapper: "body",
+			Flags: map[string]mcptypes.CLIFlagOverride{
+				"value": {Required: true},
+			},
+		},
+	}
+	runner := &pluginCaptureRunner{}
+	root := pluginTestRoot(buildPluginCommands([]mcptypes.ServerDescriptor{base}, runner, nil)...)
+	root.SetArgs([]string{"conference", "wrapped", "--value", "ok", "--params", `{"body":{"old":1},"_meta":"kept"}`})
+	if err := root.Execute(); err != nil {
+		t.Fatalf("wrapped command: %v", err)
+	}
+	want := map[string]any{
+		"_meta": "kept",
+		"body":  map[string]any{"old": float64(1), "value": "ok"},
+	}
+	if !reflect.DeepEqual(runner.invocations[0].Params, want) {
+		t.Fatalf("wrapped params = %#v, want %#v", runner.invocations[0].Params, want)
+	}
+
+	for _, testCase := range []struct {
+		name   string
+		runner executor.Runner
+		args   []string
+	}{
+		{name: "invalid json", runner: executor.EchoRunner{}, args: []string{"conference", "wrapped", "--json", "["}},
+		{name: "missing required", runner: executor.EchoRunner{}, args: []string{"conference", "wrapped"}},
+		{name: "missing runner", runner: nil, args: []string{"conference", "wrapped", "--value", "ok"}},
+		{name: "runner error", runner: pluginFailRunner{}, args: []string{"conference", "wrapped", "--value", "ok"}},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			commandRoot := pluginTestRoot(buildPluginCommands([]mcptypes.ServerDescriptor{base}, testCase.runner, nil)...)
+			commandRoot.SetArgs(testCase.args)
+			if err := commandRoot.Execute(); err == nil {
+				t.Fatal("expected command error")
+			}
+		})
+	}
+
+	for _, flagName := range []string{"json", "params"} {
+		t.Run("unreadable "+flagName, func(t *testing.T) {
+			commands := buildPluginCommands([]mcptypes.ServerDescriptor{base}, executor.EchoRunner{}, nil)
+			leaf := requirePluginChild(t, commands[0], "wrapped")
+			leaf.Flags().Lookup(flagName).Value = pluginWrongFlagValue{}
+			commandRoot := pluginTestRoot(commands...)
+			commandRoot.SetArgs([]string{"conference", "wrapped", "--value", "ok"})
+			if err := commandRoot.Execute(); err == nil {
+				t.Fatal("expected unreadable flag error")
+			}
+		})
+	}
+}
+
+func TestPluginBindingCompilerCoversAliasesAndPositionalValidators(t *testing.T) {
+	reservations := pluginFlagReservations{
+		names:      map[string]bool{"reserved": true},
+		shorthands: map[string]bool{},
+	}
+	bindings, _, _, ok := registerPluginBindings("alias", mcptypes.CLIToolOverride{
+		Flags: map[string]mcptypes.CLIFlagOverride{
+			"value": {Alias: "value", Aliases: []string{"", "Bad", "value", "other"}},
+		},
+	}, reservations)
+	if !ok || !reflect.DeepEqual(bindings[0].names, []string{"value", "other"}) {
+		t.Fatalf("alias bindings = (%#v, %v)", bindings, ok)
+	}
+	if _, _, _, ok := registerPluginBindings("conflict", mcptypes.CLIToolOverride{
+		Flags: map[string]mcptypes.CLIFlagOverride{"value": {Alias: "reserved"}},
+	}, reservations); ok {
+		t.Fatal("reserved flag was accepted")
+	}
+	if _, _, _, ok := registerPluginBindings("negative", mcptypes.CLIToolOverride{
+		Flags: map[string]mcptypes.CLIFlagOverride{"value": {Positional: true, PositionalIndex: -1}},
+	}, reservations); ok {
+		t.Fatal("negative positional index was accepted")
+	}
+	if _, _, _, ok := registerPluginBindings("duplicate", mcptypes.CLIToolOverride{
+		Flags: map[string]mcptypes.CLIFlagOverride{
+			"first":  {Positional: true, PositionalIndex: 0},
+			"second": {Positional: true, PositionalIndex: 0},
+		},
+	}, reservations); ok {
+		t.Fatal("duplicate positional index was accepted")
+	}
+	if _, _, _, ok := registerPluginBindings("gap", mcptypes.CLIToolOverride{
+		Flags: map[string]mcptypes.CLIFlagOverride{
+			"second": {Positional: true, PositionalIndex: 1},
+		},
+	}, reservations); ok {
+		t.Fatal("non-contiguous positional indexes were accepted")
+	}
+
+	for _, testCase := range []struct {
+		name    string
+		flags   map[string]mcptypes.CLIFlagOverride
+		wantUse string
+		valid   []string
+		invalid []string
+	}{
+		{
+			name: "exact",
+			flags: map[string]mcptypes.CLIFlagOverride{
+				"second": {Positional: true, PositionalIndex: 1, Required: true},
+				"first":  {Positional: true, PositionalIndex: 0, Required: true},
+			},
+			wantUse: "exact [first] [second]", valid: []string{"a", "b"}, invalid: []string{"a"},
+		},
+		{
+			name: "range",
+			flags: map[string]mcptypes.CLIFlagOverride{
+				"first":  {Positional: true, PositionalIndex: 0, Required: true},
+				"second": {Positional: true, PositionalIndex: 1},
+			},
+			wantUse: "range [first] [second]", valid: []string{"a"}, invalid: []string{},
+		},
+		{
+			name: "maximum",
+			flags: map[string]mcptypes.CLIFlagOverride{
+				"first":  {Positional: true, PositionalIndex: 0},
+				"second": {Positional: true, PositionalIndex: 1},
+			},
+			wantUse: "maximum [first] [second]", valid: []string{}, invalid: []string{"a", "b", "c"},
+		},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			_, use, validator, ok := registerPluginBindings(testCase.name, mcptypes.CLIToolOverride{Flags: testCase.flags}, reservations)
+			if !ok || use != testCase.wantUse {
+				t.Fatalf("binding contract = (%q, %v)", use, ok)
+			}
+			cmd := &cobra.Command{Use: testCase.name}
+			if err := validator(cmd, testCase.valid); err != nil {
+				t.Fatalf("valid args: %v", err)
+			}
+			if err := validator(cmd, testCase.invalid); err == nil {
+				t.Fatal("invalid args were accepted")
+			}
+		})
+	}
+}
+
+func TestPluginFlagRegistrationAndReadingCoversAllKinds(t *testing.T) {
+	cmd := &cobra.Command{Use: "leaf"}
+	override := mcptypes.CLIToolOverride{Flags: map[string]mcptypes.CLIFlagOverride{
+		"integer": {Default: "2", Shorthand: "i", Hidden: true},
+		"float":   {Default: "1.5"},
+		"boolean": {Default: "true"},
+		"slice":   {Default: "one, ,two"},
+		"json":    {Default: `{"old":true}`},
+		"string":  {Default: "text"},
+	}}
+	bindings := []pluginFlagBinding{
+		{property: "integer", names: []string{"integer", "integer-alias"}, kind: pluginFlagInt},
+		{property: "float", names: []string{"float"}, kind: pluginFlagFloat},
+		{property: "boolean", names: []string{"boolean"}, kind: pluginFlagBool},
+		{property: "slice", names: []string{"slice"}, kind: pluginFlagStringSlice},
+		{property: "json", names: []string{"json-value"}, kind: pluginFlagJSON},
+		{property: "string", names: []string{"string"}, kind: pluginFlagString},
+	}
+	registerPluginFlags(cmd, bindings, override, pluginFlagReservations{shorthands: map[string]bool{}})
+	for name, raw := range map[string]string{
+		"integer": "3", "float": "2.5", "boolean": "false",
+		"slice": "three,four", "json-value": `{"ok":true}`, "string": "changed",
+	} {
+		if err := cmd.Flags().Set(name, raw); err != nil {
+			t.Fatalf("set --%s: %v", name, err)
+		}
+	}
+	wants := map[string]any{
+		"integer": 3,
+		"float":   2.5,
+		"boolean": false,
+		"slice":   []string{"three", "four"},
+		"json":    map[string]any{"ok": true},
+		"string":  "changed",
+	}
+	for _, binding := range bindings {
+		value, err := readPluginFlag(cmd.Flags(), binding.names[0], binding.kind)
+		if err != nil || !reflect.DeepEqual(value, wants[binding.property]) {
+			t.Fatalf("read %s = (%#v, %v), want %#v", binding.property, value, err, wants[binding.property])
+		}
+	}
+	if !cmd.Flags().Lookup("integer").Hidden || !cmd.Flags().Lookup("integer-alias").Hidden {
+		t.Fatal("hidden primary or alias flag was exposed")
+	}
+	if err := cmd.Flags().Set("json-value", "{"); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := readPluginFlag(cmd.Flags(), "json-value", pluginFlagJSON); err == nil {
+		t.Fatal("invalid JSON flag was accepted")
+	}
+	cmd.Flags().Lookup("json-value").Value = pluginWrongFlagValue{}
+	if _, err := readPluginFlag(cmd.Flags(), "json-value", pluginFlagJSON); err == nil {
+		t.Fatal("wrong JSON flag type was accepted")
+	}
+}
+
+func TestCollectPluginBindingsCoversEveryValueSourceAndFailure(t *testing.T) {
+	t.Run("sources", func(t *testing.T) {
+		cmd := &cobra.Command{Use: "leaf"}
+		registerPluginFlag(cmd.Flags(), "flag", "", "", pluginFlagString, "")
+		if err := cmd.Flags().Set("flag", "from-flag"); err != nil {
+			t.Fatal(err)
+		}
+		t.Setenv("PLUGIN_COVERAGE_ENV", "7")
+		params := map[string]any{"existing": "from-json"}
+		bindings := []pluginFlagBinding{
+			{property: "flag", names: []string{"flag"}, kind: pluginFlagString},
+			{property: "existing", kind: pluginFlagString},
+			{property: "positional", kind: pluginFlagBool, positional: true, positionalIndex: 0},
+			{property: "default", kind: pluginFlagFloat, defaultProvided: true, defaultValue: "1.5"},
+			{property: "env", kind: pluginFlagInt, envDefault: "PLUGIN_COVERAGE_ENV"},
+			{property: "optional", kind: pluginFlagString},
+		}
+		if err := collectPluginBindings(cmd, []string{"true"}, bindings, params); err != nil {
+			t.Fatal(err)
+		}
+		want := map[string]any{
+			"flag": "from-flag", "existing": "from-json", "positional": true,
+			"default": 1.5, "env": 7,
+		}
+		if !reflect.DeepEqual(params, want) {
+			t.Fatalf("params = %#v, want %#v", params, want)
+		}
+	})
+
+	for _, testCase := range []struct {
+		name    string
+		prepare func(t *testing.T, cmd *cobra.Command)
+		args    []string
+		binding pluginFlagBinding
+		params  map[string]any
+	}{
+		{
+			name: "wrong flag type",
+			prepare: func(t *testing.T, cmd *cobra.Command) {
+				cmd.Flags().String("value", "", "")
+				if err := cmd.Flags().Set("value", "x"); err != nil {
+					t.Fatal(err)
+				}
+			},
+			binding: pluginFlagBinding{property: "value", names: []string{"value"}, kind: pluginFlagInt},
+		},
+		{name: "invalid positional", args: []string{"maybe"}, binding: pluginFlagBinding{property: "value", kind: pluginFlagBool, positional: true, positionalIndex: 0}},
+		{name: "invalid default", binding: pluginFlagBinding{property: "value", kind: pluginFlagInt, defaultProvided: true, defaultValue: "bad"}},
+		{
+			name:    "invalid env",
+			prepare: func(t *testing.T, _ *cobra.Command) { t.Setenv("PLUGIN_COVERAGE_BAD_ENV", "bad") },
+			binding: pluginFlagBinding{property: "value", kind: pluginFlagInt, envDefault: "PLUGIN_COVERAGE_BAD_ENV"},
+		},
+		{name: "missing named required", binding: pluginFlagBinding{property: "value", names: []string{"value"}, required: true}},
+		{name: "missing positional required", binding: pluginFlagBinding{property: "value", required: true, positional: true, positionalIndex: 0}},
+		{name: "required omitted", binding: pluginFlagBinding{property: "value", required: true, defaultProvided: true, defaultValue: "", omitWhen: "empty"}},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			cmd := &cobra.Command{Use: "leaf"}
+			if testCase.prepare != nil {
+				testCase.prepare(t, cmd)
+			}
+			if err := collectPluginBindings(cmd, testCase.args, []pluginFlagBinding{testCase.binding}, testCase.params); err == nil {
+				t.Fatal("expected binding error")
+			}
+		})
+	}
+
+	params := map[string]any{"value": ""}
+	if err := collectPluginBindings(&cobra.Command{Use: "leaf"}, nil, []pluginFlagBinding{{
+		property: "value", kind: pluginFlagString, omitWhen: "empty",
+	}}, params); err != nil {
+		t.Fatal(err)
+	}
+	if _, exists := params["value"]; exists {
+		t.Fatal("optional empty value was not omitted")
+	}
+}
+
+func TestPluginValueAndNamingHelpers(t *testing.T) {
+	parseCases := []struct {
+		kind pluginFlagKind
+		raw  string
+		want any
+	}{
+		{pluginFlagInt, " 2 ", 2},
+		{pluginFlagFloat, " 2.5 ", 2.5},
+		{pluginFlagBool, "true", true},
+		{pluginFlagStringSlice, "one, ,two", []string{"one", "two"}},
+		{pluginFlagJSON, `{"ok":true}`, map[string]any{"ok": true}},
+		{pluginFlagString, " raw ", " raw "},
+	}
+	for _, testCase := range parseCases {
+		got, err := parsePluginValue(testCase.raw, testCase.kind)
+		if err != nil || !reflect.DeepEqual(got, testCase.want) {
+			t.Fatalf("parse %q = (%#v, %v), want %#v", testCase.raw, got, err, testCase.want)
+		}
+	}
+	for _, testCase := range []struct {
+		kind pluginFlagKind
+		raw  string
+	}{
+		{pluginFlagInt, "bad"}, {pluginFlagFloat, "bad"}, {pluginFlagBool, "bad"}, {pluginFlagJSON, "{"},
+	} {
+		if _, err := parsePluginValue(testCase.raw, testCase.kind); err == nil {
+			t.Fatalf("invalid %q was accepted", testCase.raw)
+		}
+	}
+
+	omitCases := []struct {
+		value any
+		mode  string
+		want  bool
+	}{
+		{nil, "", true}, {" ", "", true}, {[]string{}, "", true},
+		{"", "never", false}, {false, "zero", true}, {0, "zero", true},
+		{float64(0), "zero", true}, {true, "zero", false}, {1, "zero", false},
+		{float64(1), "zero", false}, {[]any{}, "zero", true}, {map[string]any{}, "zero", true},
+		{[]any{"value"}, "zero", false}, {map[string]any{"value": true}, "zero", false},
+		{struct{}{}, "zero", false}, {false, "", false},
+	}
+	for _, testCase := range omitCases {
+		if got := shouldOmitPluginValue(testCase.value, testCase.mode); got != testCase.want {
+			t.Fatalf("omit (%#v, %q) = %v, want %v", testCase.value, testCase.mode, got, testCase.want)
+		}
+	}
+
+	wrapPluginParams(nil, "body")
+	untouched := map[string]any{"value": 1}
+	wrapPluginParams(untouched, " ")
+	wrapped := map[string]any{"body": map[string]any{"old": 1}, "value": 2, "_meta": 3}
+	wrapPluginParams(wrapped, "body")
+	wantWrapped := map[string]any{"body": map[string]any{"old": 1, "value": 2}, "_meta": 3}
+	if !reflect.DeepEqual(wrapped, wantWrapped) {
+		t.Fatalf("wrapped = %#v, want %#v", wrapped, wantWrapped)
+	}
+
+	kinds := map[string]pluginFlagKind{
+		"int": pluginFlagInt, "integer": pluginFlagInt,
+		"float": pluginFlagFloat, "float64": pluginFlagFloat, "number": pluginFlagFloat,
+		"bool": pluginFlagBool, "boolean": pluginFlagBool,
+		"stringSlice": pluginFlagStringSlice, "string_slice": pluginFlagStringSlice,
+		"array": pluginFlagStringSlice, "[]string": pluginFlagStringSlice,
+		"json": pluginFlagJSON, "object": pluginFlagJSON, "unknown": pluginFlagString,
+	}
+	for raw, want := range kinds {
+		if got := pluginFlagKindFromString(raw); got != want {
+			t.Fatalf("kind %q = %v, want %v", raw, got, want)
+		}
+	}
+
+	used := map[string]bool{}
+	reserved := map[string]bool{"r": true}
+	if got := safePluginShorthand(" x ", used, reserved); got != "x" || !used["x"] {
+		t.Fatalf("safe shorthand = %q / %#v", got, used)
+	}
+	for _, raw := range []string{"", "xy", "x", "r"} {
+		if got := safePluginShorthand(raw, used, reserved); got != "" {
+			t.Fatalf("unsafe shorthand %q = %q", raw, got)
+		}
+	}
+
+	baseReservations := pluginReservedFlags(nil)
+	root := &cobra.Command{Use: "dws"}
+	root.PersistentFlags().StringP("custom", "c", "", "")
+	rootReservations := pluginReservedFlags(root)
+	if !baseReservations.names["yes"] || !rootReservations.names["custom"] || !rootReservations.shorthands["c"] {
+		t.Fatalf("reservations = %#v / %#v", baseReservations, rootReservations)
+	}
+
+	if got := safePluginAliases([]string{"", "help", "auth", "cmd", "cmd", "ok", "Bad"}, "cmd"); !reflect.DeepEqual(got, []string{"ok"}) {
+		t.Fatalf("aliases = %#v", got)
+	}
+	if got := derivePluginCommandName("conference_getCurrent2Status", []string{"other", "conference"}); got != "get-current2-status" {
+		t.Fatalf("derived name = %q", got)
+	}
+	if got := pluginKebabName(" HTTP2.Foo_bar baz@ "); got != "http2-foo-bar-baz@" {
+		t.Fatalf("kebab name = %q", got)
+	}
+	for _, name := range []string{"", "1bad", "bad-", "bad--name", "bad_name", "bad@name"} {
+		if validPluginKebabName(name) {
+			t.Fatalf("invalid kebab name %q was accepted", name)
+		}
+	}
+	if !validPluginKebabName("good-name2") || validPluginCommandName("help") || validPluginFlagName("json") || validPluginFlagName("params") {
+		t.Fatal("name validation contract failed")
+	}
+	if got := firstNonEmptyPluginString(" ", " value "); got != "value" || firstNonEmptyPluginString("", " ") != "" {
+		t.Fatal("first non-empty string contract failed")
+	}
+}
+
+func TestPluginConstraintGroupAndRootHelpers(t *testing.T) {
+	cmd := &cobra.Command{Use: "leaf"}
+	for _, name := range []string{"a", "b", "c"} {
+		cmd.Flags().String(name, "", "")
+	}
+	applyPluginFlagConstraints(cmd, mcptypes.CLIToolOverride{
+		MutuallyExclusive: [][]string{{"a", "b"}, {"a", "missing"}},
+		RequireOneOf:      [][]string{{"a", "b"}, {"missing"}},
+		RequireTogether:   [][]string{{"b", "c"}, {"c", "missing"}},
+	})
+	bindings := []pluginFlagBinding{{names: []string{"a"}}, {names: []string{"b"}}, {names: []string{"c"}}}
+	if !validPluginFlagConstraints(bindings, mcptypes.CLIToolOverride{
+		MutuallyExclusive: [][]string{{"a", "b"}},
+		RequireOneOf:      [][]string{{"a"}},
+		RequireTogether:   [][]string{{"b", "c"}},
+	}) {
+		t.Fatal("valid plugin constraints were rejected")
+	}
+	for _, invalid := range []mcptypes.CLIToolOverride{
+		{MutuallyExclusive: [][]string{{"a"}}},
+		{RequireOneOf: [][]string{{"missing"}}},
+		{RequireTogether: [][]string{{"a", "a"}}},
+	} {
+		if validPluginFlagConstraints(bindings, invalid) {
+			t.Fatalf("invalid plugin constraints were accepted: %#v", invalid)
+		}
+	}
+
+	groups := map[string]*cobra.Command{}
+	root := &cobra.Command{Use: "root"}
+	group := ensurePluginGroup(root, "parent.child", "child description", groups)
+	if group.Name() != "child" || group.Short != "child description" || !cmdutil.IsPluginSourced(group) {
+		t.Fatalf("group = %#v", group)
+	}
+	if again := ensurePluginGroup(root, "parent.child", "ignored", groups); again != group {
+		t.Fatal("existing group was not reused")
+	}
+	for _, invalid := range []string{"safe.bad_name", "_bad", ".parent", "parent."} {
+		if got := ensurePluginGroup(root, invalid, "invalid", groups); got != nil {
+			t.Fatalf("invalid group path %q produced %#v", invalid, got)
+		}
+	}
+
+	mergePluginRoot(nil, root)
+	mergePluginRoot(root, nil)
+	destination := &cobra.Command{Use: "plugin", Aliases: []string{"one"}}
+	source := &cobra.Command{Use: "plugin", Aliases: []string{"one", "two"}}
+	source.AddCommand(&cobra.Command{Use: "leaf"})
+	mergePluginRoot(destination, source)
+	if !reflect.DeepEqual(destination.Aliases, []string{"one", "two"}) || requireOptionalPluginChild(destination, "leaf") == nil {
+		t.Fatalf("merged root = %#v", destination)
+	}
+
+	pruneEmptyPluginGroups(nil)
+	pruneRoot := &cobra.Command{Use: "root"}
+	empty := cobracmd.NewGroupCommand("empty", "empty")
+	nonEmpty := cobracmd.NewGroupCommand("non-empty", "non-empty")
+	nonEmpty.AddCommand(&cobra.Command{Use: "leaf"})
+	pruneRoot.AddCommand(empty, nonEmpty)
+	pruneEmptyPluginGroups(pruneRoot)
+	if requireOptionalPluginChild(pruneRoot, "empty") != nil || requireOptionalPluginChild(pruneRoot, "non-empty") == nil {
+		t.Fatal("empty plugin groups were not pruned correctly")
+	}
+
+	if pluginRootBoolFlag(nil, "yes") {
+		t.Fatal("nil command reported a root flag")
+	}
+	noFlag := &cobra.Command{Use: "root"}
+	if pluginRootBoolFlag(noFlag, "yes") {
+		t.Fatal("missing flag reported true")
+	}
+	wrongType := &cobra.Command{Use: "root"}
+	wrongType.PersistentFlags().String("yes", "true", "")
+	if pluginRootBoolFlag(wrongType, "yes") {
+		t.Fatal("wrong flag type reported true")
+	}
+	boolRoot := &cobra.Command{Use: "root"}
+	boolRoot.PersistentFlags().Bool("yes", false, "")
+	if err := boolRoot.PersistentFlags().Set("yes", "true"); err != nil {
+		t.Fatal(err)
+	}
+	if !pluginRootBoolFlag(boolRoot, "yes") {
+		t.Fatal("true root flag was not observed")
+	}
+	if err := pluginConfirmationRequired("dws plugin"); err == nil || !strings.Contains(err.Error(), "sensitive") {
+		t.Fatalf("confirmation error = %v", err)
+	}
+}
+
+func TestUnsupportedPluginSemanticsReportEveryField(t *testing.T) {
+	overlays := []struct {
+		value mcptypes.CLIOverlay
+		want  string
+	}{
+		{mcptypes.CLIOverlay{Parent: "root"}, "parent"},
+		{mcptypes.CLIOverlay{Group: "group"}, "group"},
+		{mcptypes.CLIOverlay{ServerDeps: []string{"other"}}, "serverDeps"},
+		{mcptypes.CLIOverlay{Hints: map[string]json.RawMessage{"x": json.RawMessage(`{}`)}}, "hintCommands"},
+		{mcptypes.CLIOverlay{RedirectTo: "other"}, "redirectTo"},
+		{mcptypes.CLIOverlay{}, ""},
+	}
+	for _, testCase := range overlays {
+		if got := unsupportedPluginOverlay(testCase.value); got != testCase.want {
+			t.Fatalf("unsupported overlay = %q, want %q", got, testCase.want)
+		}
+	}
+
+	tools := []struct {
+		value mcptypes.CLIToolOverride
+		want  string
+	}{
+		{mcptypes.CLIToolOverride{CLIAliases: []string{"x"}}, "cliAliases"},
+		{mcptypes.CLIToolOverride{OutputFormat: map[string]any{"x": true}}, "outputFormat"},
+		{mcptypes.CLIToolOverride{ServerOverride: "other"}, "serverOverride"},
+		{mcptypes.CLIToolOverride{RedirectTo: "x"}, "redirectTo"},
+		{mcptypes.CLIToolOverride{Pipeline: []json.RawMessage{json.RawMessage(`{}`)}}, "pipeline"},
+		{mcptypes.CLIToolOverride{}, ""},
+	}
+	for _, testCase := range tools {
+		if got := unsupportedPluginToolOverride(testCase.value); got != testCase.want {
+			t.Fatalf("unsupported tool = %q, want %q", got, testCase.want)
+		}
+	}
+
+	flags := []struct {
+		value mcptypes.CLIFlagOverride
+		want  string
+	}{
+		{mcptypes.CLIFlagOverride{MapsTo: "x"}, "mapsTo"},
+		{mcptypes.CLIFlagOverride{Transform: "x"}, "transform"},
+		{mcptypes.CLIFlagOverride{TransformArgs: map[string]any{"x": true}}, "transformArgs"},
+		{mcptypes.CLIFlagOverride{RuntimeDefault: "x"}, "runtimeDefault"},
+		{mcptypes.CLIFlagOverride{PipelineLocal: true}, "pipelineLocal"},
+		{mcptypes.CLIFlagOverride{Type: "mystery"}, "type"},
+		{mcptypes.CLIFlagOverride{OmitWhen: "sometimes"}, "omitWhen"},
+		{mcptypes.CLIFlagOverride{}, ""},
+	}
+	for _, testCase := range flags {
+		if got := unsupportedPluginFlagOverride(testCase.value); got != testCase.want {
+			t.Fatalf("unsupported flag = %q, want %q", got, testCase.want)
+		}
+	}
+	for _, value := range []string{"", "string", "integer", "float64", "boolean", "stringSlice", "array", "json", "object"} {
+		if !supportedPluginFlagType(value) {
+			t.Fatalf("supported plugin flag type %q was rejected", value)
+		}
+	}
+	for _, value := range []string{"", "empty", "zero", "never"} {
+		if !supportedPluginOmitMode(value) {
+			t.Fatalf("supported plugin omit mode %q was rejected", value)
+		}
+	}
+}
+
+func TestUnsupportedPluginDescriptorRejectsEveryInvalidLayer(t *testing.T) {
+	testCases := []struct {
+		name   string
+		mutate func(*mcptypes.ServerDescriptor)
+		want   string
+	}{
+		{name: "overlay", mutate: func(value *mcptypes.ServerDescriptor) { value.CLI.Parent = "root" }, want: "parent"},
+		{name: "no tools", mutate: func(value *mcptypes.ServerDescriptor) { value.CLI.ToolOverrides = nil }, want: ""},
+		{name: "root", mutate: func(value *mcptypes.ServerDescriptor) { value.CLI.Command = "Bad" }, want: "command"},
+		{name: "declared group", mutate: func(value *mcptypes.ServerDescriptor) {
+			value.CLI.Groups = map[string]mcptypes.CLIGroupDef{"bad_name": {}}
+		}, want: "groups"},
+		{name: "blank tool", mutate: func(value *mcptypes.ServerDescriptor) {
+			value.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{"": {}}
+		}, want: "tool"},
+		{name: "tool semantics", mutate: func(value *mcptypes.ServerDescriptor) {
+			value.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{"tool": {ServerOverride: "drive"}}
+		}, want: "serverOverride"},
+		{name: "hidden tool", mutate: func(value *mcptypes.ServerDescriptor) {
+			value.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{"tool": {Hidden: true, ServerOverride: "drive"}}
+		}, want: ""},
+		{name: "derived leaf", mutate: func(value *mcptypes.ServerDescriptor) {
+			value.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{"conference_derived_tool": {}}
+		}, want: ""},
+		{name: "leaf", mutate: func(value *mcptypes.ServerDescriptor) {
+			value.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{"tool": {CLIName: "Bad"}}
+		}, want: "cliName"},
+		{name: "leaf group", mutate: func(value *mcptypes.ServerDescriptor) {
+			value.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{"tool": {CLIName: "leaf", Group: "bad_name"}}
+		}, want: "group"},
+		{name: "flags", mutate: func(value *mcptypes.ServerDescriptor) {
+			value.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{"tool": {
+				CLIName: "leaf",
+				Flags:   map[string]mcptypes.CLIFlagOverride{"value": {Alias: "yes"}},
+			}}
+		}, want: "flags"},
+		{name: "constraints", mutate: func(value *mcptypes.ServerDescriptor) {
+			value.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{"tool": {
+				CLIName:         "leaf",
+				Flags:           map[string]mcptypes.CLIFlagOverride{"value": {}},
+				RequireTogether: [][]string{{"value", "missing"}},
+			}}
+		}, want: "constraints"},
+		{name: "valid", want: ""},
+	}
+	root := pluginTestRoot()
+	for _, testCase := range testCases {
+		t.Run(testCase.name, func(t *testing.T) {
+			descriptor := conferencePluginDescriptor()
+			if testCase.mutate != nil {
+				testCase.mutate(&descriptor)
+			}
+			if got := unsupportedPluginDescriptor(root, descriptor); got != testCase.want {
+				t.Fatalf("unsupported descriptor = %q, want %q", got, testCase.want)
+			}
+		})
+	}
+}

internal/app/plugin_commands_test.go

@@ -0,0 +1,809 @@
+// 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 (
+	"context"
+	"encoding/json"
+	"errors"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"reflect"
+	"strings"
+	"sync/atomic"
+	"testing"
+
+	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/pipeline"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/cmdutil"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/mcptypes"
+	"github.com/spf13/cobra"
+)
+
+type pluginCaptureRunner struct {
+	invocations []executor.Invocation
+}
+
+func (r *pluginCaptureRunner) Run(_ context.Context, invocation executor.Invocation) (executor.Result, error) {
+	r.invocations = append(r.invocations, invocation)
+	return executor.Result{Invocation: invocation}, nil
+}
+
+func conferencePluginDescriptor() mcptypes.ServerDescriptor {
+	return mcptypes.ServerDescriptor{
+		Key:         "conference-local",
+		DisplayName: "conference/conference-local",
+		Description: "conference plugin",
+		Endpoint:    "stdio://conference/conference-local",
+		Source:      "plugin",
+		HasCLIMeta:  true,
+		CLI: mcptypes.CLIOverlay{
+			ID:          "conference-local",
+			Command:     "conference",
+			Description: "视频会议：发起/邀请入会/会中控制",
+			Prefixes:    []string{"conference"},
+			Groups: map[string]mcptypes.CLIGroupDef{
+				"camera": {Description: "摄像头控制"},
+				"mic":    {Description: "麦克风控制"},
+				"share":  {Description: "屏幕共享"},
+			},
+			ToolOverrides: map[string]mcptypes.CLIToolOverride{
+				"create_conference": {
+					CLIName:     "start",
+					Description: "发起即时会议",
+					Flags: map[string]mcptypes.CLIFlagOverride{
+						"title": {Description: "会议标题"},
+					},
+				},
+				"get_conference_status": {
+					CLIName:     "status",
+					Description: "查询当前会议状态",
+				},
+				"ai_end_meeting_for_all": {
+					CLIName:     "end",
+					Description: "结束会议（所有人）",
+					IsSensitive: true,
+				},
+				"ai_open_camera": {
+					CLIName:     "open",
+					Group:       "camera",
+					Description: "打开摄像头",
+				},
+				"ai_mute_mic": {
+					CLIName:     "mute",
+					Group:       "mic",
+					Description: "静音自己",
+				},
+				"ai_share_desktop": {
+					CLIName:     "start",
+					Group:       "share",
+					Description: "开始共享桌面",
+					Flags: map[string]mcptypes.CLIFlagOverride{
+						"capture_speaker": {Description: "是否共享电脑音频"},
+					},
+				},
+			},
+		},
+	}
+}
+
+func pluginTestRoot(commands ...*cobra.Command) *cobra.Command {
+	root := &cobra.Command{
+		Use:           "dws",
+		SilenceErrors: true,
+		SilenceUsage:  true,
+	}
+	root.PersistentFlags().Bool("dry-run", false, "")
+	root.PersistentFlags().Bool("yes", false, "")
+	root.PersistentFlags().StringP("format", "f", "json", "")
+	root.SetOut(io.Discard)
+	root.SetErr(io.Discard)
+	root.AddCommand(commands...)
+	return root
+}
+
+func requirePluginChild(t *testing.T, parent *cobra.Command, names ...string) *cobra.Command {
+	t.Helper()
+	current := parent
+	for _, name := range names {
+		var next *cobra.Command
+		for _, child := range current.Commands() {
+			if child.Name() == name {
+				next = child
+				break
+			}
+		}
+		if next == nil {
+			t.Fatalf("missing plugin command %q below %q", name, current.CommandPath())
+		}
+		current = next
+	}
+	return current
+}
+
+func TestPluginOverlayBuildsConferenceTreeAndDispatchesOriginalProperties(t *testing.T) {
+	runner := &pluginCaptureRunner{}
+	commands := buildPluginCommands([]mcptypes.ServerDescriptor{conferencePluginDescriptor()}, runner, nil)
+	if len(commands) != 1 {
+		t.Fatalf("plugin roots = %d, want 1", len(commands))
+	}
+	conference := commands[0]
+	if conference.Name() != "conference" || conference.Short != "视频会议：发起/邀请入会/会中控制" {
+		t.Fatalf("conference root = %q / %q", conference.Name(), conference.Short)
+	}
+	if !cmdutil.IsPluginSourced(conference) {
+		t.Fatal("conference root is missing plugin provenance")
+	}
+	if got := requirePluginChild(t, conference, "camera").Short; got != "摄像头控制" {
+		t.Fatalf("camera group short = %q", got)
+	}
+	if got := requirePluginChild(t, conference, "camera", "open").Short; got != "打开摄像头" {
+		t.Fatalf("camera open short = %q", got)
+	}
+	requirePluginChild(t, conference, "mic", "mute")
+	requirePluginChild(t, conference, "status")
+	share := requirePluginChild(t, conference, "share", "start")
+	flag := share.Flags().Lookup("capture-speaker")
+	if flag == nil || flag.Usage != "是否共享电脑音频" {
+		t.Fatalf("capture-speaker flag = %#v", flag)
+	}
+
+	root := pluginTestRoot(commands...)
+	root.SetArgs([]string{
+		"conference", "start",
+		"--json", `{"from_json":"kept","title":"json"}`,
+		"--params", `{"from_params":2,"title":"params"}`,
+		"--title", "验证会议",
+		"--dry-run",
+	})
+	if err := root.Execute(); err != nil {
+		t.Fatalf("conference start: %v", err)
+	}
+	if len(runner.invocations) != 1 {
+		t.Fatalf("runner calls = %d, want 1", len(runner.invocations))
+	}
+	invocation := runner.invocations[0]
+	if invocation.Kind != "compat_invocation" ||
+		invocation.CanonicalProduct != "conference-local" ||
+		invocation.Tool != "create_conference" ||
+		!invocation.DryRun {
+		t.Fatalf("conference invocation = %#v", invocation)
+	}
+	wantParams := map[string]any{
+		"from_json":   "kept",
+		"from_params": float64(2),
+		"title":       "验证会议",
+	}
+	if !reflect.DeepEqual(invocation.Params, wantParams) {
+		t.Fatalf("conference params = %#v, want %#v", invocation.Params, wantParams)
+	}
+
+	precedenceRunner := &pluginCaptureRunner{}
+	precedenceRoot := pluginTestRoot(buildPluginCommands(
+		[]mcptypes.ServerDescriptor{conferencePluginDescriptor()},
+		precedenceRunner,
+		nil,
+	)...)
+	precedenceRoot.SetArgs([]string{
+		"conference", "start",
+		"--json", `{"title":"json"}`,
+		"--params", `{"title":"params"}`,
+		"--dry-run",
+	})
+	if err := precedenceRoot.Execute(); err != nil {
+		t.Fatalf("conference payload precedence: %v", err)
+	}
+	if got := precedenceRunner.invocations[0].Params["title"]; got != "params" {
+		t.Fatalf("conference payload title = %#v, want --params value", got)
+	}
+}
+
+func TestPluginOverlayTypedFlags(t *testing.T) {
+	descriptor := conferencePluginDescriptor()
+	descriptor.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+		"typed_tool": {
+			CLIName: "typed",
+			Flags: map[string]mcptypes.CLIFlagOverride{
+				"conversationId": {Required: true, Description: "conversation"},
+				"enabled":        {Type: "bool"},
+				"limit":          {Type: "int"},
+				"tags":           {Type: "stringSlice"},
+			},
+		},
+	}
+	runner := &pluginCaptureRunner{}
+	root := pluginTestRoot(buildPluginCommands([]mcptypes.ServerDescriptor{descriptor}, runner, nil)...)
+	root.SetArgs([]string{
+		"conference", "typed",
+		"--conversation-id", "cid",
+		"--enabled=false",
+		"--limit", "3",
+		"--tags", "one,two",
+		"--dry-run",
+	})
+	if err := root.Execute(); err != nil {
+		t.Fatalf("typed plugin command: %v", err)
+	}
+	if len(runner.invocations) != 1 {
+		t.Fatalf("runner calls = %d", len(runner.invocations))
+	}
+	invocation := runner.invocations[0]
+	if invocation.CanonicalProduct != "conference-local" {
+		t.Fatalf("canonical product = %q", invocation.CanonicalProduct)
+	}
+	want := map[string]any{
+		"conversationId": "cid",
+		"enabled":        false,
+		"limit":          3,
+		"tags":           []string{"one", "two"},
+	}
+	if !reflect.DeepEqual(invocation.Params, want) {
+		t.Fatalf("typed params = %#v, want %#v", invocation.Params, want)
+	}
+}
+
+func TestPluginSensitiveCommandRequiresConfirmation(t *testing.T) {
+	for _, testCase := range []struct {
+		name      string
+		args      []string
+		wantCalls int
+		wantDry   bool
+		wantError bool
+	}{
+		{name: "blocked", args: []string{"conference", "end"}, wantError: true},
+		{name: "preview", args: []string{"conference", "end", "--dry-run"}, wantCalls: 1, wantDry: true},
+		{name: "confirmed", args: []string{"conference", "end", "--yes"}, wantCalls: 1},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			runner := &pluginCaptureRunner{}
+			root := pluginTestRoot(buildPluginCommands(
+				[]mcptypes.ServerDescriptor{conferencePluginDescriptor()}, runner, nil)...)
+			root.SetArgs(testCase.args)
+			err := root.Execute()
+			if testCase.wantError {
+				var appErr *apperrors.Error
+				if !errors.As(err, &appErr) ||
+					appErr.Category != apperrors.CategoryValidation ||
+					appErr.Reason != "confirmation_required" {
+					t.Fatalf("sensitive error = %#v", err)
+				}
+			} else if err != nil {
+				t.Fatalf("sensitive command: %v", err)
+			}
+			if len(runner.invocations) != testCase.wantCalls {
+				t.Fatalf("runner calls = %d, want %d", len(runner.invocations), testCase.wantCalls)
+			}
+			if testCase.wantCalls == 1 && runner.invocations[0].DryRun != testCase.wantDry {
+				t.Fatalf("dry-run = %v, want %v", runner.invocations[0].DryRun, testCase.wantDry)
+			}
+		})
+	}
+}
+
+func TestPluginOverlayMergesServersWithoutProbingHTTP(t *testing.T) {
+	isolatePluginRuntime(t)
+	var calls atomic.Int32
+	server := httptest.NewServer(http.HandlerFunc(func(http.ResponseWriter, *http.Request) {
+		calls.Add(1)
+	}))
+	defer server.Close()
+
+	first := conferencePluginDescriptor()
+	first.Endpoint = server.URL
+	first.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+		"one": {CLIName: "one"},
+	}
+	second := first
+	second.Key = "conference-extra"
+	second.DisplayName = "conference/conference-extra"
+	second.Endpoint = server.URL + "/extra"
+	second.CLI.ID = "conference-extra"
+	second.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+		"two": {CLIName: "two"},
+	}
+	registerPluginHTTPServer(first)
+	registerPluginHTTPServer(second)
+	runner := &pluginCaptureRunner{}
+	commands := buildPluginCommands([]mcptypes.ServerDescriptor{second, first}, runner, nil)
+	if len(commands) != 1 {
+		t.Fatalf("merged roots = %d, want 1", len(commands))
+	}
+	requirePluginChild(t, commands[0], "one")
+	requirePluginChild(t, commands[0], "two")
+	root := pluginTestRoot(commands...)
+	root.SetArgs([]string{"conference", "--help"})
+	if err := root.Execute(); err != nil {
+		t.Fatalf("conference help: %v", err)
+	}
+	if got := calls.Load(); got != 0 {
+		t.Fatalf("HTTP calls while building help = %d, want 0", got)
+	}
+	for _, command := range []string{"one", "two"} {
+		root.SetArgs([]string{"conference", command, "--dry-run"})
+		if err := root.Execute(); err != nil {
+			t.Fatalf("conference %s: %v", command, err)
+		}
+	}
+	if len(runner.invocations) != 2 ||
+		runner.invocations[0].CanonicalProduct != "conference-local" ||
+		runner.invocations[1].CanonicalProduct != "conference-extra" {
+		t.Fatalf("merged routes = %#v", runner.invocations)
+	}
+}
+
+func TestPluginCanReplaceHiddenFallbackButNotVisibleDistributionCommand(t *testing.T) {
+	root := &cobra.Command{Use: "dws"}
+	fallback := &cobra.Command{Use: "conference", Hidden: true}
+	fallback.AddCommand(&cobra.Command{Use: "meeting"})
+	distribution := &cobra.Command{Use: "drive"}
+	root.AddCommand(fallback, distribution)
+
+	conference := buildPluginCommands(
+		[]mcptypes.ServerDescriptor{conferencePluginDescriptor()},
+		executor.EchoRunner{},
+		nil,
+	)[0]
+	drive := &cobra.Command{Use: "drive"}
+	cmdutil.MarkPluginSource(drive)
+	addPluginCommandsSafe(root, []*cobra.Command{conference, drive})
+
+	gotConference := requirePluginChild(t, root, "conference")
+	if gotConference == fallback || gotConference.Hidden {
+		t.Fatalf("conference fallback was not replaced: %#v", gotConference)
+	}
+	requirePluginChild(t, gotConference, "status")
+	if gotDrive := requirePluginChild(t, root, "drive"); gotDrive != distribution {
+		t.Fatal("visible distribution command was replaced by a plugin")
+	}
+}
+
+func TestConflictingPluginDescriptorCannotReplaceDistributionEndpoint(t *testing.T) {
+	isolatePluginRuntime(t)
+	configDir := t.TempDir()
+	t.Setenv("DWS_CONFIG_DIR", configDir)
+	pluginDir := filepath.Join(configDir, "plugins", "user", "drive-hijack")
+	if err := os.MkdirAll(pluginDir, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	manifest := `{
+		"name":"drive-hijack",
+		"version":"1.0.0",
+		"mcpServers":{
+			"drive":{
+				"type":"streamable-http",
+				"endpoint":"https://plugin.invalid/mcp",
+				"cli":{
+					"id":"drive-service",
+					"command":"drive-hijack",
+					"toolOverrides":{"plugin_tool":{"cliName":"plugin-tool"}}
+				}
+			}
+		}
+	}`
+	if err := os.WriteFile(filepath.Join(pluginDir, "plugin.json"), []byte(manifest), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	AppendDynamicServer(mcptypes.ServerDescriptor{
+		Key:      "drive",
+		Endpoint: "https://distribution.invalid/mcp",
+		CLI:      mcptypes.CLIOverlay{ID: "drive-service", Command: "drive"},
+	})
+	root := &cobra.Command{Use: "dws"}
+	root.AddCommand(&cobra.Command{Use: "drive"})
+	if commands := loadPlugins(root, nil, executor.EchoRunner{}); len(commands) != 0 {
+		t.Fatalf("conflicting plugin commands = %#v", commands)
+	}
+	if endpoint, ok := directRuntimeEndpoint("drive-service", "plugin_tool"); !ok ||
+		endpoint != "https://distribution.invalid/mcp" {
+		t.Fatalf("drive endpoint after rejected plugin = (%q, %v)", endpoint, ok)
+	}
+}
+
+func TestSchemaSourceRootDoesNotLoadRuntimePlugins(t *testing.T) {
+	isolatePluginRuntime(t)
+	previous := rootLoadPlugins
+	t.Cleanup(func() { rootLoadPlugins = previous })
+	var calls atomic.Int32
+	rootLoadPlugins = func(*cobra.Command, *pipeline.Engine, executor.Runner) []*cobra.Command {
+		calls.Add(1)
+		AppendDynamicServer(conferencePluginDescriptor())
+		return buildPluginCommands(
+			[]mcptypes.ServerDescriptor{conferencePluginDescriptor()},
+			executor.EchoRunner{},
+			nil,
+		)
+	}
+
+	base := NewSchemaSourceRootCommand()
+	if calls.Load() != 0 {
+		t.Fatalf("Schema source root loaded plugins %d times", calls.Load())
+	}
+	baseConference := requirePluginChild(t, base, "conference")
+	if !baseConference.Hidden || requireOptionalPluginChild(baseConference, "status") != nil {
+		t.Fatal("Schema source root contains installed conference plugin commands")
+	}
+
+	runtime := NewRootCommand()
+	if calls.Load() != 1 {
+		t.Fatalf("runtime root plugin loads = %d, want 1", calls.Load())
+	}
+	runtimeConference := requirePluginChild(t, runtime, "conference")
+	if runtimeConference.Hidden {
+		t.Fatal("runtime conference plugin is hidden")
+	}
+	requirePluginChild(t, runtimeConference, "status")
+}
+
+func requireOptionalPluginChild(parent *cobra.Command, name string) *cobra.Command {
+	for _, child := range parent.Commands() {
+		if child.Name() == name {
+			return child
+		}
+	}
+	return nil
+}
+
+func TestPluginDerivedNamesAndReservedAliases(t *testing.T) {
+	descriptor := conferencePluginDescriptor()
+	descriptor.CLI.Aliases = []string{"auth", "conf", "conf"}
+	descriptor.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+		"conference_getCurrentStatus": {},
+	}
+	commands := buildPluginCommands([]mcptypes.ServerDescriptor{descriptor}, executor.EchoRunner{}, nil)
+	if len(commands) != 1 || !reflect.DeepEqual(commands[0].Aliases, []string{"conf"}) {
+		t.Fatalf("plugin aliases = %#v", commands)
+	}
+	if requireOptionalPluginChild(commands[0], "get-current-status") == nil {
+		var names []string
+		for _, command := range commands[0].Commands() {
+			names = append(names, command.Name())
+		}
+		t.Fatalf("derived command missing, got %s", strings.Join(names, ", "))
+	}
+}
+
+func TestPluginFlagsCannotShadowHostControls(t *testing.T) {
+	host := pluginTestRoot()
+	host.PersistentFlags().StringP("host-extra", "x", "", "")
+	reservations := pluginReservedFlags(host)
+	for name := range reservations.names {
+		t.Run(name, func(t *testing.T) {
+			descriptor := conferencePluginDescriptor()
+			descriptor.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+				"unsafe": {
+					CLIName:     "unsafe",
+					IsSensitive: true,
+					Flags: map[string]mcptypes.CLIFlagOverride{
+						"value": {Alias: name},
+					},
+				},
+			}
+			if commands := buildPluginCommands(
+				[]mcptypes.ServerDescriptor{descriptor},
+				executor.EchoRunner{},
+				host,
+			); len(commands) != 0 {
+				t.Fatalf("reserved host flag %q produced commands %#v", name, commands)
+			}
+		})
+	}
+}
+
+func TestPluginShorthandsCannotShadowHostOrHelp(t *testing.T) {
+	host := pluginTestRoot()
+	host.PersistentFlags().StringP("host-extra", "x", "", "")
+	descriptor := conferencePluginDescriptor()
+	descriptor.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+		"safe": {
+			CLIName: "safe",
+			Flags: map[string]mcptypes.CLIFlagOverride{
+				"alpha":   {Shorthand: "f"},
+				"bravo":   {Shorthand: "h"},
+				"charlie": {Shorthand: "o"},
+				"delta":   {Shorthand: "v"},
+				"echo":    {Shorthand: "x"},
+				"foxtrot": {Shorthand: "y"},
+			},
+		},
+	}
+	runner := &pluginCaptureRunner{}
+	commands := buildPluginCommands(
+		[]mcptypes.ServerDescriptor{descriptor},
+		runner,
+		host,
+	)
+	if len(commands) != 1 {
+		t.Fatalf("plugin commands = %#v", commands)
+	}
+	host.AddCommand(commands...)
+	leaf := requirePluginChild(t, commands[0], "safe")
+	for _, name := range []string{"alpha", "bravo", "charlie", "delta", "echo", "foxtrot"} {
+		if shorthand := leaf.Flags().Lookup(name).Shorthand; shorthand != "" {
+			t.Fatalf("--%s shorthand = %q, want empty", name, shorthand)
+		}
+	}
+	host.SetArgs([]string{"conference", "safe", "-h"})
+	if err := host.Execute(); err != nil {
+		t.Fatalf("plugin help: %v", err)
+	}
+	if len(runner.invocations) != 0 {
+		t.Fatalf("help executed plugin: %#v", runner.invocations)
+	}
+}
+
+func TestPluginPayloadPrecedenceRequiredAndTypedPositionals(t *testing.T) {
+	descriptor := conferencePluginDescriptor()
+	descriptor.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+		"payload": {
+			CLIName: "payload",
+			Flags: map[string]mcptypes.CLIFlagOverride{
+				"title":   {Required: true},
+				"mode":    {Default: "fallback"},
+				"enabled": {Positional: true, PositionalIndex: 0, Alias: "enabled-value", Required: true, Type: "bool"},
+			},
+		},
+	}
+
+	for _, testCase := range []struct {
+		name        string
+		args        []string
+		wantEnabled bool
+	}{
+		{
+			name: "flag satisfies dual positional",
+			args: []string{
+				"conference", "payload",
+				"--params", `{"title":"from-json","mode":"from-json"}`,
+				"--enabled-value=true",
+				"--dry-run",
+			},
+			wantEnabled: true,
+		},
+		{
+			name: "json beats positional",
+			args: []string{
+				"conference", "payload", "true",
+				"--params", `{"title":"from-json","mode":"from-json","enabled":false}`,
+				"--dry-run",
+			},
+			wantEnabled: false,
+		},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			runner := &pluginCaptureRunner{}
+			root := pluginTestRoot(buildPluginCommands(
+				[]mcptypes.ServerDescriptor{descriptor},
+				runner,
+				nil,
+			)...)
+			root.SetArgs(testCase.args)
+			if err := root.Execute(); err != nil {
+				t.Fatalf("payload command: %v", err)
+			}
+			if len(runner.invocations) != 1 {
+				t.Fatalf("runner calls = %d", len(runner.invocations))
+			}
+			params := runner.invocations[0].Params
+			if params["title"] != "from-json" ||
+				params["mode"] != "from-json" ||
+				params["enabled"] != testCase.wantEnabled {
+				t.Fatalf("payload params = %#v", params)
+			}
+		})
+	}
+}
+
+func TestPluginDescriptorWinnerKeepsRouteAuthAndClientAtomic(t *testing.T) {
+	isolatePluginRuntime(t)
+	configDir := t.TempDir()
+	t.Setenv("DWS_CONFIG_DIR", configDir)
+
+	writeManifest := func(name, manifest string) {
+		t.Helper()
+		directory := filepath.Join(configDir, "plugins", "user", name)
+		if err := os.MkdirAll(directory, 0o755); err != nil {
+			t.Fatal(err)
+		}
+		if err := os.WriteFile(filepath.Join(directory, "plugin.json"), []byte(manifest), 0o600); err != nil {
+			t.Fatal(err)
+		}
+	}
+	writeManifest("alpha-plugin", `{
+		"name":"alpha-plugin",
+		"version":"1.0.0",
+		"mcpServers":{
+			"alpha":{
+				"type":"streamable-http",
+				"endpoint":"https://alpha.invalid/mcp",
+				"headers":{"Authorization":"Bearer alpha-secret"},
+				"cli":{
+					"id":"shared-plugin-id",
+					"command":"alpha-command",
+					"toolOverrides":{"alpha_tool":{"cliName":"alpha"}}
+				}
+			},
+			"alpha-extra":{
+				"type":"streamable-http",
+				"endpoint":"https://alpha-extra.invalid/mcp",
+				"cli":{
+					"id":"alpha-extra-id",
+					"command":"alpha-command",
+					"toolOverrides":{"extra_tool":{"cliName":"extra"}}
+				}
+			}
+		}
+	}`)
+	writeManifest("beta-plugin", `{
+		"name":"beta-plugin",
+		"version":"1.0.0",
+		"mcpServers":{
+			"beta":{
+				"type":"stdio",
+				"command":"bin/beta",
+				"cli":{
+					"id":"shared-plugin-id",
+					"command":"beta-command",
+					"toolOverrides":{"beta_tool":{"cliName":"beta"}}
+				}
+			}
+		}
+	}`)
+
+	root := pluginTestRoot()
+	commands := loadPlugins(root, nil, executor.EchoRunner{})
+	if len(commands) != 1 || commands[0].Name() != "alpha-command" {
+		t.Fatalf("plugin winner commands = %#v", commands)
+	}
+	requirePluginChild(t, commands[0], "alpha")
+	requirePluginChild(t, commands[0], "extra")
+	endpoint, ok := directRuntimeEndpoint("shared-plugin-id", "alpha_tool")
+	if !ok || endpoint != "https://alpha.invalid/mcp" {
+		t.Fatalf("winner endpoint = (%q, %v)", endpoint, ok)
+	}
+	extraEndpoint, ok := directRuntimeEndpoint("alpha-extra-id", "extra_tool")
+	if !ok || extraEndpoint != "https://alpha-extra.invalid/mcp" {
+		t.Fatalf("merged server endpoint = (%q, %v)", extraEndpoint, ok)
+	}
+	auth, ok := LookupPluginAuth("shared-plugin-id")
+	if !ok || auth.Token != "alpha-secret" {
+		t.Fatalf("winner auth = (%#v, %v)", auth, ok)
+	}
+	if _, ok := LookupStdioClient("beta-plugin/beta"); ok {
+		t.Fatal("losing stdio client was registered")
+	}
+}
+
+func TestUnsupportedPluginOverlaySemanticsFailClosed(t *testing.T) {
+	for _, mutate := range []func(*mcptypes.ServerDescriptor){
+		func(descriptor *mcptypes.ServerDescriptor) {
+			descriptor.CLI.RedirectTo = "drive"
+		},
+		func(descriptor *mcptypes.ServerDescriptor) {
+			descriptor.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+				"unsafe": {
+					CLIName: "unsafe",
+					Flags: map[string]mcptypes.CLIFlagOverride{
+						"source": {MapsTo: "target"},
+					},
+				},
+			}
+		},
+		func(descriptor *mcptypes.ServerDescriptor) {
+			descriptor.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+				"unsafe": {
+					CLIName:  "unsafe",
+					Pipeline: []json.RawMessage{json.RawMessage(`{"tool":"one"}`)},
+				},
+			}
+		},
+		func(descriptor *mcptypes.ServerDescriptor) {
+			descriptor.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+				"unsafe": {CLIName: "unsafe", ServerOverride: "drive"},
+			}
+		},
+		func(descriptor *mcptypes.ServerDescriptor) {
+			descriptor.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+				"unsafe": {
+					CLIName: "unsafe",
+					Flags: map[string]mcptypes.CLIFlagOverride{
+						"Body.query": {},
+					},
+				},
+			}
+		},
+		func(descriptor *mcptypes.ServerDescriptor) {
+			descriptor.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+				"unsafe": {CLIName: "unsafe", Group: "safe.bad_name"},
+			}
+		},
+		func(descriptor *mcptypes.ServerDescriptor) {
+			descriptor.CLI.ToolOverrides = map[string]mcptypes.CLIToolOverride{
+				"unsafe": {
+					CLIName:         "unsafe",
+					Flags:           map[string]mcptypes.CLIFlagOverride{"value": {}},
+					RequireTogether: [][]string{{"value", "missing"}},
+				},
+			}
+		},
+	} {
+		descriptor := conferencePluginDescriptor()
+		mutate(&descriptor)
+		if commands := buildPluginCommands(
+			[]mcptypes.ServerDescriptor{descriptor},
+			executor.EchoRunner{},
+			nil,
+		); len(commands) != 0 {
+			t.Fatalf("unsupported overlay produced commands %#v", commands)
+		}
+	}
+}
+
+func TestUnsupportedPluginDescriptorsDoNotRegisterRuntimeState(t *testing.T) {
+	isolatePluginRuntime(t)
+	configDir := t.TempDir()
+	t.Setenv("DWS_CONFIG_DIR", configDir)
+
+	writeManifest := func(name, manifest string) {
+		t.Helper()
+		directory := filepath.Join(configDir, "plugins", "user", name)
+		if err := os.MkdirAll(directory, 0o755); err != nil {
+			t.Fatal(err)
+		}
+		if err := os.WriteFile(filepath.Join(directory, "plugin.json"), []byte(manifest), 0o600); err != nil {
+			t.Fatal(err)
+		}
+	}
+	writeManifest("unsafe-http", `{
+		"name":"unsafe-http",
+		"version":"1.0.0",
+		"mcpServers":{"unsafe":{
+			"type":"streamable-http",
+			"endpoint":"https://unsafe.invalid/mcp",
+			"headers":{"Authorization":"Bearer unsafe-secret"},
+			"cli":{"id":"unsafe-http-id","command":"unsafe-http","toolOverrides":{
+				"unsafe_tool":{"cliName":"run","serverOverride":"drive"}
+			}}
+		}}
+	}`)
+	writeManifest("unsafe-stdio", `{
+		"name":"unsafe-stdio",
+		"version":"1.0.0",
+		"mcpServers":{"unsafe":{
+			"type":"stdio",
+			"command":"bin/unsafe",
+			"cli":{"id":"unsafe-stdio-id","command":"unsafe-stdio","toolOverrides":{
+				"unsafe_tool":{"cliName":"run","flags":{"value":{"mapsTo":"target"}}}
+			}}
+		}}
+	}`)
+
+	root := pluginTestRoot()
+	if commands := loadPlugins(root, nil, executor.EchoRunner{}); len(commands) != 0 {
+		t.Fatalf("unsupported plugin descriptors produced commands %#v", commands)
+	}
+	if endpoint, ok := directRuntimeEndpoint("unsafe-http-id", "unsafe_tool"); ok {
+		t.Fatalf("unsupported HTTP descriptor registered endpoint %q", endpoint)
+	}
+	if _, ok := LookupPluginAuth("unsafe-http-id"); ok {
+		t.Fatal("unsupported HTTP descriptor registered plugin auth")
+	}
+	if _, ok := LookupStdioClient("unsafe-stdio/unsafe"); ok {
+		t.Fatal("unsupported stdio descriptor registered a client")
+	}
+}

internal/app/plugin_input_schema.go

@@ -0,0 +1,239 @@
+// 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 (
+	"encoding/json"
+	"fmt"
+	"strconv"
+	"strings"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/cli"
+	apperrors "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/errors"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/transport"
+)
+
+func pluginToolInputSchema(
+	tools transport.ToolsListResult,
+	toolName string,
+) (map[string]any, bool) {
+	for _, tool := range tools.Tools {
+		if strings.TrimSpace(tool.Name) == strings.TrimSpace(toolName) {
+			return tool.InputSchema, true
+		}
+	}
+	return nil, false
+}
+
+func normalizePluginInputParams(
+	params map[string]any,
+	schema map[string]any,
+) (map[string]any, error) {
+	schema = canonicalPluginInputSchema(schema)
+	normalized := make(map[string]any, len(params))
+	for key, value := range params {
+		normalized[key] = value
+	}
+	if _, err := coercePluginSchemaValue(normalized, schema); err != nil {
+		return nil, cliInputValidationError(err)
+	}
+	if err := cli.ValidateInputSchema(normalized, schema); err != nil {
+		return nil, err
+	}
+	return normalized, nil
+}
+
+func canonicalPluginInputSchema(schema map[string]any) map[string]any {
+	if len(schema) == 0 {
+		return schema
+	}
+	cloned := make(map[string]any, len(schema))
+	for key, value := range schema {
+		cloned[key] = clonePluginSchemaValue(key, value)
+	}
+	return cloned
+}
+
+func clonePluginSchemaValue(key string, value any) any {
+	switch typed := value.(type) {
+	case map[string]any:
+		cloned := make(map[string]any, len(typed))
+		for childKey, childValue := range typed {
+			cloned[childKey] = clonePluginSchemaValue(childKey, childValue)
+		}
+		return cloned
+	case []any:
+		cloned := make([]any, len(typed))
+		for index, item := range typed {
+			cloned[index] = clonePluginSchemaValue(key, item)
+		}
+		return cloned
+	case []string:
+		cloned := make([]string, len(typed))
+		for index, item := range typed {
+			if key == "type" {
+				item = canonicalPluginSchemaType(item)
+			}
+			cloned[index] = item
+		}
+		return cloned
+	case string:
+		if key == "type" {
+			return canonicalPluginSchemaType(typed)
+		}
+		return typed
+	default:
+		return value
+	}
+}
+
+func canonicalPluginSchemaType(value string) string {
+	switch strings.ToLower(strings.TrimSpace(value)) {
+	case "bool":
+		return "boolean"
+	case "int":
+		return "integer"
+	case "float":
+		return "number"
+	default:
+		return value
+	}
+}
+
+func cliInputValidationError(err error) error {
+	if err == nil {
+		return nil
+	}
+	return apperrors.NewValidation(
+		fmt.Sprintf("input schema normalization failed: %v", err),
+		apperrors.WithReason("plugin_input_schema_invalid"),
+	)
+}
+
+func coercePluginSchemaValue(value any, schema map[string]any) (any, error) {
+	target := singlePluginSchemaType(schema)
+	if raw, ok := value.(string); ok {
+		trimmed := strings.TrimSpace(raw)
+		switch target {
+		case "bool", "boolean":
+			parsed, err := strconv.ParseBool(trimmed)
+			if err != nil {
+				return nil, fmt.Errorf("cannot convert %q to boolean: %w", raw, err)
+			}
+			value = parsed
+		case "int", "integer":
+			parsed, err := strconv.Atoi(trimmed)
+			if err != nil {
+				return nil, fmt.Errorf("cannot convert %q to integer: %w", raw, err)
+			}
+			value = parsed
+		case "float", "number":
+			parsed, err := strconv.ParseFloat(trimmed, 64)
+			if err != nil {
+				return nil, fmt.Errorf("cannot convert %q to number: %w", raw, err)
+			}
+			value = parsed
+		case "object":
+			var parsed map[string]any
+			if err := json.Unmarshal([]byte(trimmed), &parsed); err != nil {
+				return nil, fmt.Errorf("cannot convert plugin parameter to object: %w", err)
+			}
+			if parsed == nil {
+				return nil, fmt.Errorf("cannot convert plugin parameter to object: expected a JSON object")
+			}
+			value = parsed
+		case "array":
+			var parsed []any
+			if strings.HasPrefix(trimmed, "[") {
+				if err := json.Unmarshal([]byte(trimmed), &parsed); err != nil {
+					return nil, fmt.Errorf("cannot convert plugin parameter to array: %w", err)
+				}
+			} else if trimmed != "" {
+				for _, item := range strings.Split(trimmed, ",") {
+					if item = strings.TrimSpace(item); item != "" {
+						parsed = append(parsed, item)
+					}
+				}
+			}
+			value = parsed
+		}
+	}
+
+	switch typed := value.(type) {
+	case map[string]any:
+		properties, _ := schema["properties"].(map[string]any)
+		for key, propertyValue := range typed {
+			propertySchema, _ := properties[key].(map[string]any)
+			if len(propertySchema) == 0 {
+				continue
+			}
+			coerced, err := coercePluginSchemaValue(propertyValue, propertySchema)
+			if err != nil {
+				return nil, fmt.Errorf("%s: %w", key, err)
+			}
+			typed[key] = coerced
+		}
+		return typed, nil
+	case []string:
+		items := make([]any, len(typed))
+		for index, item := range typed {
+			items[index] = item
+		}
+		value = items
+	}
+
+	if items, ok := value.([]any); ok {
+		itemSchema, _ := schema["items"].(map[string]any)
+		if len(itemSchema) == 0 {
+			return items, nil
+		}
+		for index, item := range items {
+			coerced, err := coercePluginSchemaValue(item, itemSchema)
+			if err != nil {
+				return nil, fmt.Errorf("item %d: %w", index, err)
+			}
+			items[index] = coerced
+		}
+		return items, nil
+	}
+	return value, nil
+}
+
+func singlePluginSchemaType(schema map[string]any) string {
+	var types []string
+	switch typed := schema["type"].(type) {
+	case string:
+		types = []string{typed}
+	case []string:
+		types = typed
+	case []any:
+		for _, value := range typed {
+			if text, ok := value.(string); ok {
+				types = append(types, text)
+			}
+		}
+	}
+	var target string
+	for _, candidate := range types {
+		candidate = strings.TrimSpace(candidate)
+		if candidate == "" || candidate == "null" {
+			continue
+		}
+		if target != "" && target != candidate {
+			return ""
+		}
+		target = candidate
+	}
+	return target
+}

internal/app/plugin_input_schema_coverage_test.go

@@ -0,0 +1,229 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0
+
+package app
+
+import (
+	"errors"
+	"reflect"
+	"strings"
+	"testing"
+
+	apperrors "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/errors"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/transport"
+)
+
+func TestPluginToolInputSchemaMatchesTrimmedName(t *testing.T) {
+	want := map[string]any{"type": "object"}
+	tools := transport.ToolsListResult{Tools: []transport.ToolDescriptor{
+		{Name: "other", InputSchema: map[string]any{"type": "string"}},
+		{Name: "  create_conference  ", InputSchema: want},
+	}}
+
+	got, ok := pluginToolInputSchema(tools, " create_conference ")
+	if !ok || !reflect.DeepEqual(got, want) {
+		t.Fatalf("pluginToolInputSchema() = (%#v, %v), want (%#v, true)", got, ok, want)
+	}
+	if got, ok := pluginToolInputSchema(tools, "missing"); ok || got != nil {
+		t.Fatalf("missing pluginToolInputSchema() = (%#v, %v), want (nil, false)", got, ok)
+	}
+}
+
+func TestNormalizePluginInputParamsCoercesNestedValues(t *testing.T) {
+	schema := map[string]any{
+		"type":     "object",
+		"required": []string{"enabled"},
+		"properties": map[string]any{
+			"enabled": map[string]any{"type": []any{"null", "bool"}},
+			"count":   map[string]any{"type": "int"},
+			"ratio":   map[string]any{"type": "float"},
+			"settings": map[string]any{
+				"type": "object",
+				"properties": map[string]any{
+					"active": map[string]any{"type": "bool"},
+				},
+			},
+			"ids": map[string]any{
+				"type":  []string{"array", "null"},
+				"items": map[string]any{"type": "int"},
+			},
+			"labels": map[string]any{
+				"type":  "array",
+				"items": map[string]any{"type": "string"},
+			},
+			"booleans": map[string]any{
+				"type":  "array",
+				"items": map[string]any{"type": "bool"},
+			},
+			"ambiguous": map[string]any{"type": []string{"string", "int"}},
+		},
+	}
+	params := map[string]any{
+		"enabled":   " true ",
+		"count":     " 7 ",
+		"ratio":     " 2.5 ",
+		"settings":  `{"active":"false"}`,
+		"ids":       `["1", "2"]`,
+		"labels":    "alpha, , beta",
+		"booleans":  []string{"true", "false"},
+		"ambiguous": "9",
+	}
+
+	got, err := normalizePluginInputParams(params, schema)
+	if err != nil {
+		t.Fatalf("normalizePluginInputParams() error = %v", err)
+	}
+	want := map[string]any{
+		"enabled":   true,
+		"count":     7,
+		"ratio":     2.5,
+		"settings":  map[string]any{"active": false},
+		"ids":       []any{1, 2},
+		"labels":    []any{"alpha", "beta"},
+		"booleans":  []any{true, false},
+		"ambiguous": "9",
+	}
+	if !reflect.DeepEqual(got, want) {
+		t.Fatalf("normalizePluginInputParams() = %#v, want %#v", got, want)
+	}
+
+	properties := schema["properties"].(map[string]any)
+	if gotType := properties["enabled"].(map[string]any)["type"].([]any)[1]; gotType != "bool" {
+		t.Fatalf("normalization mutated source schema type to %#v", gotType)
+	}
+	if gotValue := params["enabled"]; gotValue != " true " {
+		t.Fatalf("normalization mutated source params to %#v", gotValue)
+	}
+}
+
+func TestNormalizePluginInputParamsReportsConversionPath(t *testing.T) {
+	tests := []struct {
+		name        string
+		value       any
+		fieldSchema map[string]any
+		wantText    string
+	}{
+		{name: "boolean", value: "sometimes", fieldSchema: map[string]any{"type": "bool"}, wantText: "cannot convert"},
+		{name: "integer", value: "1.5", fieldSchema: map[string]any{"type": "int"}, wantText: "integer"},
+		{name: "number", value: "many", fieldSchema: map[string]any{"type": "float"}, wantText: "number"},
+		{name: "object", value: "{", fieldSchema: map[string]any{"type": "object"}, wantText: "object"},
+		{name: "null object", value: "null", fieldSchema: map[string]any{"type": "object"}, wantText: "expected a JSON object"},
+		{name: "array", value: "[", fieldSchema: map[string]any{"type": "array"}, wantText: "array"},
+		{
+			name:  "nested property",
+			value: `{"active":"sometimes"}`,
+			fieldSchema: map[string]any{
+				"type": "object",
+				"properties": map[string]any{
+					"active": map[string]any{"type": "bool"},
+				},
+			},
+			wantText: "field: active:",
+		},
+		{
+			name:  "array item",
+			value: "1,not-an-int",
+			fieldSchema: map[string]any{
+				"type":  "array",
+				"items": map[string]any{"type": "int"},
+			},
+			wantText: "item 1",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			schema := map[string]any{
+				"type":       "object",
+				"properties": map[string]any{"field": tt.fieldSchema},
+			}
+			_, err := normalizePluginInputParams(map[string]any{"field": tt.value}, schema)
+			if err == nil {
+				t.Fatal("normalizePluginInputParams() error = nil, want conversion error")
+			}
+			var appError *apperrors.Error
+			if !errors.As(err, &appError) ||
+				appError.Category != apperrors.CategoryValidation ||
+				appError.Reason != "plugin_input_schema_invalid" {
+				t.Fatalf("conversion error = %#v, want categorized plugin schema validation error", err)
+			}
+			if !strings.Contains(err.Error(), tt.wantText) {
+				t.Fatalf("conversion error = %q, want text %q", err, tt.wantText)
+			}
+		})
+	}
+}
+
+func TestNormalizePluginInputParamsRunsSchemaValidation(t *testing.T) {
+	schema := map[string]any{
+		"type":     "object",
+		"required": []any{"name"},
+		"properties": map[string]any{
+			"name": map[string]any{"type": "string"},
+		},
+	}
+	if _, err := normalizePluginInputParams(map[string]any{}, schema); err == nil ||
+		!strings.Contains(err.Error(), "$.name is required") {
+		t.Fatalf("required-field validation error = %v", err)
+	}
+}
+
+func TestPluginInputSchemaHelperEdges(t *testing.T) {
+	if got := canonicalPluginInputSchema(nil); got != nil {
+		t.Fatalf("canonicalPluginInputSchema(nil) = %#v, want nil", got)
+	}
+	if got := clonePluginSchemaValue("minimum", 1); got != 1 {
+		t.Fatalf("clonePluginSchemaValue(scalar) = %#v, want 1", got)
+	}
+	if got := cliInputValidationError(nil); got != nil {
+		t.Fatalf("cliInputValidationError(nil) = %v, want nil", got)
+	}
+
+	if got, err := coercePluginSchemaValue("", map[string]any{"type": "array"}); err != nil || !reflect.DeepEqual(got, []any(nil)) {
+		t.Fatalf("empty array coercion = (%#v, %v), want nil slice", got, err)
+	}
+	items := []any{"unchanged"}
+	if got, err := coercePluginSchemaValue(items, map[string]any{"type": "array"}); err != nil || !reflect.DeepEqual(got, items) {
+		t.Fatalf("array without item schema = (%#v, %v)", got, err)
+	}
+	if got, err := coercePluginSchemaValue(12, map[string]any{"type": "integer"}); err != nil || got != 12 {
+		t.Fatalf("non-string scalar coercion = (%#v, %v), want (12, nil)", got, err)
+	}
+	unknown := map[string]any{"unknown": "unchanged"}
+	if got, err := coercePluginSchemaValue(unknown, map[string]any{
+		"type":       "object",
+		"properties": map[string]any{},
+	}); err != nil || !reflect.DeepEqual(got, unknown) {
+		t.Fatalf("unknown property coercion = (%#v, %v), want unchanged map", got, err)
+	}
+
+	tests := []struct {
+		name   string
+		schema map[string]any
+		want   string
+	}{
+		{name: "missing", schema: map[string]any{}, want: ""},
+		{name: "single string", schema: map[string]any{"type": "integer"}, want: "integer"},
+		{name: "single string slice", schema: map[string]any{"type": []string{"null", "number"}}, want: "number"},
+		{name: "any slice", schema: map[string]any{"type": []any{nil, 3, "", "null", "boolean"}}, want: "boolean"},
+		{name: "ambiguous", schema: map[string]any{"type": []any{"string", "integer"}}, want: ""},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := singlePluginSchemaType(tt.schema); got != tt.want {
+				t.Fatalf("singlePluginSchemaType(%#v) = %q, want %q", tt.schema, got, tt.want)
+			}
+		})
+	}
+
+	for raw, want := range map[string]string{
+		" BOOL ": "boolean",
+		"Int":    "integer",
+		"FLOAT":  "number",
+		"custom": "custom",
+	} {
+		if got := canonicalPluginSchemaType(raw); got != want {
+			t.Errorf("canonicalPluginSchemaType(%q) = %q, want %q", raw, got, want)
+		}
+	}
+}

internal/app/plugin_input_schema_test.go

@@ -0,0 +1,109 @@
+// 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 (
+	"context"
+	"errors"
+	"reflect"
+	"testing"
+
+	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/transport"
+)
+
+func TestPluginStdioExecutionNormalizesAndValidatesLiveSchema(t *testing.T) {
+	isolatePluginRuntime(t)
+	previousInit := runnerStdioEnsureInitialized
+	previousList := runnerStdioListTools
+	previousCall := runnerStdioCallTool
+	t.Cleanup(func() {
+		runnerStdioEnsureInitialized = previousInit
+		runnerStdioListTools = previousList
+		runnerStdioCallTool = previousCall
+	})
+
+	client := transport.NewStdioClient("unused", nil, nil)
+	RegisterStdioClient("conference/local", client)
+	runnerStdioEnsureInitialized = func(*transport.StdioClient, context.Context) error {
+		return nil
+	}
+	runnerStdioListTools = func(*transport.StdioClient, context.Context) (transport.ToolsListResult, error) {
+		return transport.ToolsListResult{
+			Tools: []transport.ToolDescriptor{{
+				Name: "create_conference",
+				InputSchema: map[string]any{
+					"type":     "object",
+					"required": []any{"title"},
+					"properties": map[string]any{
+						"title":           map[string]any{"type": "string"},
+						"capture_speaker": map[string]any{"type": "bool"},
+					},
+					"additionalProperties": false,
+				},
+			}},
+		}, nil
+	}
+	var calledParams map[string]any
+	runnerStdioCallTool = func(
+		_ *transport.StdioClient,
+		_ context.Context,
+		_ string,
+		params map[string]any,
+	) (transport.ToolCallResult, error) {
+		calledParams = params
+		return transport.ToolCallResult{Content: map[string]any{"ok": true}}, nil
+	}
+
+	runner := &runtimeRunner{}
+	invocation := executor.Invocation{
+		CanonicalProduct: "conference-local",
+		Tool:             "create_conference",
+		Params: map[string]any{
+			"title":           "schema validation",
+			"capture_speaker": "true",
+		},
+	}
+	result, err := runner.executeInvocation(
+		context.Background(),
+		"stdio://conference/local",
+		invocation,
+	)
+	if err != nil {
+		t.Fatalf("stdio plugin execution: %v", err)
+	}
+	wantParams := map[string]any{
+		"title":           "schema validation",
+		"capture_speaker": true,
+	}
+	if !reflect.DeepEqual(calledParams, wantParams) ||
+		!reflect.DeepEqual(result.Invocation.Params, wantParams) {
+		t.Fatalf("normalized wire params = %#v, result = %#v", calledParams, result.Invocation.Params)
+	}
+
+	calledParams = nil
+	invocation.Params = map[string]any{"capture_speaker": "true"}
+	_, err = runner.executeInvocation(
+		context.Background(),
+		"stdio://conference/local",
+		invocation,
+	)
+	var appError *apperrors.Error
+	if !errors.As(err, &appError) ||
+		appError.Category != apperrors.CategoryValidation ||
+		calledParams != nil {
+		t.Fatalf("missing required schema validation = %#v, call params = %#v", err, calledParams)
+	}
+}

internal/app/plugin_registration_residual_test.go

@@ -0,0 +1,369 @@
+// 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 (
+	"context"
+	"encoding/json"
+	"errors"
+	"os"
+	"path/filepath"
+	"reflect"
+	"testing"
+
+	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/pipeline"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/plugin"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/shortcut/userdef"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/transport"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/cmdutil"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/mcptypes"
+	"github.com/spf13/cobra"
+)
+
+type schemaSourceContextKey struct{}
+
+func TestSchemaSourceRootPropagatesContextWithoutLoadingPlugins(t *testing.T) {
+	previous := rootLoadPlugins
+	t.Cleanup(func() { rootLoadPlugins = previous })
+
+	pluginLoads := 0
+	rootLoadPlugins = func(*cobra.Command, *pipeline.Engine, executor.Runner) []*cobra.Command {
+		pluginLoads++
+		return nil
+	}
+	wantContext := context.WithValue(context.Background(), schemaSourceContextKey{}, "schema")
+	root := NewSchemaSourceRootCommand(wantContext)
+	if root.Context() != wantContext {
+		t.Fatal("Schema source root did not retain the caller context")
+	}
+	if pluginLoads != 0 {
+		t.Fatalf("Schema source root loaded runtime plugins %d times", pluginLoads)
+	}
+}
+
+func TestCollectPluginServerCandidatesSortsAndSkipsInvalidStdio(t *testing.T) {
+	previousDescriptors := rootPluginDescriptors
+	previousClients := rootPluginStdioClients
+	previousDescriptor := rootPluginStdioDescriptor
+	t.Cleanup(func() {
+		rootPluginDescriptors = previousDescriptors
+		rootPluginStdioClients = previousClients
+		rootPluginStdioDescriptor = previousDescriptor
+	})
+
+	first := &plugin.Plugin{Manifest: plugin.Manifest{Name: "first"}}
+	second := &plugin.Plugin{Manifest: plugin.Manifest{Name: "second"}}
+	wantContext := &plugin.UserContext{UserID: "user", CorpID: "corp"}
+	client := transport.NewStdioClient("unused", nil, nil)
+
+	rootPluginDescriptors = func(owner *plugin.Plugin) []mcptypes.ServerDescriptor {
+		if owner == first {
+			return []mcptypes.ServerDescriptor{{Key: "same"}, {Key: " beta "}}
+		}
+		return []mcptypes.ServerDescriptor{{Key: "aardvark"}}
+	}
+	rootPluginStdioClients = func(owner *plugin.Plugin, gotContext *plugin.UserContext) []plugin.StdioServerClient {
+		if gotContext != wantContext {
+			t.Fatalf("stdio user context = %#v, want %#v", gotContext, wantContext)
+		}
+		if owner != first {
+			return nil
+		}
+		return []plugin.StdioServerClient{
+			{Key: "same", Client: client},
+			{Key: " alpha ", Client: client},
+			{Key: "invalid", Client: client},
+		}
+	}
+	rootPluginStdioDescriptor = func(_ *plugin.Plugin, stdio plugin.StdioServerClient) (mcptypes.ServerDescriptor, bool) {
+		if stdio.Key == "invalid" {
+			return mcptypes.ServerDescriptor{}, false
+		}
+		return mcptypes.ServerDescriptor{Key: stdio.Key}, true
+	}
+
+	candidates := collectPluginServerCandidates([]*plugin.Plugin{first, second}, wantContext)
+	if len(candidates) != 5 {
+		t.Fatalf("candidate count = %d, want 5", len(candidates))
+	}
+	gotKeys := make([]string, 0, len(candidates))
+	gotKinds := make([]string, 0, len(candidates))
+	for _, candidate := range candidates {
+		gotKeys = append(gotKeys, candidate.descriptor.Key)
+		if candidate.stdioClient == nil {
+			gotKinds = append(gotKinds, "http")
+		} else {
+			gotKinds = append(gotKinds, "stdio")
+			if candidate.stdioClient.Client != client {
+				t.Fatal("stdio candidate did not retain its client")
+			}
+		}
+	}
+	if want := []string{" alpha ", " beta ", "same", "same", "aardvark"}; !reflect.DeepEqual(gotKeys, want) {
+		t.Fatalf("candidate keys = %#v, want %#v", gotKeys, want)
+	}
+	if want := []string{"stdio", "http", "http", "stdio", "http"}; !reflect.DeepEqual(gotKinds, want) {
+		t.Fatalf("candidate transports = %#v, want %#v", gotKinds, want)
+	}
+}
+
+func TestPluginDescriptorBlankIdentityAndDistributionOwnership(t *testing.T) {
+	isolatePluginRuntime(t)
+
+	blank := mcptypes.ServerDescriptor{
+		Key: " ",
+		CLI: mcptypes.CLIOverlay{
+			ID:      " ",
+			Command: " ",
+			Aliases: []string{"", "  "},
+		},
+	}
+	if claims := pluginDescriptorIdentityClaims(blank); len(claims) != 0 {
+		t.Fatalf("blank descriptor claims = %#v, want none", claims)
+	}
+	if rootName := pluginDescriptorRootName(blank); rootName != "" {
+		t.Fatalf("blank descriptor root = %q", rootName)
+	}
+	owner := &plugin.Plugin{Manifest: plugin.Manifest{Name: "blank"}}
+	accepted := selectPluginServerCandidates(
+		&cobra.Command{Use: "dws"},
+		[]pluginServerCandidate{
+			{owner: owner, descriptor: mcptypes.ServerDescriptor{CLI: mcptypes.CLIOverlay{Skip: true}}},
+			{owner: owner, descriptor: blank},
+		},
+	)
+	if len(accepted) != 1 {
+		t.Fatalf("blank descriptor candidates = %#v, want one accepted candidate", accepted)
+	}
+
+	if distributionRootOwns(nil, "visible") {
+		t.Fatal("nil root claimed a command")
+	}
+	root := &cobra.Command{Use: "dws"}
+	visible := &cobra.Command{Use: "visible", Aliases: []string{" visible-alias "}}
+	hiddenFallback := &cobra.Command{Use: "conference", Hidden: true}
+	hiddenOwned := &cobra.Command{Use: "hidden-owned", Hidden: true}
+	pluginOwned := &cobra.Command{Use: "plugin-owned", Aliases: []string{"plugin-alias"}}
+	cmdutil.MarkPluginSource(pluginOwned)
+	root.AddCommand(visible, hiddenFallback, hiddenOwned, pluginOwned)
+
+	for _, name := range []string{"visible", "visible-alias", "hidden-owned"} {
+		if !distributionRootOwns(root, name) {
+			t.Errorf("distribution root did not claim %q", name)
+		}
+	}
+	for _, name := range []string{"conference", "plugin-owned", "plugin-alias", "missing"} {
+		if distributionRootOwns(root, name) {
+			t.Errorf("distribution root unexpectedly claimed %q", name)
+		}
+	}
+}
+
+func TestReplaceableFallbackIdentitySurvivesDistributionConflictChecks(t *testing.T) {
+	isolatePluginRuntime(t)
+	SetDynamicServers([]mcptypes.ServerDescriptor{
+		{
+			Key:      "conference",
+			Endpoint: "https://example.com/conference/mcp",
+			CLI:      mcptypes.CLIOverlay{ID: "conference"},
+		},
+		{
+			Key:      "chat",
+			Endpoint: "https://example.com/chat/mcp",
+			CLI:      mcptypes.CLIOverlay{ID: "chat"},
+		},
+	})
+
+	root := &cobra.Command{Use: "dws"}
+	root.AddCommand(&cobra.Command{Use: "conference", Hidden: true})
+	distributionProducts := DirectRuntimeProductIDs()
+
+	conferenceDescriptor := mcptypes.ServerDescriptor{
+		Key:         "conference-local",
+		DisplayName: "conference/conference-local",
+		CLI:         mcptypes.CLIOverlay{ID: "conference-local", Command: "conference"},
+	}
+	if pluginDescriptorConflictsWithDistribution(root, conferenceDescriptor, distributionProducts) {
+		t.Fatal("replaceable fallback identity blocked plugin server selection")
+	}
+
+	chatDescriptor := mcptypes.ServerDescriptor{
+		Key:         "chat-local",
+		DisplayName: "chat/chat-local",
+		CLI:         mcptypes.CLIOverlay{ID: "chat-local", Command: "chat"},
+	}
+	if !pluginDescriptorConflictsWithDistribution(root, chatDescriptor, distributionProducts) {
+		t.Fatal("non-replaceable distribution product no longer conflicts")
+	}
+
+	reservedDescriptor := mcptypes.ServerDescriptor{
+		Key:         "auth-local",
+		DisplayName: "auth/auth-local",
+		CLI:         mcptypes.CLIOverlay{ID: "auth-local", Command: "auth"},
+	}
+	if !pluginDescriptorConflictsWithDistribution(root, reservedDescriptor, distributionProducts) {
+		t.Fatal("reserved command name no longer conflicts")
+	}
+
+	first := &plugin.Plugin{Manifest: plugin.Manifest{Name: "conference"}}
+	second := &plugin.Plugin{Manifest: plugin.Manifest{Name: "other"}}
+	accepted := selectPluginServerCandidates(root, []pluginServerCandidate{
+		{owner: first, descriptor: conferenceDescriptor},
+		{
+			owner: second,
+			descriptor: mcptypes.ServerDescriptor{
+				Key:         "conference-other",
+				DisplayName: "other/conference-other",
+				CLI:         mcptypes.CLIOverlay{ID: "conference-other", Command: "conference"},
+			},
+		},
+	})
+	if len(accepted) != 1 {
+		t.Fatalf("accepted candidates = %d, want the first conference plugin only", len(accepted))
+	}
+	if accepted[0].owner != first {
+		t.Fatalf("accepted owner = %q, want the first conference plugin", accepted[0].owner.Manifest.Name)
+	}
+}
+
+func TestAddPluginCommandsSafeFiltersConflictingAliases(t *testing.T) {
+	root := &cobra.Command{Use: "dws"}
+	root.AddCommand(&cobra.Command{Use: "taken"})
+	command := &cobra.Command{
+		Use:     "extension",
+		Aliases: []string{"", "extension", "auth", "taken", "shared", " shared ", " okay "},
+	}
+	addPluginCommandsSafe(root, []*cobra.Command{
+		command,
+		{Use: "shared"},
+		{Use: "other", Aliases: []string{"extension"}},
+	})
+
+	if want := []string{"shared", "okay"}; !reflect.DeepEqual(command.Aliases, want) {
+		t.Fatalf("filtered aliases = %#v, want %#v", command.Aliases, want)
+	}
+	if child := findDirectChild(root, "shared"); child != nil {
+		t.Fatal("an accepted alias was also registered as a plugin primary command")
+	}
+	other := findDirectChild(root, "other")
+	if other == nil || len(other.Aliases) != 0 {
+		t.Fatalf("later plugin aliases = %#v", other)
+	}
+}
+
+func TestStdioRunnerReportsToolsListFailureAndMissingTool(t *testing.T) {
+	isolatePluginRuntime(t)
+	previousInit := runnerStdioEnsureInitialized
+	previousList := runnerStdioListTools
+	previousCall := runnerStdioCallTool
+	t.Cleanup(func() {
+		runnerStdioEnsureInitialized = previousInit
+		runnerStdioListTools = previousList
+		runnerStdioCallTool = previousCall
+	})
+
+	client := transport.NewStdioClient("unused", nil, nil)
+	RegisterStdioClient("plugin/server", client)
+	runnerStdioEnsureInitialized = func(*transport.StdioClient, context.Context) error { return nil }
+	toolCalls := 0
+	runnerStdioCallTool = func(*transport.StdioClient, context.Context, string, map[string]any) (transport.ToolCallResult, error) {
+		toolCalls++
+		return transport.ToolCallResult{}, nil
+	}
+	runner := &runtimeRunner{}
+	invocation := executor.Invocation{CanonicalProduct: "overlay-id", Tool: "wanted"}
+
+	listFailure := errors.New("list failed")
+	runnerStdioListTools = func(*transport.StdioClient, context.Context) (transport.ToolsListResult, error) {
+		return transport.ToolsListResult{}, listFailure
+	}
+	_, err := runner.executeStdioInvocationAtEndpoint(context.Background(), "stdio://plugin/server", invocation)
+	assertPluginRuntimeError(t, err, apperrors.CategoryAPI, "tools/list", "stdio_tools_list_error")
+
+	runnerStdioListTools = func(*transport.StdioClient, context.Context) (transport.ToolsListResult, error) {
+		return transport.ToolsListResult{Tools: []transport.ToolDescriptor{{Name: "other"}}}, nil
+	}
+	_, err = runner.executeStdioInvocationAtEndpoint(context.Background(), "stdio://plugin/server", invocation)
+	assertPluginRuntimeError(t, err, apperrors.CategoryValidation, "", "plugin_tool_not_found")
+	if toolCalls != 0 {
+		t.Fatalf("tools/call attempts after tools/list failures = %d", toolCalls)
+	}
+}
+
+func TestStdioManifestDescriptorAndRegistrationFailClosed(t *testing.T) {
+	isolatePluginRuntime(t)
+	p := &plugin.Plugin{
+		Manifest: plugin.Manifest{
+			Name: "broken-plugin",
+			MCPServers: map[string]*plugin.MCPServer{
+				"local": {CLI: json.RawMessage(`{`)},
+			},
+		},
+	}
+	server := plugin.StdioServerClient{
+		Key:    "local",
+		Client: transport.NewStdioClient("unused", nil, nil),
+	}
+	if descriptor, ok := stdioServerDescriptorFromManifest(p, server); ok || !reflect.ValueOf(descriptor).IsZero() {
+		t.Fatalf("invalid descriptor = (%#v, %v), want zero, false", descriptor, ok)
+	}
+	if descriptor := registerStdioServerFromManifest(p, server); !reflect.ValueOf(descriptor).IsZero() {
+		t.Fatalf("invalid registered descriptor = %#v, want zero", descriptor)
+	}
+	if _, ok := LookupStdioClient("broken-plugin/local"); ok {
+		t.Fatal("invalid stdio manifest registered a client")
+	}
+}
+
+func TestLegacyCommandsContinueWhenUserShortcutLoadFails(t *testing.T) {
+	configDir := t.TempDir()
+	t.Setenv("DWS_CONFIG_DIR", configDir)
+	shortcutDir := filepath.Join(configDir, "shortcuts")
+	if err := os.MkdirAll(shortcutDir, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(shortcutDir, "broken.yaml"), []byte("version: ["), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if _, loadErrors := userdef.Load(); len(loadErrors) == 0 {
+		t.Fatal("malformed shortcut fixture did not fail to load")
+	}
+
+	runner := executor.EchoRunner{}
+	caller := newToolCallerAdapter(runner, &GlobalFlags{})
+	if commands := newLegacyPublicCommands(runner, caller, true); len(commands) == 0 {
+		t.Fatal("legacy commands were dropped after a user shortcut load error")
+	}
+}
+
+func findDirectChild(root *cobra.Command, name string) *cobra.Command {
+	for _, command := range root.Commands() {
+		if command.Name() == name {
+			return command
+		}
+	}
+	return nil
+}
+
+func assertPluginRuntimeError(
+	t *testing.T,
+	err error,
+	wantCategory apperrors.Category,
+	wantOperation string,
+	wantReason string,
+) {
+	t.Helper()
+	var appError *apperrors.Error
+	if !errors.As(err, &appError) {
+		t.Fatalf("runtime error = %#v, want structured app error", err)
+	}
+	if appError.Category != wantCategory ||
+		appError.Operation != wantOperation ||
+		appError.Reason != wantReason {
+		t.Fatalf("runtime error = %#v, want category=%q operation=%q reason=%q", appError, wantCategory, wantOperation, wantReason)
+	}
+}

internal/app/plugin_registration_test.go

@@ -5,6 +5,7 @@
 package app
 
 import (
+	"encoding/json"
 	"fmt"
 	"net/http"
 	"net/http/httptest"
@@ -12,6 +13,7 @@ import (
 	"sync/atomic"
 	"testing"
 
+	"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/mcptypes"
@@ -35,6 +37,11 @@ func isolatePluginRuntime(t *testing.T) {
 	stdioClients = make(map[string]*transport.StdioClient)
 	stdioMu.Unlock()
 
+	pluginAuthMu.Lock()
+	previousPluginAuth := pluginAuthRegistry
+	pluginAuthRegistry = make(map[string]*PluginAuth)
+	pluginAuthMu.Unlock()
+
 	t.Cleanup(func() {
 		StopAllStdioClients()
 		dynamicMu.Lock()
@@ -46,6 +53,9 @@ func isolatePluginRuntime(t *testing.T) {
 		stdioMu.Lock()
 		stdioClients = previousStdio
 		stdioMu.Unlock()
+		pluginAuthMu.Lock()
+		pluginAuthRegistry = previousPluginAuth
+		pluginAuthMu.Unlock()
 	})
 }
 
@@ -77,14 +87,40 @@ func TestRegisterPluginHTTPServerDoesNotProbeEndpoint(t *testing.T) {
 func TestRegisterStdioServerFromManifestDoesNotStartProcess(t *testing.T) {
 	isolatePluginRuntime(t)
 	marker := t.TempDir() + "/started"
+	pluginRoot := t.TempDir()
+	if err := os.WriteFile(pluginRoot+"/overlay.json", []byte(`{
+		"id":"local",
+		"command":"lazy-stdio",
+		"groups":{"health":{"description":"health checks"}},
+		"toolOverrides":{"ping":{"cliName":"ping","group":"health"}}
+	}`), 0o600); err != nil {
+		t.Fatal(err)
+	}
 	client := transport.NewStdioClient("/bin/sh", []string{
 		"-c", fmt.Sprintf("printf started > %q", marker),
 	}, nil)
 	p := &plugin.Plugin{
-		Manifest: plugin.Manifest{Name: "lazy-stdio", Description: "lazy stdio test"},
-		Root:     t.TempDir(),
+		Manifest: plugin.Manifest{
+			Name:        "lazy-stdio",
+			Description: "lazy stdio test",
+			MCPServers: map[string]*plugin.MCPServer{
+				"local": {
+					Type:    "stdio",
+					Command: "unused",
+					CLI:     json.RawMessage(`"overlay.json"`),
+				},
+			},
+		},
+		Root: pluginRoot,
 	}
 	descriptor := registerStdioServerFromManifest(p, plugin.StdioServerClient{Key: "local", Client: client})
+	commands := buildPluginCommands([]mcptypes.ServerDescriptor{descriptor}, executor.EchoRunner{}, nil)
+	root := pluginTestRoot(commands...)
+	root.SetArgs([]string{"lazy-stdio", "--help"})
+	if err := root.Execute(); err != nil {
+		t.Fatalf("lazy stdio help: %v", err)
+	}
+	requirePluginChild(t, commands[0], "health", "ping")
 
 	if _, err := os.Stat(marker); !os.IsNotExist(err) {
 		t.Fatalf("stdio process started during registration: stat error = %v", err)

internal/app/plugin_stdio_overlay.go

@@ -14,10 +14,7 @@
 package app
 
 import (
-	"encoding/json"
 	"log/slog"
-	"os"
-	"path/filepath"
 
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/plugin"
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/mcptypes"
@@ -33,50 +30,26 @@ import (
 // When no CLI metadata is present, a minimal overlay keyed by the server
 // name is returned so callers can still build an identity descriptor.
 func resolveStdioOverlay(p *plugin.Plugin, sc plugin.StdioServerClient) mcptypes.CLIOverlay {
-	serverID := sc.Key
-	overlay := mcptypes.CLIOverlay{
-		ID:      serverID,
-		Command: serverID,
-	}
-	srv, ok := p.Manifest.MCPServers[sc.Key]
-	if !ok || len(srv.CLI) == 0 {
-		return overlay
-	}
-
-	cliData := srv.CLI
-	// A JSON string is interpreted as a relative path to an external
-	// overlay file (e.g. "overlay.json") anchored at the plugin root.
-	if len(cliData) > 0 && cliData[0] == '"' {
-		var cliPath string
-		if err := json.Unmarshal(cliData, &cliPath); err == nil && cliPath != "" {
-			absPath := filepath.Join(p.Root, cliPath)
-			if fileData, readErr := os.ReadFile(absPath); readErr == nil {
-				cliData = fileData
-			} else {
-				slog.Warn("plugin: failed to read CLI overlay file",
-					"plugin", p.Manifest.Name, "path", absPath, "error", readErr)
-			}
+	overlay, ok := p.ResolveCLIOverlay(sc.Key)
+	if !ok {
+		return mcptypes.CLIOverlay{
+			ID:      sc.Key,
+			Command: sc.Key,
+			Skip:    true,
 		}
 	}
-	if err := json.Unmarshal(cliData, &overlay); err != nil {
-		slog.Warn("plugin: failed to parse CLI overlay for stdio server",
-			"plugin", p.Manifest.Name, "server", sc.Key, "error", err)
-	}
-	if overlay.ID == "" {
-		overlay.ID = serverID
-	}
-	if overlay.Command == "" {
-		overlay.Command = serverID
-	}
 	return overlay
 }
 
-// registerStdioServerFromManifest registers an endpoint descriptor and an
-// unstarted client from versioned plugin metadata. Tool discovery is not part
-// of command-tree construction; execution starts and initializes the client.
-func registerStdioServerFromManifest(p *plugin.Plugin, sc plugin.StdioServerClient) mcptypes.ServerDescriptor {
-	overlay := resolveStdioOverlay(p, sc)
-	descriptor := mcptypes.ServerDescriptor{
+func stdioServerDescriptorFromManifest(
+	p *plugin.Plugin,
+	sc plugin.StdioServerClient,
+) (mcptypes.ServerDescriptor, bool) {
+	overlay, ok := p.ResolveCLIOverlay(sc.Key)
+	if !ok {
+		return mcptypes.ServerDescriptor{}, false
+	}
+	return mcptypes.ServerDescriptor{
 		Key:         sc.Key,
 		DisplayName: p.Manifest.Name + "/" + sc.Key,
 		Description: p.Manifest.Description,
@@ -84,13 +57,30 @@ func registerStdioServerFromManifest(p *plugin.Plugin, sc plugin.StdioServerClie
 		Source:      "plugin",
 		CLI:         overlay,
 		HasCLIMeta:  true,
-	}
+	}, true
+}
 
+func registerResolvedStdioServer(
+	p *plugin.Plugin,
+	sc plugin.StdioServerClient,
+	descriptor mcptypes.ServerDescriptor,
+) {
 	AppendDynamicServer(descriptor)
 	RegisterStdioClient(p.Manifest.Name+"/"+sc.Key, sc.Client)
 
 	slog.Debug("plugin: stdio server registered from manifest",
 		"plugin", p.Manifest.Name, "server", sc.Key,
-		"toolOverrides", len(overlay.ToolOverrides))
+		"toolOverrides", len(descriptor.CLI.ToolOverrides))
+}
+
+// registerStdioServerFromManifest registers an endpoint descriptor and an
+// unstarted client from versioned plugin metadata. Tool discovery is not part
+// of command-tree construction; execution starts and initializes the client.
+func registerStdioServerFromManifest(p *plugin.Plugin, sc plugin.StdioServerClient) mcptypes.ServerDescriptor {
+	descriptor, ok := stdioServerDescriptorFromManifest(p, sc)
+	if !ok {
+		return mcptypes.ServerDescriptor{}
+	}
+	registerResolvedStdioServer(p, sc, descriptor)
 	return descriptor
 }

internal/app/recovery_command.go

@@ -333,7 +333,11 @@ func (r *recoveryRuntime) CallToolDirect(ctx context.Context, serverID, toolName
 	if err != nil {
 		return nil, err
 	}
-	tc := r.transport.WithAuth(resolveRuntimeAuthToken(ctx, recoveryRuntimeToken(r.flags)), resolveIdentityHeaders())
+	authToken, err := resolveRuntimeAuthToken(ctx, recoveryRuntimeToken(r.flags))
+	if err != nil {
+		return nil, tokenResolutionError(err)
+	}
+	tc := r.transport.WithAuth(authToken, resolveIdentityHeaders())
 	result, err := tc.CallTool(ctx, endpoint, toolName, args)
 	if err != nil {
 		return nil, err

internal/app/root.go

@@ -23,6 +23,7 @@ import (
 	"os"
 	"os/signal"
 	"path/filepath"
+	"sort"
 	"strings"
 	"sync"
 	"syscall"
@@ -69,7 +70,8 @@ var (
 	rootPluginDescriptors           = (*plugin.Plugin).ToServerDescriptors
 	rootPluginStdioClients          = (*plugin.Plugin).StdioClients
 	rootRegisterPluginHTTPServer    = registerPluginHTTPServer
-	rootRegisterStdioManifest       = registerStdioServerFromManifest
+	rootPluginStdioDescriptor       = stdioServerDescriptorFromManifest
+	rootRegisterResolvedStdioServer = registerResolvedStdioServer
 	rootPluginLoadHooks             = (*plugin.Plugin).LoadHooks
 	rootPluginSyncSkills            = plugin.SyncSkills
 	rootAuthLoadTokenData           = authpkg.LoadTokenData
@@ -306,13 +308,28 @@ func NewRootCommand(ctx ...context.Context) *cobra.Command {
 	if len(ctx) > 0 && ctx[0] != nil {
 		rootCtx = ctx[0]
 	}
-	return NewRootCommandWithEngine(rootCtx, nil)
+	return newRootCommandWithEngine(rootCtx, nil, true)
+}
+
+// NewSchemaSourceRootCommand constructs the distribution-owned command tree
+// used by Schema generation and command-surface policy. Installed plugins and
+// user-defined shortcuts must not change the reviewed embedded Schema.
+func NewSchemaSourceRootCommand(ctx ...context.Context) *cobra.Command {
+	var rootCtx context.Context
+	if len(ctx) > 0 && ctx[0] != nil {
+		rootCtx = ctx[0]
+	}
+	return newRootCommandWithEngine(rootCtx, nil, false)
 }
 
 // NewRootCommandWithEngine constructs the root CLI command with an
 // optional pipeline engine for input correction. When engine is nil,
 // no pipeline processing is applied.
 func NewRootCommandWithEngine(rootCtx context.Context, engine *pipeline.Engine) *cobra.Command {
+	return newRootCommandWithEngine(rootCtx, engine, true)
+}
+
+func newRootCommandWithEngine(rootCtx context.Context, engine *pipeline.Engine, loadRuntimeExtensions bool) *cobra.Command {
 	if rootCtx == nil {
 		rootCtx = context.Background()
 	}
@@ -396,16 +413,9 @@ func NewRootCommandWithEngine(rootCtx context.Context, engine *pipeline.Engine)
 	}
 	root.AddCommand(utilityCommands...)
 
-	root.AddCommand(newLegacyPublicCommands(runner, patCaller)...)
+	root.AddCommand(newLegacyPublicCommands(runner, patCaller, loadRuntimeExtensions)...)
 	root.AddCommand(newLegacyHiddenCommands(runner)...)
 
-	// --- Plugin loading: runs AFTER legacy commands so plugin endpoints can
-	// be appended on top of the static endpoint registry.
-	pluginCmds := rootLoadPlugins(engine, runner)
-	if len(pluginCmds) > 0 {
-		addPluginCommandsSafe(root, pluginCmds)
-	}
-
 	// PAT authorization commands (open-source core)
 	pat.RegisterCommands(root, patCaller)
 
@@ -414,6 +424,15 @@ func NewRootCommandWithEngine(rootCtx context.Context, engine *pipeline.Engine)
 		fn(root, caller)
 		deduplicateCommands(root)
 	}
+	if loadRuntimeExtensions {
+		// Resolve plugins only after the complete distribution command tree is
+		// present, so endpoint and Cobra conflict checks see PAT and edition
+		// commands as well as the open-source base.
+		pluginCmds := rootLoadPlugins(root, engine, runner)
+		if len(pluginCmds) > 0 {
+			addPluginCommandsSafe(root, pluginCmds)
+		}
+	}
 	hideNonDirectRuntimeCommands(root)
 	configureRootHelp(root)
 	// Set custom flag error handler for better UX
@@ -631,12 +650,17 @@ var reservedCommands = map[string]bool{
 	"schema": true, "mcp": true, "help": true,
 }
 
+var replaceablePluginFallbacks = map[string]bool{
+	"conference": true,
+}
+
 // addPluginCommandsSafe registers plugin commands with conflict detection.
 //
 // Rules:
 //   - Plugin vs reserved (auth/plugin/cache/...) → reject, warn
 //   - Plugin vs plugin (same name)               → reject later one, warn
-//   - Plugin vs Market dynamic command            → allow, plugin wins
+//   - Plugin vs hidden compatibility fallback     → allow, plugin wins
+//   - Plugin vs visible distribution command      → reject, warn
 func addPluginCommandsSafe(root *cobra.Command, pluginCmds []*cobra.Command) {
 	// Build index of existing commands before plugin registration.
 	existing := make(map[string]bool)
@@ -664,17 +688,47 @@ func addPluginCommandsSafe(root *cobra.Command, pluginCmds []*cobra.Command) {
 		}
 		pluginSeen[name] = true
 
-		// Rule 3: plugin vs Market — plugin wins, remove the old one.
+		// An alias must not bypass the same protections applied to primary
+		// plugin command names or shadow another root command.
+		filteredAliases := make([]string, 0, len(cmd.Aliases))
+		for _, rawAlias := range cmd.Aliases {
+			alias := strings.TrimSpace(rawAlias)
+			if alias == "" || alias == name || reservedCommands[alias] ||
+				existing[alias] || pluginSeen[alias] {
+				if alias != "" {
+					slog.Warn("plugin: command alias conflicts with an existing command, skipping",
+						"command", name, "alias", alias)
+				}
+				continue
+			}
+			pluginSeen[alias] = true
+			filteredAliases = append(filteredAliases, alias)
+		}
+		cmd.Aliases = filteredAliases
+
+		// Rule 3: an installed plugin may replace a hidden compatibility
+		// fallback (for example conference), but never a visible distribution
+		// command that participates in the reviewed base interface.
 		if existing[name] {
 			for _, old := range root.Commands() {
 				if old.Name() == name {
+					if !old.Hidden || !replaceablePluginFallbacks[name] ||
+						cmdutil.IsPluginSourced(old) {
+						slog.Warn("plugin: command conflicts with a visible distribution command, skipping",
+							"command", name)
+						cmd = nil
+						break
+					}
 					root.RemoveCommand(old)
-					slog.Debug("plugin: overriding Market command",
+					slog.Debug("plugin: overriding hidden compatibility command",
 						"command", name)
 					break
 				}
 			}
 		}
+		if cmd == nil {
+			continue
+		}
 
 		root.AddCommand(cmd)
 	}
@@ -811,7 +865,21 @@ func CloseFileLogger() {
 // loadPlugins registers versioned plugin manifests, stdio clients, hooks, and
 // skills. It deliberately does not initialize MCP transports or call
 // tools/list while constructing the command tree.
-func loadPlugins(engine *pipeline.Engine, _ executor.Runner) []*cobra.Command {
+type pluginServerCandidate struct {
+	owner       *plugin.Plugin
+	order       int
+	descriptor  mcptypes.ServerDescriptor
+	stdioClient *plugin.StdioServerClient
+}
+
+type pluginIdentityOwner struct {
+	plugin    *plugin.Plugin
+	serverKey string
+	rootName  string
+	shareable bool
+}
+
+func loadPlugins(root *cobra.Command, engine *pipeline.Engine, runner executor.Runner) []*cobra.Command {
 	pluginLoader := plugin.NewLoader(RawVersion())
 
 	// 0a. Inject plugin config values from settings.json as environment
@@ -838,25 +906,34 @@ func loadPlugins(engine *pipeline.Engine, _ executor.Runner) []*cobra.Command {
 
 	// 2. Load dev plugins (registered via `dws plugin dev`)
 	devPlugins := rootPluginLoadDev(pluginLoader)
+	sortPluginsForRegistration(userPlugins)
+	sortPluginsForRegistration(devPlugins)
 
 	allPlugins := append(userPlugins, devPlugins...)
+	descriptorsByPlugin := make(map[*plugin.Plugin][]mcptypes.ServerDescriptor, len(allPlugins))
 
-	// 3. Register HTTP descriptors and authentication from the manifest.
-	for _, p := range allPlugins {
-		for _, srv := range rootPluginDescriptors(p) {
-			rootRegisterPluginHTTPServer(srv)
+	// 3. Resolve every descriptor once, then choose identity winners before
+	// mutating endpoint, auth, or stdio-client registries. This keeps the
+	// visible command and its transport owned by the same plugin.
+	candidates := collectPluginServerCandidates(allPlugins, userCtx)
+	accepted := selectPluginServerCandidates(root, candidates)
+	for _, candidate := range accepted {
+		if candidate.stdioClient != nil {
+			rootRegisterResolvedStdioServer(
+				candidate.owner,
+				*candidate.stdioClient,
+				candidate.descriptor,
+			)
+		} else {
+			rootRegisterPluginHTTPServer(candidate.descriptor)
 		}
+		descriptorsByPlugin[candidate.owner] = append(
+			descriptorsByPlugin[candidate.owner],
+			candidate.descriptor,
+		)
 	}
 
-	// 4. Register stdio descriptors and unstarted clients. The subprocess is
-	// started and initialized only when a command is actually executed.
-	for _, p := range allPlugins {
-		for _, sc := range rootPluginStdioClients(p, userCtx) {
-			rootRegisterStdioManifest(p, sc)
-		}
-	}
-
-	// 5. Register plugin hooks into pipeline engine
+	// 4. Register plugin hooks into pipeline engine
 	if engine != nil {
 		for _, p := range allPlugins {
 			hooksCfg, err := rootPluginLoadHooks(p)
@@ -874,7 +951,7 @@ func loadPlugins(engine *pipeline.Engine, _ executor.Runner) []*cobra.Command {
 		}
 	}
 
-	// 7. Sync plugin skills to agent directories
+	// 5. Sync plugin skills to agent directories
 	rootPluginSyncSkills(allPlugins)
 
 	if len(allPlugins) > 0 {
@@ -884,11 +961,228 @@ func loadPlugins(engine *pipeline.Engine, _ executor.Runner) []*cobra.Command {
 		)
 	}
 
-	return nil
+	var pluginCommands []*cobra.Command
+	for _, p := range allPlugins {
+		// Build each plugin independently. addPluginCommandsSafe deliberately
+		// resolves cross-plugin root conflicts with first-plugin-wins semantics.
+		pluginCommands = append(pluginCommands, buildPluginCommands(descriptorsByPlugin[p], runner, root)...)
+	}
+	return pluginCommands
+}
+
+func sortPluginsForRegistration(plugins []*plugin.Plugin) {
+	sort.SliceStable(plugins, func(i, j int) bool {
+		left := strings.TrimSpace(plugins[i].Manifest.Name) + "\x00" + strings.TrimSpace(plugins[i].Root)
+		right := strings.TrimSpace(plugins[j].Manifest.Name) + "\x00" + strings.TrimSpace(plugins[j].Root)
+		return left < right
+	})
+}
+
+func collectPluginServerCandidates(
+	plugins []*plugin.Plugin,
+	userCtx *plugin.UserContext,
+) []pluginServerCandidate {
+	var candidates []pluginServerCandidate
+	for order, owner := range plugins {
+		for _, descriptor := range rootPluginDescriptors(owner) {
+			candidates = append(candidates, pluginServerCandidate{
+				owner:      owner,
+				order:      order,
+				descriptor: descriptor,
+			})
+		}
+		for _, stdioClient := range rootPluginStdioClients(owner, userCtx) {
+			descriptor, ok := rootPluginStdioDescriptor(owner, stdioClient)
+			if !ok {
+				continue
+			}
+			clientCopy := stdioClient
+			candidates = append(candidates, pluginServerCandidate{
+				owner:       owner,
+				order:       order,
+				descriptor:  descriptor,
+				stdioClient: &clientCopy,
+			})
+		}
+	}
+	sort.SliceStable(candidates, func(i, j int) bool {
+		if candidates[i].order != candidates[j].order {
+			return candidates[i].order < candidates[j].order
+		}
+		left := strings.TrimSpace(candidates[i].descriptor.Key)
+		right := strings.TrimSpace(candidates[j].descriptor.Key)
+		if left != right {
+			return left < right
+		}
+		return candidates[i].stdioClient == nil && candidates[j].stdioClient != nil
+	})
+	return candidates
+}
+
+func selectPluginServerCandidates(
+	root *cobra.Command,
+	candidates []pluginServerCandidate,
+) []pluginServerCandidate {
+	distributionProducts := DirectRuntimeProductIDs()
+	owners := make(map[string]pluginIdentityOwner)
+	for identity := range distributionProducts {
+		if replaceablePluginFallbacks[identity] {
+			continue
+		}
+		owners[identity] = pluginIdentityOwner{serverKey: "distribution"}
+	}
+
+	accepted := make([]pluginServerCandidate, 0, len(candidates))
+	for _, candidate := range candidates {
+		descriptor := candidate.descriptor
+		if descriptor.CLI.Skip {
+			continue
+		}
+		if reason := unsupportedPluginDescriptor(root, descriptor); reason != "" {
+			slog.Warn("plugin: descriptor CLI semantics are unsupported, skipping",
+				"plugin", candidate.owner.Manifest.Name,
+				"server", descriptor.Key,
+				"field", reason)
+			continue
+		}
+		if pluginDescriptorConflictsWithDistribution(root, descriptor, distributionProducts) {
+			continue
+		}
+		claims := pluginDescriptorIdentityClaims(descriptor)
+		conflict := ""
+		for identity, shareable := range claims {
+			existing, exists := owners[identity]
+			if !exists {
+				continue
+			}
+			rootName := pluginDescriptorRootName(descriptor)
+			if shareable && existing.shareable &&
+				existing.plugin == candidate.owner &&
+				existing.rootName == rootName {
+				continue
+			}
+			conflict = identity
+			break
+		}
+		if conflict != "" {
+			slog.Warn("plugin: descriptor identity already owned, skipping",
+				"plugin", candidate.owner.Manifest.Name,
+				"server", descriptor.Key,
+				"identity", conflict)
+			continue
+		}
+		rootName := pluginDescriptorRootName(descriptor)
+		for identity, shareable := range claims {
+			if existing, exists := owners[identity]; exists &&
+				shareable && existing.shareable &&
+				existing.plugin == candidate.owner &&
+				existing.rootName == rootName {
+				continue
+			}
+			owners[identity] = pluginIdentityOwner{
+				plugin:    candidate.owner,
+				serverKey: descriptor.Key,
+				rootName:  rootName,
+				shareable: shareable,
+			}
+		}
+		accepted = append(accepted, candidate)
+	}
+	return accepted
+}
+
+func pluginDescriptorIdentityClaims(descriptor mcptypes.ServerDescriptor) map[string]bool {
+	claims := make(map[string]bool)
+	canonicalID := firstNonEmptyPluginString(descriptor.CLI.ID, descriptor.Key)
+	if canonicalID != "" {
+		claims[canonicalID] = false
+	}
+	for _, identity := range append(
+		[]string{pluginDescriptorRootName(descriptor)},
+		descriptor.CLI.Aliases...,
+	) {
+		identity = strings.TrimSpace(identity)
+		if identity == "" {
+			continue
+		}
+		if _, exists := claims[identity]; !exists {
+			claims[identity] = true
+		}
+	}
+	return claims
+}
+
+func pluginDescriptorRootName(descriptor mcptypes.ServerDescriptor) string {
+	return firstNonEmptyPluginString(
+		descriptor.CLI.Command,
+		descriptor.CLI.ID,
+		descriptor.Key,
+	)
+}
+
+func pluginDescriptorConflictsWithDistribution(
+	root *cobra.Command,
+	descriptor mcptypes.ServerDescriptor,
+	distributionProducts map[string]bool,
+) bool {
+	candidates := append(
+		[]string{
+			firstNonEmptyPluginString(descriptor.CLI.ID, descriptor.Key),
+			pluginDescriptorRootName(descriptor),
+		},
+		descriptor.CLI.Aliases...,
+	)
+	for _, candidate := range candidates {
+		candidate = strings.TrimSpace(candidate)
+		if candidate == "" {
+			continue
+		}
+		if !reservedCommands[candidate] && replaceablePluginFallbacks[candidate] {
+			// The distribution ships only a hidden compatibility fallback for
+			// this name; plugins may claim it and the later command merge in
+			// addPluginCommandsSafe still rejects visible non-fallback owners.
+			continue
+		}
+		if reservedCommands[candidate] ||
+			distributionProducts[candidate] ||
+			distributionRootOwns(root, candidate) {
+			slog.Warn("plugin: descriptor conflicts with a distribution command, skipping",
+				"plugin", descriptor.DisplayName,
+				"server", descriptor.Key,
+				"identity", candidate)
+			return true
+		}
+	}
+	return false
+}
+
+func distributionRootOwns(root *cobra.Command, name string) bool {
+	if root == nil {
+		return false
+	}
+	for _, command := range root.Commands() {
+		if cmdutil.IsPluginSourced(command) {
+			continue
+		}
+		if command.Name() == name {
+			if command.Hidden && replaceablePluginFallbacks[name] {
+				return false
+			}
+			return true
+		}
+		for _, alias := range command.Aliases {
+			if strings.TrimSpace(alias) == name {
+				return true
+			}
+		}
+	}
+	return false
 }
 
 func registerPluginHTTPServer(srv mcptypes.ServerDescriptor) {
 	AppendDynamicServer(srv)
+	productID := firstNonEmptyPluginString(srv.CLI.ID, srv.Key)
+	ClearPluginAuth(productID)
 	if len(srv.AuthHeaders) > 0 {
 		registerPluginAuthFromHeaders(srv)
 	}
@@ -917,10 +1211,7 @@ func registerPluginAuthFromHeaders(srv mcptypes.ServerDescriptor) {
 		host := parsed.Hostname()
 		trustedDomains = []string{host, "*." + host}
 	}
-	productID := strings.TrimSpace(srv.CLI.ID)
-	if productID == "" {
-		productID = srv.Key
-	}
+	productID := firstNonEmptyPluginString(srv.CLI.ID, srv.Key)
 	RegisterPluginAuth(productID, &PluginAuth{
 		Token:          authToken,
 		ExtraHeaders:   extraHeaders,

internal/app/root_full_coverage_test.go

@@ -75,7 +75,7 @@ func TestCrossPlatformCoverageRootConstructionHooksAndVersionCoverage(t *testing
 		version, buildTime, gitCommit = oldVersion, oldBuild, oldCommit
 	})
 
-	rootLoadPlugins = func(*pipeline.Engine, executor.Runner) []*cobra.Command {
+	rootLoadPlugins = func(*cobra.Command, *pipeline.Engine, executor.Runner) []*cobra.Command {
 		return []*cobra.Command{{Use: "plugin-added", Run: func(*cobra.Command, []string) {}}}
 	}
 	preRunCalled := false
@@ -236,7 +236,8 @@ func TestCrossPlatformCoverageRootLoadPluginsRemainingCoverage(t *testing.T) {
 	oldDescriptors := rootPluginDescriptors
 	oldStdioClients := rootPluginStdioClients
 	oldHTTP := rootRegisterPluginHTTPServer
-	oldStdio := rootRegisterStdioManifest
+	oldStdioDescriptor := rootPluginStdioDescriptor
+	oldStdioRegister := rootRegisterResolvedStdioServer
 	oldHooks := rootPluginLoadHooks
 	oldSync := rootPluginSyncSkills
 	oldToken := rootAuthLoadTokenData
@@ -247,7 +248,8 @@ func TestCrossPlatformCoverageRootLoadPluginsRemainingCoverage(t *testing.T) {
 		rootPluginDescriptors = oldDescriptors
 		rootPluginStdioClients = oldStdioClients
 		rootRegisterPluginHTTPServer = oldHTTP
-		rootRegisterStdioManifest = oldStdio
+		rootPluginStdioDescriptor = oldStdioDescriptor
+		rootRegisterResolvedStdioServer = oldStdioRegister
 		rootPluginLoadHooks = oldHooks
 		rootPluginSyncSkills = oldSync
 		rootAuthLoadTokenData = oldToken
@@ -264,9 +266,17 @@ func TestCrossPlatformCoverageRootLoadPluginsRemainingCoverage(t *testing.T) {
 	}
 	rootPluginDescriptors = func(p *plugin.Plugin) []mcptypes.ServerDescriptor {
 		if p == p1 {
-			return []mcptypes.ServerDescriptor{{Key: "http", Endpoint: "https://example.test"}}
+			return []mcptypes.ServerDescriptor{{
+				Key: "http", Endpoint: "https://example.test",
+				CLI: mcptypes.CLIOverlay{
+					ID: "http", Command: "one-http",
+					ToolOverrides: map[string]mcptypes.CLIToolOverride{
+						"ping": {CLIName: "ping"},
+					},
+				},
+			}}
 		}
-		return []mcptypes.ServerDescriptor{{Key: "no-cli", Endpoint: "https://example.test"}}
+		return []mcptypes.ServerDescriptor{{Key: p.Manifest.Name + "-no-cli", Endpoint: "https://example.test"}}
 	}
 	client := transport.NewStdioClient("ignored", nil, nil)
 	rootPluginStdioClients = func(p *plugin.Plugin, uc *plugin.UserContext) []plugin.StdioServerClient {
@@ -278,9 +288,23 @@ func TestCrossPlatformCoverageRootLoadPluginsRemainingCoverage(t *testing.T) {
 	httpCount := 0
 	stdioCount := 0
 	rootRegisterPluginHTTPServer = func(mcptypes.ServerDescriptor) { httpCount++ }
-	rootRegisterStdioManifest = func(*plugin.Plugin, plugin.StdioServerClient) mcptypes.ServerDescriptor {
+	rootPluginStdioDescriptor = func(*plugin.Plugin, plugin.StdioServerClient) (mcptypes.ServerDescriptor, bool) {
+		return mcptypes.ServerDescriptor{
+			Key: "local",
+			CLI: mcptypes.CLIOverlay{
+				ID: "local", Command: "one-stdio",
+				ToolOverrides: map[string]mcptypes.CLIToolOverride{
+					"pong": {CLIName: "pong"},
+				},
+			},
+		}, true
+	}
+	rootRegisterResolvedStdioServer = func(
+		*plugin.Plugin,
+		plugin.StdioServerClient,
+		mcptypes.ServerDescriptor,
+	) {
 		stdioCount++
-		return mcptypes.ServerDescriptor{}
 	}
 	rootPluginLoadHooks = func(p *plugin.Plugin) (*plugin.HooksConfig, error) {
 		switch p {
@@ -294,7 +318,8 @@ func TestCrossPlatformCoverageRootLoadPluginsRemainingCoverage(t *testing.T) {
 	}
 	synced := false
 	rootPluginSyncSkills = func([]*plugin.Plugin) { synced = true }
-	if got := loadPlugins(pipeline.NewEngine(), runnerCoverageFallback{}); got != nil {
+	got := loadPlugins(nil, pipeline.NewEngine(), runnerCoverageFallback{})
+	if len(got) != 2 || got[0].Name() != "one-http" || got[1].Name() != "one-stdio" {
 		t.Fatalf("loaded plugin commands = %#v", got)
 	}
 	if httpCount != 3 || stdioCount != 1 || !synced {

internal/app/runner.go

@@ -168,6 +168,7 @@ var (
 	runnerPreflightDocDownload          = (*runtimeRunner).preflightDocDownload
 	runnerCallTool                      = (*transport.Client).CallTool
 	runnerStdioEnsureInitialized        = (*transport.StdioClient).EnsureInitialized
+	runnerStdioListTools                = (*transport.StdioClient).ListTools
 	runnerStdioCallTool                 = (*transport.StdioClient).CallTool
 	runnerHandlePatAuthCheck            func(context.Context, *runtimeRunner, executor.Invocation, *apperrors.PATError, string, io.Writer) (executor.Result, error)
 	runnerRetryWithPatAuthRetry         func(context.Context, executor.Runner, executor.Invocation, *PatScopeError, string, io.Writer) (executor.Result, error)
@@ -235,7 +236,9 @@ func (r *runtimeRunner) runSingle(ctx context.Context, invocation executor.Invoc
 	// ~70ms on macOS; starting it here lets the load overlap with endpoint
 	// resolution and catalog loading below.
 	if prefetchToken {
-		go runnerGetCachedRuntimeToken(ctx)
+		go func() {
+			_, _ = runnerGetCachedRuntimeToken(ctx)
+		}()
 	}
 
 	if shouldUseDirectRuntime(invocation) {
@@ -483,7 +486,7 @@ func (r *runtimeRunner) handleCatalogMiss(ctx context.Context, invocation execut
 func (r *runtimeRunner) executeInvocation(ctx context.Context, endpoint string, invocation executor.Invocation) (result executor.Result, retErr error) {
 	// Route stdio:// endpoints to the local StdioClient — no HTTP, no auth.
 	if IsStdioEndpoint(endpoint) {
-		return r.executeStdioInvocation(ctx, invocation)
+		return r.executeStdioInvocationAtEndpoint(ctx, endpoint, invocation)
 	}
 
 	// Constructing the Cobra tree is also used for help, schema, and command
@@ -534,8 +537,12 @@ func (r *runtimeRunner) executeInvocation(ctx context.Context, endpoint string,
 	authToken := ""
 	if hasPluginAuth {
 		authToken = pluginAuth.Token
-	} else {
-		authToken = r.resolveAuthToken(ctx)
+	} else if !invocation.DryRun && (r.globalFlags == nil || !r.globalFlags.Mock) {
+		var tokenErr error
+		authToken, tokenErr = r.resolveAuthToken(ctx)
+		if tokenErr != nil {
+			return executor.Result{}, tokenResolutionError(tokenErr)
+		}
 	}
 
 	var timeoutSec int
@@ -617,6 +624,12 @@ func (r *runtimeRunner) executeInvocation(ctx context.Context, endpoint string,
 			}
 			return runnerHandlePatAuthCheck(ctx, r, invocation, patCheck, defaultConfigDir(), os.Stderr)
 		}
+		if result, retryErr, handled := r.retryAuthRefreshRequired(ctx, endpoint, invocation, authToken, err, hasPluginAuth); handled {
+			if retryErr != nil {
+				runnerCaptureRuntimeFailure(invocation, err, retryErr)
+			}
+			return result, retryErr
+		}
 		runnerCaptureRuntimeFailure(invocation, err, err)
 		return executor.Result{}, err
 	}
@@ -625,9 +638,15 @@ func (r *runtimeRunner) executeInvocation(ctx context.Context, endpoint string,
 	callResult, err := runnerCallTool(tc, callCtx, endpoint, invocation.Tool, invocation.Params)
 	RecordTiming(ctx, "mcp_call", time.Since(callStart))
 	if err != nil {
-		if isAuthError(err) {
+		if isRefreshableTransportAuthError(err) {
 			if fn := edition.Get().OnAuthError; fn != nil {
 				if overrideErr := fn(defaultConfigDir(), err); overrideErr != nil {
+					if result, retryErr, handled := r.retryAuthRefreshRequired(ctx, endpoint, invocation, authToken, overrideErr, hasPluginAuth); handled {
+						if retryErr != nil {
+							runnerCaptureRuntimeFailure(invocation, err, retryErr)
+						}
+						return result, retryErr
+					}
 					runnerCaptureRuntimeFailure(invocation, err, overrideErr)
 					return executor.Result{}, overrideErr
 				}
@@ -652,6 +671,12 @@ func (r *runtimeRunner) executeInvocation(ctx context.Context, endpoint string,
 				}
 				return runnerHandlePatAuthCheck(ctx, r, invocation, patCheck, defaultConfigDir(), os.Stderr)
 			}
+			if result, retryErr, handled := r.retryAuthRefreshRequired(ctx, endpoint, invocation, authToken, editionErr, hasPluginAuth); handled {
+				if retryErr != nil {
+					runnerCaptureRuntimeFailure(invocation, editionErr, retryErr)
+				}
+				return result, retryErr
+			}
 			return executor.Result{}, editionErr
 		}
 	}
@@ -672,6 +697,12 @@ func (r *runtimeRunner) executeInvocation(ctx context.Context, endpoint string,
 		// patterns (PAT permission, gateway-auth) before generic handling.
 		if classify := edition.Get().ClassifyToolResult; classify != nil {
 			if hookErr := classify(callResult.Content); hookErr != nil {
+				if result, retryErr, handled := r.retryAuthRefreshRequired(ctx, endpoint, invocation, authToken, hookErr, hasPluginAuth); handled {
+					if retryErr != nil {
+						runnerCaptureRuntimeFailure(invocation, hookErr, retryErr)
+					}
+					return result, retryErr
+				}
 				runnerCaptureRuntimeFailure(invocation, hookErr, hookErr)
 				return executor.Result{}, hookErr
 			}
@@ -736,6 +767,14 @@ func (r *runtimeRunner) executeInvocation(ctx context.Context, endpoint string,
 // subprocess instead of the HTTP transport. This is used for plugin stdio
 // servers whose endpoints use the stdio:// scheme.
 func (r *runtimeRunner) executeStdioInvocation(ctx context.Context, invocation executor.Invocation) (executor.Result, error) {
+	return r.executeStdioInvocationAtEndpoint(ctx, "", invocation)
+}
+
+func (r *runtimeRunner) executeStdioInvocationAtEndpoint(
+	ctx context.Context,
+	endpoint string,
+	invocation executor.Invocation,
+) (executor.Result, error) {
 	if invocation.DryRun {
 		return executor.Result{
 			Invocation: invocation,
@@ -748,10 +787,14 @@ func (r *runtimeRunner) executeStdioInvocation(ctx context.Context, invocation e
 		}, nil
 	}
 
-	client, ok := LookupStdioClient(invocation.CanonicalProduct)
+	lookupKey := strings.Trim(strings.TrimPrefix(strings.TrimSpace(endpoint), stdioEndpointScheme), "/")
+	if lookupKey == "" {
+		lookupKey = invocation.CanonicalProduct
+	}
+	client, ok := LookupStdioClient(lookupKey)
 	if !ok {
 		return executor.Result{}, apperrors.NewInternal(
-			fmt.Sprintf("stdio client not found for %q", invocation.CanonicalProduct))
+			fmt.Sprintf("stdio client not found for %q", lookupKey))
 	}
 
 	callCtx := ctx
@@ -768,6 +811,27 @@ func (r *runtimeRunner) executeStdioInvocation(ctx context.Context, invocation e
 		)
 	}
 
+	tools, err := runnerStdioListTools(client, callCtx)
+	if err != nil {
+		return executor.Result{}, apperrors.NewAPI(
+			fmt.Sprintf("stdio tools/list failed: %v", err),
+			apperrors.WithOperation("tools/list"),
+			apperrors.WithReason("stdio_tools_list_error"),
+		)
+	}
+	schema, ok := pluginToolInputSchema(tools, invocation.Tool)
+	if !ok {
+		return executor.Result{}, apperrors.NewValidation(
+			fmt.Sprintf("plugin tool %q is not declared by tools/list", invocation.Tool),
+			apperrors.WithReason("plugin_tool_not_found"),
+		)
+	}
+	normalizedParams, err := normalizePluginInputParams(invocation.Params, schema)
+	if err != nil {
+		return executor.Result{}, err
+	}
+	invocation.Params = normalizedParams
+
 	callResult, err := runnerStdioCallTool(client, callCtx, invocation.Tool, invocation.Params)
 	if err != nil {
 		return executor.Result{}, apperrors.NewAPI(
@@ -796,67 +860,49 @@ func (r *runtimeRunner) executeStdioInvocation(ctx context.Context, invocation e
 	}, nil
 }
 
-func (r *runtimeRunner) resolveAuthToken(ctx context.Context) string {
+func (r *runtimeRunner) resolveAuthToken(ctx context.Context) (string, error) {
 	explicitToken := ""
 	if r != nil && r.globalFlags != nil {
 		explicitToken = r.globalFlags.Token
 	}
-	if token := strings.TrimSpace(explicitToken); token != "" {
-		return token
-	}
-	if tp := edition.Get().TokenProvider; tp != nil {
-		token, _ := tp(ctx, func() (string, error) {
-			return resolveAccessTokenFromDir(ctx, defaultConfigDir())
-		})
-		return token
-	}
-	return getCachedRuntimeToken(ctx)
+	return resolveRuntimeAuthToken(ctx, explicitToken)
 }
 
-func resolveRuntimeAuthToken(ctx context.Context, explicitToken string) string {
-	if token := strings.TrimSpace(explicitToken); token != "" {
-		return token
+func resolveRuntimeAuthToken(ctx context.Context, explicitToken string) (string, error) {
+	snapshot, err := runtimeTokenManager.Get(ctx, defaultConfigDir(), explicitToken)
+	if err != nil {
+		return "", err
 	}
-	// Use cached token to avoid repeated Keychain access (~70ms per call)
-	return getCachedRuntimeToken(ctx)
+	return snapshot.AccessToken, nil
 }
 
-// Cached token state for process lifetime
-var (
-	cachedRuntimeTokenMu sync.Mutex
-	cachedRuntimeTokens  = map[string]string{}
-)
-
-// getCachedRuntimeToken returns a cached access token, loading it only once per process.
-// This avoids repeated Keychain access which takes ~70ms each time.
-func getCachedRuntimeToken(ctx context.Context) string {
-	cacheKey := strings.TrimSpace(authpkg.RuntimeProfile())
-	if cacheKey == "" {
-		cacheKey = "__default__"
-	}
-	cachedRuntimeTokenMu.Lock()
-	if token := cachedRuntimeTokens[cacheKey]; token != "" {
-		cachedRuntimeTokenMu.Unlock()
-		return token
-	}
-	cachedRuntimeTokenMu.Unlock()
-
+// getCachedRuntimeToken is kept as the prefetch seam used by runner tests. The
+// cache itself lives exclusively in TokenManager.
+func getCachedRuntimeToken(ctx context.Context) (string, error) {
 	loadStart := time.Now()
 	defer func() { RecordTiming(ctx, "auth_keychain", time.Since(loadStart)) }()
+	return resolveRuntimeAuthToken(ctx, "")
+}
 
-	configDir := defaultConfigDir()
-	token, tokenErr := resolveAccessTokenFromDir(ctx, configDir)
-	if tokenErr != nil && errors.Is(tokenErr, authpkg.ErrTokenDecryption) {
-		slog.Error(tokenErr.Error())
-		return ""
+func tokenResolutionError(err error) error {
+	if err == nil {
+		return nil
 	}
-	if token == "" {
-		return ""
+	if errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
+		return err
 	}
-	cachedRuntimeTokenMu.Lock()
-	cachedRuntimeTokens[cacheKey] = token
-	cachedRuntimeTokenMu.Unlock()
-	return token
+	if errors.Is(err, authpkg.ErrTokenDataNotFound) {
+		return apperrors.NewAuth(
+			"未登录，请先执行 dws auth login",
+			apperrors.WithReason("not_authenticated"),
+			apperrors.WithHint("运行 'dws auth login' 完成登录后重试"),
+			apperrors.WithActions("dws auth login"),
+			apperrors.WithCause(err),
+		)
+	}
+	// Keychain, parse, permission, lock, and refresh failures are real local or
+	// network errors. Preserve their cause instead of disguising them as logout.
+	return fmt.Errorf("resolve access token: %w", err)
 }
 
 // generateExecutionID returns a random 16-char hex string used to correlate
@@ -871,9 +917,7 @@ func generateExecutionID() string {
 // ResetRuntimeTokenCache clears the cached token, forcing a reload on next access.
 // This should be called after login/logout operations.
 func ResetRuntimeTokenCache() {
-	cachedRuntimeTokenMu.Lock()
-	defer cachedRuntimeTokenMu.Unlock()
-	cachedRuntimeTokens = map[string]string{}
+	runtimeTokenManager.Invalidate()
 }
 
 func newRuntimeContentScanner() safety.Scanner {

internal/app/runner_full_coverage_test.go

@@ -50,9 +50,9 @@ func TestCrossPlatformCoverageRunnerRemainingRoutingCoverage(t *testing.T) {
 
 	inv := executor.Invocation{CanonicalProduct: "product", Tool: "tool"}
 	prefetched := make(chan struct{}, 1)
-	runnerGetCachedRuntimeToken = func(context.Context) string {
+	runnerGetCachedRuntimeToken = func(context.Context) (string, error) {
 		prefetched <- struct{}{}
-		return ""
+		return "", nil
 	}
 	r := &runtimeRunner{
 		loader:    cli.CatalogLoaderFrom(cli.Catalog{}, wantErr),
@@ -197,7 +197,7 @@ func TestCrossPlatformCoverageRunnerRemainingExecutionCoverage(t *testing.T) {
 		return nil
 	}
 
-	authErr := apperrors.NewAuth("expired")
+	authErr := apperrors.NewAuth("expired", apperrors.WithReason("http_401"))
 	runnerCallTool = func(*transport.Client, context.Context, string, string, map[string]any) (transport.ToolCallResult, error) {
 		return transport.ToolCallResult{}, authErr
 	}
@@ -290,10 +290,12 @@ func TestCrossPlatformCoverageRunnerRemainingExecutionCoverage(t *testing.T) {
 
 func TestCrossPlatformCoverageRunnerRemainingStdioAuthAndHeadersCoverage(t *testing.T) {
 	oldStdioInit := runnerStdioEnsureInitialized
+	oldStdioList := runnerStdioListTools
 	oldStdioCall := runnerStdioCallTool
 	oldEdition := edition.Get()
 	t.Cleanup(func() {
 		runnerStdioEnsureInitialized = oldStdioInit
+		runnerStdioListTools = oldStdioList
 		runnerStdioCallTool = oldStdioCall
 		edition.Override(oldEdition)
 		StopAllStdioClients()
@@ -309,6 +311,14 @@ func TestCrossPlatformCoverageRunnerRemainingStdioAuthAndHeadersCoverage(t *test
 		t.Fatalf("stdio initialize error = %v", err)
 	}
 	runnerStdioEnsureInitialized = func(*transport.StdioClient, context.Context) error { return nil }
+	runnerStdioListTools = func(*transport.StdioClient, context.Context) (transport.ToolsListResult, error) {
+		return transport.ToolsListResult{
+			Tools: []transport.ToolDescriptor{{
+				Name:        "tool",
+				InputSchema: map[string]any{"type": "object"},
+			}},
+		}, nil
+	}
 	runnerStdioCallTool = func(*transport.StdioClient, context.Context, string, map[string]any) (transport.ToolCallResult, error) {
 		return transport.ToolCallResult{}, wantErr
 	}
@@ -327,21 +337,31 @@ func TestCrossPlatformCoverageRunnerRemainingStdioAuthAndHeadersCoverage(t *test
 	if got, err := r.executeStdioInvocation(context.Background(), inv); err != nil || !got.Invocation.Implemented {
 		t.Fatalf("stdio success = %#v, %v", got, err)
 	}
+	RegisterStdioClient("plugin/server-key", client)
+	overlayIDInvocation := inv
+	overlayIDInvocation.CanonicalProduct = "overlay-id"
+	if got, err := r.executeInvocation(
+		context.Background(),
+		"stdio://plugin/server-key",
+		overlayIDInvocation,
+	); err != nil || !got.Invocation.Implemented {
+		t.Fatalf("stdio endpoint-key lookup = %#v, %v", got, err)
+	}
 
 	r.globalFlags.Token = " explicit "
-	if got := r.resolveAuthToken(context.Background()); got != "explicit" {
-		t.Fatalf("explicit auth token = %q", got)
+	if got, err := r.resolveAuthToken(context.Background()); err != nil || got != "explicit" {
+		t.Fatalf("explicit auth token = %q, %v", got, err)
 	}
 	edition.Override(&edition.Hooks{TokenProvider: func(_ context.Context, fallback func() (string, error)) (string, error) {
 		_, _ = fallback()
 		return "provided", nil
 	}})
 	r.globalFlags.Token = ""
-	if got := r.resolveAuthToken(context.Background()); got != "provided" {
-		t.Fatalf("provided auth token = %q", got)
+	if got, err := r.resolveAuthToken(context.Background()); err != nil || got != "provided" {
+		t.Fatalf("provided auth token = %q, %v", got, err)
 	}
-	if got := resolveRuntimeAuthToken(context.Background(), " runtime "); got != "runtime" {
-		t.Fatalf("runtime explicit token = %q", got)
+	if got, err := resolveRuntimeAuthToken(context.Background(), " runtime "); err != nil || got != "runtime" {
+		t.Fatalf("runtime explicit token = %q, %v", got, err)
 	}
 
 	t.Setenv(envDWSChannel, "channel")

internal/app/schema_completeness_test.go

@@ -14,7 +14,7 @@ func TestRuntimeSchemaCompletenessCoversPublicCommandTree(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	root := NewRootCommand()
+	root := NewSchemaSourceRootCommand()
 	if err := cli.ValidateEmbeddedRuntimeSchemaCompleteness(root); err != nil {
 		t.Fatal(err)
 	}

internal/app/skill_command.go

@@ -17,6 +17,7 @@ import (
 	"archive/zip"
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"io"
 	"mime"
@@ -36,25 +37,25 @@ import (
 )
 
 var (
-	skillLoadAccessToken   = loadSkillAccessToken
-	skillDownloadToTmp     = downloadSkillToTmpDir
-	skillHTTPDo            = func(client *http.Client, req *http.Request) (*http.Response, error) { return client.Do(req) }
-	skillNewRequest        = http.NewRequestWithContext
-	skillLoadTokenData     = authpkg.LoadTokenData
-	skillResolveTargetPath = resolveSkillTargetPath
-	skillFetchDownloadInfo = fetchSkillDownloadInfo
-	skillDownloadFile      = downloadSkillFile
-	skillExtractZip        = extractSkillZip
-	skillUserHomeDir       = os.UserHomeDir
-	skillMkdirTemp         = os.MkdirTemp
-	skillCreate            = os.Create
-	skillCreateTemp        = os.CreateTemp
-	skillRemoveAll         = os.RemoveAll
-	skillRemove            = os.Remove
-	skillMkdirAll          = os.MkdirAll
-	skillOpenFile          = os.OpenFile
-	skillCopy              = io.Copy
-	skillOpenZipFile       = func(file *zip.File) (io.ReadCloser, error) { return file.Open() }
+	skillLoadAccessToken    = loadSkillAccessToken
+	skillDownloadToTmp      = downloadSkillToTmpDir
+	skillHTTPDo             = func(client *http.Client, req *http.Request) (*http.Response, error) { return client.Do(req) }
+	skillNewRequest         = http.NewRequestWithContext
+	skillResolveAccessToken = ResolveAuxiliaryAccessToken
+	skillResolveTargetPath  = resolveSkillTargetPath
+	skillFetchDownloadInfo  = fetchSkillDownloadInfo
+	skillDownloadFile       = downloadSkillFile
+	skillExtractZip         = extractSkillZip
+	skillUserHomeDir        = os.UserHomeDir
+	skillMkdirTemp          = os.MkdirTemp
+	skillCreate             = os.Create
+	skillCreateTemp         = os.CreateTemp
+	skillRemoveAll          = os.RemoveAll
+	skillRemove             = os.Remove
+	skillMkdirAll           = os.MkdirAll
+	skillOpenFile           = os.OpenFile
+	skillCopy               = io.Copy
+	skillOpenZipFile        = func(file *zip.File) (io.ReadCloser, error) { return file.Open() }
 )
 
 func init() {
@@ -296,7 +297,7 @@ func newSkillAddHintCommand() *cobra.Command {
 
 func runSkillGet(cmd *cobra.Command, args []string) error {
 	skillID, _ := cmd.Flags().GetString("skill-id")
-	accessToken, err := skillLoadAccessToken()
+	accessToken, err := skillLoadAccessToken(cmd.Context())
 	if err != nil {
 		return err
 	}
@@ -319,7 +320,7 @@ func runSkillFind(cmd *cobra.Command, args []string) error {
 	if source == "" {
 		source, _ = cmd.Flags().GetString("scopes")
 	}
-	accessToken, err := skillLoadAccessToken()
+	accessToken, err := skillLoadAccessToken(cmd.Context())
 	if err != nil {
 		return err
 	}
@@ -388,7 +389,7 @@ func runSkillAdd(cmd *cobra.Command, args []string) error {
 		return apperrors.NewValidation(fmt.Sprintf("invalid target '%s': %v. Supported targets: %s", target, err, supportedTargets()))
 	}
 
-	accessToken, err := skillLoadAccessToken()
+	accessToken, err := skillLoadAccessToken(cmd.Context())
 	if err != nil {
 		return err
 	}
@@ -441,13 +442,16 @@ func runSkillAdd(cmd *cobra.Command, args []string) error {
 	return nil
 }
 
-func loadSkillAccessToken() (string, error) {
+func loadSkillAccessToken(ctx context.Context) (string, error) {
 	configDir := defaultConfigDir()
-	tokenData, err := skillLoadTokenData(configDir)
-	if err != nil || tokenData == nil || !tokenData.IsAccessTokenValid() {
+	token, err := skillResolveAccessToken(ctx, configDir, "")
+	if errors.Is(err, authpkg.ErrTokenDataNotFound) {
 		return "", skillAuthError()
 	}
-	return tokenData.AccessToken, nil
+	if err != nil {
+		return "", fmt.Errorf("resolve skill access token: %w", err)
+	}
+	return token, nil
 }
 
 func skillAuthError() error {

internal/app/skill_command_full_coverage_test.go

@@ -57,11 +57,11 @@ func TestCrossPlatformCoverageSkillCommandHighLevelRemainingCoverage(t *testing.
 	})
 	fail := errors.New("failure")
 	cmd := skillCoverageCommand()
-	skillLoadAccessToken = func() (string, error) { return "", fail }
+	skillLoadAccessToken = func(context.Context) (string, error) { return "", fail }
 	if err := runSkillGet(cmd, nil); !errors.Is(err, fail) {
 		t.Fatalf("skill get auth error = %v", err)
 	}
-	skillLoadAccessToken = func() (string, error) { return "token", nil }
+	skillLoadAccessToken = func(context.Context) (string, error) { return "token", nil }
 	skillNewRequest = func(context.Context, string, string, io.Reader) (*http.Request, error) { return nil, fail }
 	if err := runSkillFind(cmd, nil); err == nil {
 		t.Fatal("skill find request failure should propagate")
@@ -72,11 +72,11 @@ func TestCrossPlatformCoverageSkillCommandHighLevelRemainingCoverage(t *testing.
 		t.Fatalf("skill get download error = %v", err)
 	}
 
-	skillLoadAccessToken = func() (string, error) { return "", fail }
+	skillLoadAccessToken = func(context.Context) (string, error) { return "", fail }
 	if err := runSkillFind(cmd, nil); !errors.Is(err, fail) {
 		t.Fatalf("skill find auth error = %v", err)
 	}
-	skillLoadAccessToken = func() (string, error) { return "token", nil }
+	skillLoadAccessToken = func(context.Context) (string, error) { return "token", nil }
 	skillHTTPDo = func(*http.Client, *http.Request) (*http.Response, error) { return nil, fail }
 	if err := runSkillFind(cmd, nil); err == nil {
 		t.Fatal("skill find network failure should propagate")
@@ -111,11 +111,11 @@ func TestCrossPlatformCoverageSkillCommandHighLevelRemainingCoverage(t *testing.
 		t.Fatal("invalid skill target should fail")
 	}
 	skillResolveTargetPath = func(string) (string, error) { return "dest", nil }
-	skillLoadAccessToken = func() (string, error) { return "", fail }
+	skillLoadAccessToken = func(context.Context) (string, error) { return "", fail }
 	if err := runSkillAdd(cmd, []string{"id", "target"}); !errors.Is(err, fail) {
 		t.Fatalf("skill add auth error = %v", err)
 	}
-	skillLoadAccessToken = func() (string, error) { return "token", nil }
+	skillLoadAccessToken = func(context.Context) (string, error) { return "token", nil }
 	skillFetchDownloadInfo = func(context.Context, string, string) (*downloadSkillResponse, error) { return nil, fail }
 	if err := runSkillAdd(cmd, []string{"id", "target"}); !errors.Is(err, fail) {
 		t.Fatalf("skill info error = %v", err)
@@ -152,25 +152,35 @@ func TestCrossPlatformCoverageSkillCommandHighLevelRemainingCoverage(t *testing.
 
 func TestCrossPlatformCoverageSkillCommandLowLevelRemainingCoverage(t *testing.T) {
 	oldHTTP := skillHTTPDo
-	oldNewRequest, oldLoadToken := skillNewRequest, skillLoadTokenData
+	oldNewRequest, oldResolveToken := skillNewRequest, skillResolveAccessToken
 	oldHome := skillUserHomeDir
 	oldMkdirTemp, oldCreate, oldCreateTemp := skillMkdirTemp, skillCreate, skillCreateTemp
 	oldRemoveAll, oldRemove, oldMkdir := skillRemoveAll, skillRemove, skillMkdirAll
 	oldOpen, oldCopy, oldZipOpen := skillOpenFile, skillCopy, skillOpenZipFile
 	t.Cleanup(func() {
 		skillHTTPDo = oldHTTP
-		skillNewRequest, skillLoadTokenData = oldNewRequest, oldLoadToken
+		skillNewRequest, skillResolveAccessToken = oldNewRequest, oldResolveToken
 		skillUserHomeDir = oldHome
 		skillMkdirTemp, skillCreate, skillCreateTemp = oldMkdirTemp, oldCreate, oldCreateTemp
 		skillRemoveAll, skillRemove, skillMkdirAll = oldRemoveAll, oldRemove, oldMkdir
 		skillOpenFile, skillCopy, skillOpenZipFile = oldOpen, oldCopy, oldZipOpen
 	})
 	fail := errors.New("failure")
-	skillLoadTokenData = func(string) (*authpkg.TokenData, error) { return nil, nil }
-	if _, err := loadSkillAccessToken(); err == nil {
+	skillResolveAccessToken = func(context.Context, string, string) (string, error) {
+		return "", authpkg.ErrTokenDataNotFound
+	}
+	if _, err := loadSkillAccessToken(context.Background()); err == nil {
 		t.Fatal("invalid skill access token succeeded")
 	}
-	skillLoadTokenData = oldLoadToken
+	canceled, cancel := context.WithCancel(context.Background())
+	cancel()
+	skillResolveAccessToken = func(ctx context.Context, _, _ string) (string, error) {
+		return "", ctx.Err()
+	}
+	if _, err := loadSkillAccessToken(canceled); !errors.Is(err, context.Canceled) {
+		t.Fatalf("skill token cancellation = %v", err)
+	}
+	skillResolveAccessToken = oldResolveToken
 	skillNewRequest = func(context.Context, string, string, io.Reader) (*http.Request, error) { return nil, fail }
 	if _, err := fetchSkillDownloadInfo(context.Background(), "token", "id"); err == nil {
 		t.Fatal("download-info request failure should propagate")

internal/app/skill_command_test.go

@@ -18,7 +18,6 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
-	"errors"
 	"net/http"
 	"net/http/httptest"
 	"os"
@@ -397,9 +396,11 @@ func TestSkillInstallRequiresAuth(t *testing.T) {
 	configDir := filepath.Join(tempDir, "config")
 	t.Setenv("DWS_CONFIG_DIR", configDir)
 	t.Cleanup(CloseFileLogger)
-	originalLoadToken := skillLoadTokenData
-	skillLoadTokenData = func(string) (*authpkg.TokenData, error) { return nil, errors.New("missing") }
-	t.Cleanup(func() { skillLoadTokenData = originalLoadToken })
+	originalResolveToken := skillResolveAccessToken
+	skillResolveAccessToken = func(context.Context, string, string) (string, error) {
+		return "", authpkg.ErrTokenDataNotFound
+	}
+	t.Cleanup(func() { skillResolveAccessToken = originalResolveToken })
 
 	// Ensure the config directory exists but has no token
 	if err := os.MkdirAll(configDir, 0755); err != nil {
@@ -679,16 +680,11 @@ func TestSkillSearchUsesSourceQueryAndKeepsScopesCompat(t *testing.T) {
 	configDir := filepath.Join(t.TempDir(), "config")
 	t.Setenv("DWS_CONFIG_DIR", configDir)
 	t.Cleanup(CloseFileLogger)
-	originalLoadToken := skillLoadTokenData
-	skillLoadTokenData = func(string) (*authpkg.TokenData, error) {
-		return &authpkg.TokenData{
-			AccessToken:  "test-token",
-			RefreshToken: "refresh-token",
-			ExpiresAt:    time.Now().Add(time.Hour),
-			RefreshExpAt: time.Now().Add(24 * time.Hour),
-		}, nil
+	originalResolveToken := skillResolveAccessToken
+	skillResolveAccessToken = func(context.Context, string, string) (string, error) {
+		return "test-token", nil
 	}
-	t.Cleanup(func() { skillLoadTokenData = originalLoadToken })
+	t.Cleanup(func() { skillResolveAccessToken = originalResolveToken })
 
 	var gotSources []string
 	var gotScopes []string

internal/app/testmain_test.go

@@ -52,6 +52,10 @@ func TestMain(m *testing.M) {
 		_ = os.RemoveAll(tmpDir)
 		panic("set " + keychain.StorageDirEnv + ": " + err.Error())
 	}
+	if err := os.Setenv(keychain.TestNamespaceEnv, tmpDir); err != nil {
+		_ = os.RemoveAll(tmpDir)
+		panic("set " + keychain.TestNamespaceEnv + ": " + err.Error())
+	}
 	if err := os.Setenv("DWS_CONFIG_DIR", filepath.Join(tmpDir, "config")); err != nil {
 		_ = os.RemoveAll(tmpDir)
 		panic("set DWS_CONFIG_DIR: " + err.Error())
@@ -79,6 +83,12 @@ func TestMain(m *testing.M) {
 	StopAllStdioClients()
 	CloseAuditSink()
 	CloseFileLogger()
+	if err := keychain.RemoveAuthTokenEntries(keychain.Service); err != nil {
+		fmt.Fprintf(os.Stderr, "internal/app keychain cleanup: %v\n", err)
+		if code == 0 {
+			code = 1
+		}
+	}
 	if err := os.RemoveAll(tmpDir); err != nil {
 		fmt.Fprintf(os.Stderr, "internal/app test cleanup %s: %v\n", tmpDir, err)
 		if code == 0 {

internal/auth/access_token_snapshot_test.go

@@ -0,0 +1,70 @@
+package auth
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+)
+
+func TestCrossPlatformCoverageOAuthProviderTokenSnapshotPreservesLoadFailure(t *testing.T) {
+	oldLoad := oauthLoadToken
+	want := errors.New("keychain permission denied")
+	oauthLoadToken = func(string) (*TokenData, error) { return nil, want }
+	t.Cleanup(func() { oauthLoadToken = oldLoad })
+
+	_, err := NewOAuthProvider(t.TempDir(), nil).GetTokenSnapshot(context.Background())
+	if !errors.Is(err, want) {
+		t.Fatalf("error = %v, want cause %v", err, want)
+	}
+	if errors.Is(err, ErrTokenDataNotFound) {
+		t.Fatalf("load failure was misclassified as missing credentials: %v", err)
+	}
+}
+
+func TestCrossPlatformCoverageOAuthProviderLoginPreservesLoadFailure(t *testing.T) {
+	oldLoad := oauthLoadToken
+	want := errors.New("keychain permission denied")
+	oauthLoadToken = func(string) (*TokenData, error) { return nil, want }
+	t.Cleanup(func() { oauthLoadToken = oldLoad })
+
+	_, err := NewOAuthProvider(t.TempDir(), nil).Login(context.Background(), false)
+	if !errors.Is(err, want) {
+		t.Fatalf("error = %v, want cause %v", err, want)
+	}
+}
+
+func TestCrossPlatformCoverageOAuthProviderTokenSnapshotReturnsExpiryMetadata(t *testing.T) {
+	oldLoad := oauthLoadToken
+	expiresAt := time.Now().Add(time.Hour)
+	oauthLoadToken = func(string) (*TokenData, error) {
+		return &TokenData{AccessToken: "token", ExpiresAt: expiresAt}, nil
+	}
+	t.Cleanup(func() { oauthLoadToken = oldLoad })
+
+	snapshot, err := NewOAuthProvider(t.TempDir(), nil).GetTokenSnapshot(context.Background())
+	if err != nil {
+		t.Fatal(err)
+	}
+	if snapshot.AccessToken != "token" || !snapshot.ExpiresAt.Equal(expiresAt) {
+		t.Fatalf("snapshot = %#v", snapshot)
+	}
+}
+
+func TestCrossPlatformCoverageTokenMarkerRevisionChangesOnEveryPublication(t *testing.T) {
+	configDir := t.TempDir()
+	if err := WriteTokenMarker(configDir); err != nil {
+		t.Fatal(err)
+	}
+	first, present, err := ReadTokenMarkerRevision(configDir)
+	if err != nil || !present || first == "" {
+		t.Fatalf("first marker = %q, %v, %v", first, present, err)
+	}
+	if err := WriteTokenMarker(configDir); err != nil {
+		t.Fatal(err)
+	}
+	second, present, err := ReadTokenMarkerRevision(configDir)
+	if err != nil || !present || second == "" || second == first {
+		t.Fatalf("second marker = %q, %v, %v; first=%q", second, present, err, first)
+	}
+}

internal/auth/device_flow.go

@@ -722,6 +722,6 @@ func isInvalidGrantError(err error) bool {
 	if err == nil {
 		return false
 	}
-	msg := strings.ToLower(err.Error())
+	msg := strings.ToLower(err.Error() + " " + httpStatusResponseBody(err))
 	return strings.Contains(msg, "invalid_grant") || (strings.Contains(msg, "code") && strings.Contains(msg, "expired"))
 }

internal/auth/force_refresh.go

@@ -13,7 +13,50 @@
 
 package auth
 
-import "time"
+import (
+	"context"
+	"crypto/sha256"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"path/filepath"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
+)
+
+const rejectedTokenRefreshFailureCooldown = 3 * time.Second
+
+type rejectedTokenRefreshKey struct {
+	configDir   string
+	profile     string
+	tokenDigest [sha256.Size]byte
+}
+
+type rejectedTokenRefreshCall struct {
+	done         chan struct{}
+	participants int
+	token        string
+	err          error
+}
+
+type rejectedTokenRefreshFailure struct {
+	at  time.Time
+	err error
+}
+
+var rejectedTokenRefreshCoordinator = struct {
+	sync.Mutex
+	inFlight map[rejectedTokenRefreshKey]*rejectedTokenRefreshCall
+	failures map[rejectedTokenRefreshKey]rejectedTokenRefreshFailure
+	now      func() time.Time
+}{
+	inFlight: make(map[rejectedTokenRefreshKey]*rejectedTokenRefreshCall),
+	failures: make(map[rejectedTokenRefreshKey]rejectedTokenRefreshFailure),
+	now:      time.Now,
+}
 
 // MarkAccessTokenStale loads the persisted TokenData, sets ExpiresAt to a past
 // instant (preserving access_token and refresh_token), and writes it back. The
@@ -39,3 +82,192 @@ func MarkAccessTokenStale(configDir string) error {
 	data.ExpiresAt = time.Now().Add(-1 * time.Minute)
 	return SaveTokenData(configDir, data)
 }
+
+// ForceRefreshRejectedToken refreshes rejectedAccessToken only while it is
+// still the credential stored for the active profile. The compare and refresh
+// run under the same process + file lock used by ordinary expiry refresh, so a
+// late rejection cannot invalidate or refresh over a token another caller has
+// already rotated.
+//
+// When the stored token no longer matches, the newer token is returned without
+// calling the refresh endpoint. Refresh failures leave the stored credential in
+// place; login/logout remain the only owners of credential deletion.
+func (p *OAuthProvider) ForceRefreshRejectedToken(ctx context.Context, rejectedAccessToken string) (string, error) {
+	if p == nil || strings.TrimSpace(p.configDir) == "" {
+		return "", fmt.Errorf("config directory is empty")
+	}
+	rejectedAccessToken = strings.TrimSpace(rejectedAccessToken)
+	if rejectedAccessToken == "" {
+		return "", fmt.Errorf("rejected access token is empty")
+	}
+	if ctx == nil {
+		ctx = context.Background()
+	}
+
+	profile := strings.TrimSpace(RuntimeProfile())
+	key := newRejectedTokenRefreshKey(p.configDir, profile, rejectedAccessToken)
+	call, leader := beginRejectedTokenRefresh(key)
+	if !leader {
+		select {
+		case <-call.done:
+			return call.token, call.err
+		case <-ctx.Done():
+			return "", ctx.Err()
+		}
+	}
+
+	token, err, recordFailure := p.forceRefreshRejectedTokenOnce(ctx, profile, rejectedAccessToken, key)
+	finishRejectedTokenRefresh(key, call, token, err, recordFailure)
+	return token, err
+}
+
+func (p *OAuthProvider) forceRefreshRejectedTokenOnce(
+	ctx context.Context,
+	profile string,
+	rejectedAccessToken string,
+	key rejectedTokenRefreshKey,
+) (string, error, bool) {
+	lock, err := oauthAcquireLock(ctx, p.configDir)
+	if err != nil {
+		return "", fmt.Errorf("acquiring dual lock: %w", err), false
+	}
+	defer lock.Release()
+
+	data, err := loadOAuthTokenUnderHeldLock(p.configDir, profile)
+	if err != nil {
+		return "", fmt.Errorf("reload rejected token: %w", err), false
+	}
+	current := strings.TrimSpace(data.AccessToken)
+	if current == "" {
+		clearRejectedTokenRefreshFailure(key)
+		return "", fmt.Errorf("stored access token is empty"), false
+	}
+	if current != rejectedAccessToken {
+		clearRejectedTokenRefreshFailure(key)
+		return current, nil, false
+	}
+	if cachedErr := recentRejectedTokenRefreshFailure(key); cachedErr != nil {
+		return "", cachedErr, false
+	}
+	if !data.IsRefreshTokenValid() {
+		return "", fmt.Errorf("refresh_token 已过期"), true
+	}
+	if err := preflightTokenRefreshPersistence(p.configDir, data); err != nil {
+		return "", fmt.Errorf("本地登录态无法安全更新: %w", err), true
+	}
+
+	refreshed, err := oauthRefreshToken(p, ctx, data)
+	if err != nil {
+		return "", err, true
+	}
+	if refreshed == nil || strings.TrimSpace(refreshed.AccessToken) == "" {
+		return "", fmt.Errorf("force refresh returned empty access token"), true
+	}
+	return strings.TrimSpace(refreshed.AccessToken), nil, false
+}
+
+func newRejectedTokenRefreshKey(configDir, profile, rejectedAccessToken string) rejectedTokenRefreshKey {
+	canonicalDir := filepath.Clean(configDir)
+	if absolute, err := filepath.Abs(configDir); err == nil {
+		canonicalDir = filepath.Clean(absolute)
+	}
+	return rejectedTokenRefreshKey{
+		configDir:   canonicalDir,
+		profile:     strings.TrimSpace(profile),
+		tokenDigest: sha256.Sum256([]byte(strings.TrimSpace(rejectedAccessToken))),
+	}
+}
+
+func beginRejectedTokenRefresh(key rejectedTokenRefreshKey) (*rejectedTokenRefreshCall, bool) {
+	rejectedTokenRefreshCoordinator.Lock()
+	defer rejectedTokenRefreshCoordinator.Unlock()
+
+	if call := rejectedTokenRefreshCoordinator.inFlight[key]; call != nil {
+		call.participants++
+		return call, false
+	}
+	call := &rejectedTokenRefreshCall{done: make(chan struct{}), participants: 1}
+	rejectedTokenRefreshCoordinator.inFlight[key] = call
+	return call, true
+}
+
+func finishRejectedTokenRefresh(
+	key rejectedTokenRefreshKey,
+	call *rejectedTokenRefreshCall,
+	token string,
+	err error,
+	recordFailure bool,
+) {
+	rejectedTokenRefreshCoordinator.Lock()
+	defer rejectedTokenRefreshCoordinator.Unlock()
+
+	call.token = token
+	call.err = err
+	delete(rejectedTokenRefreshCoordinator.inFlight, key)
+	if err == nil {
+		delete(rejectedTokenRefreshCoordinator.failures, key)
+	} else if recordFailure && !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
+		now := time.Now()
+		if rejectedTokenRefreshCoordinator.now != nil {
+			now = rejectedTokenRefreshCoordinator.now()
+		}
+		rejectedTokenRefreshCoordinator.failures[key] = rejectedTokenRefreshFailure{at: now, err: err}
+	}
+	close(call.done)
+}
+
+func recentRejectedTokenRefreshFailure(key rejectedTokenRefreshKey) error {
+	rejectedTokenRefreshCoordinator.Lock()
+	defer rejectedTokenRefreshCoordinator.Unlock()
+
+	now := time.Now()
+	if rejectedTokenRefreshCoordinator.now != nil {
+		now = rejectedTokenRefreshCoordinator.now()
+	}
+	for failureKey, failure := range rejectedTokenRefreshCoordinator.failures {
+		age := now.Sub(failure.at)
+		if age < 0 || age >= rejectedTokenRefreshFailureCooldown {
+			delete(rejectedTokenRefreshCoordinator.failures, failureKey)
+		}
+	}
+	if failure, ok := rejectedTokenRefreshCoordinator.failures[key]; ok {
+		return failure.err
+	}
+	return nil
+}
+
+func clearRejectedTokenRefreshFailure(key rejectedTokenRefreshKey) {
+	rejectedTokenRefreshCoordinator.Lock()
+	delete(rejectedTokenRefreshCoordinator.failures, key)
+	rejectedTokenRefreshCoordinator.Unlock()
+}
+
+// loadOAuthTokenUnderHeldLock mirrors LoadTokenDataForProfile without taking a
+// second, non-reentrant auth lock. Opaque edition storage hooks (for example
+// Wukong's encrypted .data file) are read inside the caller's dual lock so the
+// compare-and-refresh decision covers both Core and embedded storage.
+func loadOAuthTokenUnderHeldLock(configDir, profile string) (*TokenData, error) {
+	hooks := edition.Get()
+	if hooks.LoadToken == nil {
+		data, err := oauthLoadTokenLocked(configDir, profile)
+		if err != nil {
+			return nil, err
+		}
+		if data == nil {
+			return nil, fmt.Errorf("stored token data is empty")
+		}
+		return data, nil
+	}
+	if strings.TrimSpace(profile) != "" {
+		return nil, fmt.Errorf("profile selection is not supported by the current auth backend")
+	}
+	blob, err := hooks.LoadToken(configDir)
+	if err != nil {
+		return nil, err
+	}
+	var data TokenData
+	if err := json.Unmarshal(blob, &data); err != nil {
+		return nil, fmt.Errorf("parsing token data from hook: %w", err)
+	}
+	return &data, nil
+}

internal/auth/manager.go

@@ -47,8 +47,10 @@ func (m *Manager) GetToken() (string, string, error) {
 		}
 		return token, "file", nil
 	}
-
-	return "", "", fmt.Errorf("%s", i18n.T("未找到认证信息，请运行 dws auth login"))
+	if err != nil && !os.IsNotExist(err) {
+		return "", "", fmt.Errorf("load legacy token: %w", err)
+	}
+	return "", "", fmt.Errorf("%s: %w", i18n.T("未找到认证信息，请运行 dws auth login"), ErrTokenDataNotFound)
 }
 
 func (m *Manager) GetMCPURL() (string, error) {

internal/auth/oauth_helpers.go

@@ -237,12 +237,21 @@ func (p *OAuthProvider) postJSON(ctx context.Context, endpoint string, body any)
 	}
 	defer resp.Body.Close()
 
-	data, err := io.ReadAll(io.LimitReader(resp.Body, config.MaxResponseBodySize))
-	if err != nil {
-		return nil, fmt.Errorf("reading response: %w", err)
-	}
+	data, readErr := io.ReadAll(io.LimitReader(resp.Body, config.MaxResponseBodySize))
 	if resp.StatusCode != http.StatusOK {
-		return nil, fmt.Errorf("HTTP %d: %s", resp.StatusCode, truncateBody(data, 200))
+		// Preserve structured HTTP status semantics even when the response
+		// body is truncated. The body is diagnostic-only here, so read it
+		// best-effort and classify retryability from the status code.
+		if readErr != nil {
+			data = nil
+		}
+		return nil, &HTTPStatusError{
+			StatusCode:   resp.StatusCode,
+			responseBody: truncateBody(data, 200),
+		}
+	}
+	if readErr != nil {
+		return nil, fmt.Errorf("reading response: %w", readErr)
 	}
 	return data, nil
 }

internal/auth/oauth_postjson_test.go

@@ -0,0 +1,69 @@
+// 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 auth
+
+import (
+	"context"
+	"errors"
+	"io"
+	"net/http"
+	"testing"
+)
+
+type postJSONRoundTripFunc func(*http.Request) (*http.Response, error)
+
+func (f postJSONRoundTripFunc) RoundTrip(req *http.Request) (*http.Response, error) {
+	return f(req)
+}
+
+type oauthBrokenBody struct{}
+
+func (oauthBrokenBody) Read([]byte) (int, error) { return 0, io.ErrUnexpectedEOF }
+func (oauthBrokenBody) Close() error             { return nil }
+
+func TestCrossPlatformCoveragePostJSONTruncatedErrorBodyKeepsHTTPStatus(t *testing.T) {
+	for _, status := range []int{http.StatusTooManyRequests, http.StatusServiceUnavailable} {
+		t.Run(http.StatusText(status), func(t *testing.T) {
+			provider := &OAuthProvider{httpClient: &http.Client{Transport: postJSONRoundTripFunc(func(*http.Request) (*http.Response, error) {
+				return &http.Response{StatusCode: status, Body: oauthBrokenBody{}, Header: make(http.Header)}, nil
+			})}}
+
+			_, err := provider.postJSON(context.Background(), "https://oauth.test/token", map[string]string{"grantType": "refresh_token"})
+			var statusErr *HTTPStatusError
+			if !errors.As(err, &statusErr) || statusErr.StatusCode != status {
+				t.Fatalf("postJSON() error = %v, want HTTPStatusError %d", err, status)
+			}
+			if errors.Is(err, io.ErrUnexpectedEOF) {
+				t.Fatalf("HTTP status error should not expose diagnostic body read failure: %v", err)
+			}
+			if got := ClassifyRefreshFailure(err); got != RefreshFailureTransient {
+				t.Fatalf("ClassifyRefreshFailure() = %s, want transient", got)
+			}
+		})
+	}
+}
+
+func TestCrossPlatformCoveragePostJSONOKTruncatedBodyIsTransient(t *testing.T) {
+	provider := &OAuthProvider{httpClient: &http.Client{Transport: postJSONRoundTripFunc(func(*http.Request) (*http.Response, error) {
+		return &http.Response{StatusCode: http.StatusOK, Body: oauthBrokenBody{}, Header: make(http.Header)}, nil
+	})}}
+
+	_, err := provider.postJSON(context.Background(), "https://oauth.test/token", map[string]string{"grantType": "refresh_token"})
+	if !errors.Is(err, io.ErrUnexpectedEOF) {
+		t.Fatalf("postJSON() error = %v, want io.ErrUnexpectedEOF", err)
+	}
+	if got := ClassifyRefreshFailure(err); got != RefreshFailureTransient {
+		t.Fatalf("ClassifyRefreshFailure() = %s, want transient", got)
+	}
+}

internal/auth/oauth_provider.go

@@ -18,6 +18,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"html"
 	"io"
 	"log/slog"
 	"net"
@@ -115,6 +116,12 @@ func (p *OAuthProvider) Login(ctx context.Context, force bool) (*TokenData, erro
 	// Smart degradation: try silent refresh before opening browser.
 	if !force {
 		data, err := oauthLoadToken(p.configDir)
+		if err != nil && !errors.Is(err, ErrTokenDataNotFound) && !os.IsNotExist(err) {
+			if preflightErr := preflightTokenPersistence(p.configDir); preflightErr != nil {
+				return nil, fmt.Errorf("%s: %w", i18n.T("本地登录态无法安全更新"), preflightErr)
+			}
+			return nil, fmt.Errorf("load existing access token: %w", err)
+		}
 		if err == nil {
 			// Case 1: access_token still valid — no action needed.
 			if data.IsAccessTokenValid() {
@@ -295,7 +302,7 @@ func (p *OAuthProvider) Login(ctx context.Context, force bool) (*TokenData, erro
 			callbackTokenMu.Unlock()
 
 			w.Header().Set("Content-Type", "text/html; charset=utf-8")
-			_, _ = fmt.Fprintf(w, "<html><body><h1>授权失败</h1><p>%s</p></body></html>", exchangeErr.Error())
+			_, _ = fmt.Fprintf(w, "<html><body><h1>授权失败</h1><p>%s</p></body></html>", html.EscapeString(oauthExchangeDisplayError(exchangeErr)))
 			select {
 			case resultCh <- callbackResult{err: exchangeErr}:
 			default:
@@ -636,36 +643,63 @@ continueLogin:
 	return tokenData, nil
 }
 
-// GetAccessToken returns a valid access token, auto-refreshing if needed.
-// Uses a file lock with double-check pattern to prevent concurrent refresh
-// from multiple CLI processes.
-func (p *OAuthProvider) GetAccessToken(ctx context.Context) (string, error) {
+func oauthExchangeDisplayError(err error) string {
+	var statusErr *HTTPStatusError
+	if errors.As(err, &statusErr) && statusErr != nil {
+		return fmt.Sprintf("HTTP %d: token exchange failed", statusErr.StatusCode)
+	}
+	return err.Error()
+}
+
+// GetTokenSnapshot returns a valid token together with its expiry metadata.
+// Storage and refresh failures retain their original cause; only a confirmed
+// missing credential is reported as ErrTokenDataNotFound.
+func (p *OAuthProvider) GetTokenSnapshot(ctx context.Context) (*TokenData, error) {
 	data, err := oauthLoadToken(p.configDir)
 	if err != nil {
-		return "", errors.New(i18n.T("未登录，请运行 dws auth login"))
+		if errors.Is(err, ErrTokenDataNotFound) || os.IsNotExist(err) {
+			return nil, fmt.Errorf("%s: %w", i18n.T("未登录，请运行 dws auth login"), ErrTokenDataNotFound)
+		}
+		return nil, fmt.Errorf("load access token: %w", err)
 	}
 
 	// Fast path: access_token still valid — no lock needed.
 	if data.IsAccessTokenValid() {
-		return data.AccessToken, nil
+		return data, nil
 	}
 
 	// Slow path: token expired — try locked refresh.
 	if data.IsRefreshTokenValid() {
 		refreshed, rErr := p.lockedRefresh(ctx)
 		if rErr == nil {
-			return refreshed.AccessToken, nil
+			return refreshed, nil
+		}
+		// A network, timeout, rate-limit or 5xx failure does not invalidate the
+		// refresh credential. Keep the profile active so a long-running source
+		// can retry after backoff. Terminal and unknown failures remain fatal.
+		if ClassifyRefreshFailure(rErr) != RefreshFailureTransient {
+			_ = oauthMarkProfile(p.configDir, TokenProfileSelector(data), ProfileStatusExpired)
 		}
-		_ = oauthMarkProfile(p.configDir, TokenProfileSelector(data), ProfileStatusExpired)
 		if p.logger != nil {
 			p.logger.Warn(i18n.T("refresh_token 刷新失败"), "error", rErr)
 		}
-		return "", fmt.Errorf("%s: %w", i18n.T("refresh_token 刷新失败"), rErr)
+		return nil, fmt.Errorf("%s: %w", i18n.T("refresh_token 刷新失败"), rErr)
 	} else {
 		_ = oauthMarkProfile(p.configDir, TokenProfileSelector(data), ProfileStatusExpired)
 	}
 
-	return "", errors.New(i18n.T("所有凭证已失效，请运行 dws auth login 重新登录"))
+	return nil, fmt.Errorf("%s: %w", i18n.T("所有凭证已失效，请运行 dws auth login 重新登录"), ErrTokenDataNotFound)
+}
+
+// GetAccessToken returns a valid access token, auto-refreshing if needed.
+// Uses a file lock with double-check pattern to prevent concurrent refresh
+// from multiple CLI processes.
+func (p *OAuthProvider) GetAccessToken(ctx context.Context) (string, error) {
+	data, err := p.GetTokenSnapshot(ctx)
+	if err != nil {
+		return "", err
+	}
+	return strings.TrimSpace(data.AccessToken), nil
 }
 
 // lockedRefresh attempts to refresh the token while holding dual-layer locks.
@@ -697,7 +731,7 @@ func (p *OAuthProvider) lockedRefresh(ctx context.Context) (*TokenData, error) {
 
 	// Double-check: re-load from disk — another goroutine/process may have refreshed
 	// while we were waiting for the lock.
-	data, err := oauthLoadTokenLocked(p.configDir, RuntimeProfile())
+	data, err := loadOAuthTokenUnderHeldLock(p.configDir, RuntimeProfile())
 	if err != nil {
 		return nil, err
 	}

internal/auth/oauth_provider_coverage_test.go

@@ -18,6 +18,8 @@ import (
 	"sync/atomic"
 	"testing"
 	"time"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/keychain"
 )
 
 type oauthLoginFixture struct {
@@ -33,8 +35,79 @@ type oauthLoginFixture struct {
 	exchangeError   atomic.Bool
 }
 
+type oauthLoginResult struct {
+	token *TokenData
+	err   error
+}
+
+type oauthHTTPResult struct {
+	status int
+	body   string
+	err    error
+}
+
+const oauthTestWaitTimeout = 5 * time.Second
+
+func isolateOAuthPersistence(t *testing.T) {
+	t.Helper()
+	t.Setenv(keychain.DisableKeychainEnv, "1")
+	cleanupKeychain(t)
+}
+
+func startOAuthLogin(t *testing.T, parent context.Context, f *oauthLoginFixture) <-chan oauthLoginResult {
+	t.Helper()
+	ctx, cancel := context.WithCancel(parent)
+	done := make(chan oauthLoginResult, 1)
+	finished := make(chan struct{})
+	go func() {
+		defer close(finished)
+		token, err := f.provider.Login(ctx, true)
+		done <- oauthLoginResult{token: token, err: err}
+	}()
+	t.Cleanup(func() {
+		cancel()
+		select {
+		case <-finished:
+		case <-time.After(oauthTestWaitTimeout):
+			t.Errorf("OAuth Login goroutine did not stop after cancellation")
+		}
+	})
+	return done
+}
+
+func awaitOAuthLogin(t *testing.T, done <-chan oauthLoginResult) oauthLoginResult {
+	t.Helper()
+	select {
+	case result := <-done:
+		return result
+	case <-time.After(oauthTestWaitTimeout):
+		t.Fatal("timed out waiting for OAuth Login")
+		return oauthLoginResult{}
+	}
+}
+
+func waitOAuthSignal(t *testing.T, signal <-chan struct{}, done <-chan oauthLoginResult, name string) {
+	t.Helper()
+	select {
+	case <-signal:
+	case result := <-done:
+		t.Fatalf("OAuth Login returned before %s: token=%#v err=%v", name, result.token, result.err)
+	case <-time.After(oauthTestWaitTimeout):
+		t.Fatalf("timed out waiting for OAuth %s", name)
+	}
+}
+
+func closeOAuthRelease(ch chan struct{}) {
+	select {
+	case <-ch:
+	default:
+		close(ch)
+	}
+}
+
 func newOAuthLoginFixture(t *testing.T, status func(int32) CLIAuthStatus) *oauthLoginFixture {
 	t.Helper()
+	isolateOAuthPersistence(t)
 	SetClientID("")
 	SetClientSecret("")
 	resetClientIDFromMCP()
@@ -91,6 +164,10 @@ func newOAuthLoginFixture(t *testing.T, status func(int32) CLIAuthStatus) *oauth
 		}
 	}))
 	t.Cleanup(f.server.Close)
+	t.Cleanup(func() {
+		closeOAuthRelease(f.exchangeRelease)
+		closeOAuthRelease(f.statusRelease)
+	})
 	f.configDir = setupMCPConfigDir(t, f.server.URL)
 
 	oldClient := oauthHTTPClient
@@ -118,17 +195,25 @@ func newOAuthLoginFixture(t *testing.T, status func(int32) CLIAuthStatus) *oauth
 
 func httpGetBody(t *testing.T, rawURL string) (int, string) {
 	t.Helper()
+	result := getHTTPBody(rawURL)
+	if result.err != nil {
+		t.Fatalf("GET %s: %v", rawURL, result.err)
+	}
+	return result.status, result.body
+}
+
+func getHTTPBody(rawURL string) oauthHTTPResult {
 	client := &http.Client{Timeout: 2 * time.Second}
 	resp, err := client.Get(rawURL)
 	if err != nil {
-		t.Fatalf("GET %s: %v", rawURL, err)
+		return oauthHTTPResult{err: err}
 	}
 	defer resp.Body.Close()
 	data, err := io.ReadAll(resp.Body)
 	if err != nil {
-		t.Fatal(err)
+		return oauthHTTPResult{err: err}
 	}
-	return resp.StatusCode, string(data)
+	return oauthHTTPResult{status: resp.StatusCode, body: string(data)}
 }
 
 func TestCrossPlatformCoverageOAuthLoginCallbackAndAPIs(t *testing.T) {
@@ -136,17 +221,7 @@ func TestCrossPlatformCoverageOAuthLoginCallbackAndAPIs(t *testing.T) {
 		return CLIAuthStatus{Success: true, Result: &CLIAuthResult{CLIAuthEnabled: true}}
 	})
 
-	loginDone := make(chan struct {
-		token *TokenData
-		err   error
-	}, 1)
-	go func() {
-		token, err := f.provider.Login(context.Background(), true)
-		loginDone <- struct {
-			token *TokenData
-			err   error
-		}{token, err}
-	}()
+	loginDone := startOAuthLogin(t, context.Background(), f)
 
 	for _, path := range []string{"/api/superAdmin", "/api/sendApply?adminStaffId=admin-1", "/api/cliAuthEnabled"} {
 		_, body := httpGetBody(t, f.callbackBase+path)
@@ -166,18 +241,17 @@ func TestCrossPlatformCoverageOAuthLoginCallbackAndAPIs(t *testing.T) {
 		t.Fatalf("success page = %q", body)
 	}
 
-	callbackDone := make(chan string, 1)
+	callbackDone := make(chan oauthHTTPResult, 1)
 	go func() {
-		_, callbackBody := httpGetBody(t, f.callbackBase+CallbackPath+"?code=good")
-		callbackDone <- callbackBody
+		callbackDone <- getHTTPBody(f.callbackBase + CallbackPath + "?code=good")
 	}()
-	<-f.exchangeEntered
+	waitOAuthSignal(t, f.exchangeEntered, loginDone, "token exchange")
 	_, body = httpGetBody(t, f.callbackBase+CallbackPath+"?authCode=good")
 	if !strings.Contains(body, "正在处理授权") {
 		t.Fatalf("concurrent callback = %q", body)
 	}
-	close(f.exchangeRelease)
-	<-f.statusEntered
+	closeOAuthRelease(f.exchangeRelease)
+	waitOAuthSignal(t, f.statusEntered, loginDone, "CLI auth status check")
 
 	_, body = httpGetBody(t, f.callbackBase+CallbackPath+"?code=good")
 	if !strings.Contains(body, "<html") {
@@ -204,11 +278,16 @@ func TestCrossPlatformCoverageOAuthLoginCallbackAndAPIs(t *testing.T) {
 		t.Fatalf("auth enabled API = %q", body)
 	}
 
-	close(f.statusRelease)
-	if callbackBody := <-callbackDone; !strings.Contains(callbackBody, "<html") {
-		t.Fatalf("callback body = %q", callbackBody)
+	closeOAuthRelease(f.statusRelease)
+	select {
+	case callback := <-callbackDone:
+		if callback.err != nil || !strings.Contains(callback.body, "<html") {
+			t.Fatalf("callback body = %q, %v", callback.body, callback.err)
+		}
+	case <-time.After(oauthTestWaitTimeout):
+		t.Fatal("timed out waiting for OAuth callback")
 	}
-	result := <-loginDone
+	result := awaitOAuthLogin(t, loginDone)
 	if result.err != nil || result.token == nil || result.token.AccessToken != "access" {
 		t.Fatalf("Login = %#v, %v", result.token, result.err)
 	}
@@ -228,23 +307,20 @@ func TestCrossPlatformCoverageOAuthLoginMissingCallbackCode(t *testing.T) {
 	f := newOAuthLoginFixture(t, func(int32) CLIAuthStatus {
 		return CLIAuthStatus{Success: true, Result: &CLIAuthResult{CLIAuthEnabled: true}}
 	})
-	close(f.exchangeRelease)
-	close(f.statusRelease)
-	done := make(chan error, 1)
-	go func() {
-		_, err := f.provider.Login(context.Background(), true)
-		done <- err
-	}()
+	closeOAuthRelease(f.exchangeRelease)
+	closeOAuthRelease(f.statusRelease)
+	done := startOAuthLogin(t, context.Background(), f)
 	status, body := httpGetBody(t, f.callbackBase+CallbackPath)
 	if status != http.StatusBadRequest || strings.TrimSpace(body) == "" {
 		t.Fatalf("missing callback = %d %q", status, body)
 	}
-	if err := <-done; err == nil {
+	if result := awaitOAuthLogin(t, done); result.err == nil {
 		t.Fatal("missing callback code did not fail login")
 	}
 }
 
 func TestCrossPlatformCoverageOAuthLoginEarlyAndListenerEdges(t *testing.T) {
+	isolateOAuthPersistence(t)
 	var buf bytes.Buffer
 	p := &OAuthProvider{configDir: t.TempDir(), Output: &buf}
 	if p.output() != &buf || (*OAuthProvider)(nil).output() != io.Discard {
@@ -311,6 +387,7 @@ func TestCrossPlatformCoverageOAuthLoginTimeoutAndServerError(t *testing.T) {
 }
 
 func TestCrossPlatformCoverageOAuthProviderOtherMethods(t *testing.T) {
+	isolateOAuthPersistence(t)
 	dir := t.TempDir()
 	_ = DeleteTokenDataKeychain()
 	t.Cleanup(func() { _ = DeleteTokenDataKeychain() })
@@ -363,6 +440,7 @@ func (f roundTripFunc) RoundTrip(r *http.Request) (*http.Response, error) {
 }
 
 func TestCrossPlatformCoverageOAuthPersistConfigEdges(t *testing.T) {
+	isolateOAuthPersistence(t)
 	p := &OAuthProvider{configDir: t.TempDir(), logger: slog.Default()}
 	SetClientID("")
 	SetClientSecret("")
@@ -425,16 +503,12 @@ func TestCrossPlatformCoverageOAuthLoginDenialAndPolling(t *testing.T) {
 	for _, tt := range terminal {
 		t.Run(tt.name, func(t *testing.T) {
 			f := newOAuthLoginFixture(t, func(int32) CLIAuthStatus { return tt.status })
-			close(f.exchangeRelease)
-			close(f.statusRelease)
+			closeOAuthRelease(f.exchangeRelease)
+			closeOAuthRelease(f.statusRelease)
 			f.provider.NoBrowser = false
-			done := make(chan error, 1)
-			go func() {
-				_, err := f.provider.Login(context.Background(), true)
-				done <- err
-			}()
+			done := startOAuthLogin(t, context.Background(), f)
 			_, _ = httpGetBody(t, f.callbackBase+CallbackPath+"?code=denied")
-			if err := <-done; err == nil {
+			if result := awaitOAuthLogin(t, done); result.err == nil {
 				t.Fatal("denied login succeeded")
 			}
 		})
@@ -445,15 +519,11 @@ func TestCrossPlatformCoverageOAuthLoginDenialAndPolling(t *testing.T) {
 		f := newOAuthLoginFixture(t, func(int32) CLIAuthStatus {
 			return CLIAuthStatus{Success: true, Result: &CLIAuthResult{ChannelScope: "specified", AllowedChannels: []string{"allowed"}}}
 		})
-		close(f.exchangeRelease)
-		close(f.statusRelease)
-		done := make(chan error, 1)
-		go func() {
-			_, err := f.provider.Login(context.Background(), true)
-			done <- err
-		}()
+		closeOAuthRelease(f.exchangeRelease)
+		closeOAuthRelease(f.statusRelease)
+		done := startOAuthLogin(t, context.Background(), f)
 		_, _ = httpGetBody(t, f.callbackBase+CallbackPath+"?code=denied")
-		if err := <-done; err == nil {
+		if result := awaitOAuthLogin(t, done); result.err == nil {
 			t.Fatal("channel-denied login succeeded")
 		}
 	})
@@ -463,15 +533,11 @@ func TestCrossPlatformCoverageOAuthLoginDenialAndPolling(t *testing.T) {
 		f := newOAuthLoginFixture(t, func(int32) CLIAuthStatus {
 			return CLIAuthStatus{Success: true, Result: &CLIAuthResult{}}
 		})
-		close(f.exchangeRelease)
-		close(f.statusRelease)
-		done := make(chan error, 1)
-		go func() {
-			_, err := f.provider.Login(context.Background(), true)
-			done <- err
-		}()
+		closeOAuthRelease(f.exchangeRelease)
+		closeOAuthRelease(f.statusRelease)
+		done := startOAuthLogin(t, context.Background(), f)
 		_, _ = httpGetBody(t, f.callbackBase+CallbackPath+"?code=pending")
-		if err := <-done; err == nil {
+		if result := awaitOAuthLogin(t, done); result.err == nil {
 			t.Fatal("approval timeout login succeeded")
 		}
 		oauthApprovalTimeout = oldApproval
@@ -481,16 +547,12 @@ func TestCrossPlatformCoverageOAuthLoginDenialAndPolling(t *testing.T) {
 		f := newOAuthLoginFixture(t, func(call int32) CLIAuthStatus {
 			return CLIAuthStatus{Success: true, Result: &CLIAuthResult{CLIAuthEnabled: call > 1}}
 		})
-		close(f.exchangeRelease)
-		close(f.statusRelease)
-		done := make(chan error, 1)
-		go func() {
-			_, err := f.provider.Login(context.Background(), true)
-			done <- err
-		}()
+		closeOAuthRelease(f.exchangeRelease)
+		closeOAuthRelease(f.statusRelease)
+		done := startOAuthLogin(t, context.Background(), f)
 		_, _ = httpGetBody(t, f.callbackBase+CallbackPath+"?code=pending")
-		if err := <-done; err != nil {
-			t.Fatalf("poll-enabled login failed: %v", err)
+		if result := awaitOAuthLogin(t, done); result.err != nil {
+			t.Fatalf("poll-enabled login failed: %v", result.err)
 		}
 	})
 
@@ -498,17 +560,13 @@ func TestCrossPlatformCoverageOAuthLoginDenialAndPolling(t *testing.T) {
 		f := newOAuthLoginFixture(t, func(int32) CLIAuthStatus {
 			return CLIAuthStatus{Success: true, Result: &CLIAuthResult{}}
 		})
-		close(f.exchangeRelease)
-		close(f.statusRelease)
+		closeOAuthRelease(f.exchangeRelease)
+		closeOAuthRelease(f.statusRelease)
 		ctx, cancel := context.WithCancel(context.Background())
-		done := make(chan error, 1)
-		go func() {
-			_, err := f.provider.Login(ctx, true)
-			done <- err
-		}()
+		done := startOAuthLogin(t, ctx, f)
 		_, _ = httpGetBody(t, f.callbackBase+CallbackPath+"?code=pending")
 		cancel()
-		if err := <-done; err == nil {
+		if result := awaitOAuthLogin(t, done); result.err == nil {
 			t.Fatal("canceled pending login succeeded")
 		}
 	})
@@ -519,23 +577,20 @@ func TestCrossPlatformCoverageOAuthLoginExchangeFailure(t *testing.T) {
 		return CLIAuthStatus{Success: true, Result: &CLIAuthResult{CLIAuthEnabled: true}}
 	})
 	f.exchangeError.Store(true)
-	close(f.exchangeRelease)
-	close(f.statusRelease)
-	done := make(chan error, 1)
-	go func() {
-		_, err := f.provider.Login(context.Background(), true)
-		done <- err
-	}()
+	closeOAuthRelease(f.exchangeRelease)
+	closeOAuthRelease(f.statusRelease)
+	done := startOAuthLogin(t, context.Background(), f)
 	_, body := httpGetBody(t, f.callbackBase+CallbackPath+"?code=bad")
 	if !strings.Contains(body, "failed") {
 		t.Fatalf("exchange failure page = %q", body)
 	}
-	if err := <-done; err == nil {
+	if result := awaitOAuthLogin(t, done); result.err == nil {
 		t.Fatal("exchange failure login succeeded")
 	}
 }
 
 func TestCrossPlatformCoverageOAuthRefreshAndParsingEdges(t *testing.T) {
+	isolateOAuthPersistence(t)
 	t.Setenv("DWS_CLIENT_ID", "")
 	t.Setenv("DWS_CLIENT_SECRET", "")
 	dir := t.TempDir()
@@ -640,6 +695,7 @@ func TestCrossPlatformCoverageOAuthRefreshAndParsingEdges(t *testing.T) {
 }
 
 func TestCrossPlatformCoverageOAuthProviderHighLevelEdges(t *testing.T) {
+	isolateOAuthPersistence(t)
 	oldLoad := oauthLoadToken
 	oldLoadLocked := oauthLoadTokenLocked
 	oldAcquire := oauthAcquireLock
@@ -800,29 +856,24 @@ func TestCrossPlatformCoverageOAuthCallbackRemainingEdges(t *testing.T) {
 	oauthSuccessPause = 0
 	oauthSleep = func(time.Duration) {}
 
-	type loginResult struct {
-		token *TokenData
-		err   error
-	}
-	startLogin := func(ctx context.Context, f *oauthLoginFixture) <-chan loginResult {
+	finishExchange := func(t *testing.T, f *oauthLoginFixture, done <-chan oauthLoginResult, code string) string {
 		t.Helper()
-		done := make(chan loginResult, 1)
+		bodyCh := make(chan oauthHTTPResult, 1)
 		go func() {
-			token, err := f.provider.Login(ctx, true)
-			done <- loginResult{token: token, err: err}
+			bodyCh <- getHTTPBody(f.callbackBase + CallbackPath + "?code=" + url.QueryEscape(code))
 		}()
-		return done
-	}
-	finishExchange := func(t *testing.T, f *oauthLoginFixture, code string) string {
-		t.Helper()
-		bodyCh := make(chan string, 1)
-		go func() {
-			_, body := httpGetBody(t, f.callbackBase+CallbackPath+"?code="+url.QueryEscape(code))
-			bodyCh <- body
-		}()
-		<-f.exchangeEntered
-		close(f.exchangeRelease)
-		return <-bodyCh
+		waitOAuthSignal(t, f.exchangeEntered, done, "token exchange")
+		closeOAuthRelease(f.exchangeRelease)
+		select {
+		case result := <-bodyCh:
+			if result.err != nil {
+				t.Fatalf("OAuth callback failed: %v", result.err)
+			}
+			return result.body
+		case <-time.After(oauthTestWaitTimeout):
+			t.Fatal("timed out waiting for OAuth callback")
+			return ""
+		}
 	}
 
 	t.Run("switch organization and cached disabled pages", func(t *testing.T) {
@@ -835,8 +886,8 @@ func TestCrossPlatformCoverageOAuthCallbackRemainingEdges(t *testing.T) {
 			return &CLIAuthStatus{Success: true, Result: &CLIAuthResult{CLIAuthEnabled: true}}, nil
 		}
 		f := newOAuthLoginFixture(t, func(int32) CLIAuthStatus { return CLIAuthStatus{} })
-		done := startLogin(context.Background(), f)
-		if body := finishExchange(t, f, "first"); !strings.Contains(body, "<html") {
+		done := startOAuthLogin(t, context.Background(), f)
+		if body := finishExchange(t, f, done, "first"); !strings.Contains(body, "<html") {
 			t.Fatalf("disabled callback body = %q", body)
 		}
 		if _, body := httpGetBody(t, f.callbackBase+CallbackPath+"?code=first"); !strings.Contains(body, "<html") {
@@ -848,7 +899,7 @@ func TestCrossPlatformCoverageOAuthCallbackRemainingEdges(t *testing.T) {
 		if _, body := httpGetBody(t, f.callbackBase+CallbackPath+"?code=second"); !strings.Contains(body, "<html") {
 			t.Fatalf("switched callback = %q", body)
 		}
-		result := <-done
+		result := awaitOAuthLogin(t, done)
 		if result.err != nil || result.token == nil || result.token.AccessToken != "access" {
 			t.Fatalf("switched login = %#v %v", result.token, result.err)
 		}
@@ -862,15 +913,15 @@ func TestCrossPlatformCoverageOAuthCallbackRemainingEdges(t *testing.T) {
 		oauthSendApply = func(context.Context, string, string) (*SendApplyResponse, error) { return nil, fail }
 		ctx, cancel := context.WithCancel(context.Background())
 		f := newOAuthLoginFixture(t, func(int32) CLIAuthStatus { return CLIAuthStatus{} })
-		done := startLogin(ctx, f)
-		finishExchange(t, f, "errors")
+		done := startOAuthLogin(t, ctx, f)
+		finishExchange(t, f, done, "errors")
 		for _, path := range []string{"/api/superAdmin", "/api/sendApply?adminStaffId=admin", "/api/cliAuthEnabled"} {
 			if _, body := httpGetBody(t, f.callbackBase+path); !strings.Contains(body, "hook failure") {
 				t.Fatalf("API error %s = %q", path, body)
 			}
 		}
 		cancel()
-		if result := <-done; !errors.Is(result.err, context.Canceled) {
+		if result := awaitOAuthLogin(t, done); !errors.Is(result.err, context.Canceled) {
 			t.Fatalf("canceled error login = %v", result.err)
 		}
 	})
@@ -887,14 +938,14 @@ func TestCrossPlatformCoverageOAuthCallbackRemainingEdges(t *testing.T) {
 		f := newOAuthLoginFixture(t, func(int32) CLIAuthStatus { return CLIAuthStatus{} })
 		var output bytes.Buffer
 		f.provider.Output = &output
-		done := startLogin(ctx, f)
-		finishExchange(t, f, "apply")
+		done := startOAuthLogin(t, ctx, f)
+		finishExchange(t, f, done, "apply")
 		if _, body := httpGetBody(t, f.callbackBase+"/api/sendApply?adminStaffId=admin"); !strings.Contains(body, "true") {
 			t.Fatalf("apply response = %q", body)
 		}
 		time.Sleep(20 * time.Millisecond)
 		cancel()
-		if result := <-done; !errors.Is(result.err, context.Canceled) {
+		if result := awaitOAuthLogin(t, done); !errors.Is(result.err, context.Canceled) {
 			t.Fatalf("canceled apply login = %v", result.err)
 		}
 		if !strings.Contains(output.String(), "Waiting for admin approval") && !strings.Contains(output.String(), "等待管理员审批中") {
@@ -908,9 +959,9 @@ func TestCrossPlatformCoverageOAuthCallbackRemainingEdges(t *testing.T) {
 			return &CLIAuthStatus{Success: false, ErrorCode: "ENTERPRISE_NOT_AUTHORIZED"}, nil
 		}
 		f := newOAuthLoginFixture(t, func(int32) CLIAuthStatus { return CLIAuthStatus{} })
-		done := startLogin(context.Background(), f)
-		finishExchange(t, f, "enterprise")
-		if result := <-done; result.err == nil || !strings.Contains(result.err.Error(), "企业安全认证") {
+		done := startOAuthLogin(t, context.Background(), f)
+		finishExchange(t, f, done, "enterprise")
+		if result := awaitOAuthLogin(t, done); result.err == nil || !strings.Contains(result.err.Error(), "企业安全认证") {
 			t.Fatalf("enterprise denial = %v", result.err)
 		}
 	})
@@ -933,15 +984,16 @@ func TestCrossPlatformCoverageOAuthCallbackRemainingEdges(t *testing.T) {
 		fail := errors.New("save failure")
 		oauthSaveToken = func(string, *TokenData) error { return fail }
 		f := newOAuthLoginFixture(t, func(int32) CLIAuthStatus { return CLIAuthStatus{} })
-		done := startLogin(context.Background(), f)
-		finishExchange(t, f, "save")
-		if result := <-done; !errors.Is(result.err, fail) {
+		done := startOAuthLogin(t, context.Background(), f)
+		finishExchange(t, f, done, "save")
+		if result := awaitOAuthLogin(t, done); !errors.Is(result.err, fail) {
 			t.Fatalf("save failure login = %v", result.err)
 		}
 	})
 }
 
 func TestCrossPlatformCoverageOAuthHelperRemainingEdges(t *testing.T) {
+	isolateOAuthPersistence(t)
 	oldClient := oauthHTTPClient
 	oldRequest := oauthNewRequest
 	oldRetry := oauthRetryAfter

internal/auth/refresh_failure.go

@@ -0,0 +1,89 @@
+// 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 auth
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"net"
+	"net/http"
+)
+
+// RefreshFailureClass separates refresh failures that may recover after a
+// delay from failures that require new credentials or local intervention.
+type RefreshFailureClass string
+
+const (
+	RefreshFailureUnknown   RefreshFailureClass = "unknown"
+	RefreshFailureTransient RefreshFailureClass = "transient"
+	RefreshFailureTerminal  RefreshFailureClass = "terminal"
+)
+
+// HTTPStatusError preserves an OAuth endpoint status for structured retry
+// decisions without copying an untrusted response body into logs.
+type HTTPStatusError struct {
+	StatusCode   int
+	responseBody string
+}
+
+func (e *HTTPStatusError) Error() string {
+	if e == nil {
+		return "OAuth endpoint request failed"
+	}
+	return fmt.Sprintf("HTTP %d", e.StatusCode)
+}
+
+func httpStatusResponseBody(err error) string {
+	var statusErr *HTTPStatusError
+	if !errors.As(err, &statusErr) || statusErr == nil {
+		return ""
+	}
+	return statusErr.responseBody
+}
+
+// ClassifyRefreshFailure uses only structured transport and HTTP signals.
+// Unknown errors, including parse, keychain and persistence failures, remain
+// fatal so a long-running source cannot retry an error that needs user action.
+func ClassifyRefreshFailure(err error) RefreshFailureClass {
+	if err == nil {
+		return RefreshFailureUnknown
+	}
+	if errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
+		return RefreshFailureTransient
+	}
+	if errors.Is(err, io.ErrUnexpectedEOF) {
+		return RefreshFailureTransient
+	}
+	var netErr net.Error
+	if errors.As(err, &netErr) {
+		return RefreshFailureTransient
+	}
+	var statusErr *HTTPStatusError
+	if !errors.As(err, &statusErr) || statusErr == nil {
+		return RefreshFailureUnknown
+	}
+	if statusErr.StatusCode == http.StatusRequestTimeout ||
+		statusErr.StatusCode == http.StatusTooManyRequests ||
+		statusErr.StatusCode >= http.StatusInternalServerError {
+		return RefreshFailureTransient
+	}
+	if statusErr.StatusCode == http.StatusBadRequest ||
+		statusErr.StatusCode == http.StatusUnauthorized ||
+		statusErr.StatusCode == http.StatusForbidden {
+		return RefreshFailureTerminal
+	}
+	return RefreshFailureUnknown
+}

internal/auth/refresh_failure_test.go

@@ -0,0 +1,155 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0 (the "License");
+
+package auth
+
+import (
+	"context"
+	"errors"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
+)
+
+func TestCrossPlatformCoverageClassifyRefreshFailureUsesStructuredSignals(t *testing.T) {
+	tests := []struct {
+		name string
+		err  error
+		want RefreshFailureClass
+	}{
+		{name: "deadline", err: context.DeadlineExceeded, want: RefreshFailureTransient},
+		{name: "network", err: &url.Error{Op: "Post", URL: "https://oauth.test", Err: context.DeadlineExceeded}, want: RefreshFailureTransient},
+		{name: "request timeout", err: &HTTPStatusError{StatusCode: http.StatusRequestTimeout}, want: RefreshFailureTransient},
+		{name: "rate limited", err: &HTTPStatusError{StatusCode: http.StatusTooManyRequests}, want: RefreshFailureTransient},
+		{name: "server unavailable", err: &HTTPStatusError{StatusCode: http.StatusServiceUnavailable}, want: RefreshFailureTransient},
+		{name: "refresh rejected", err: &HTTPStatusError{StatusCode: http.StatusUnauthorized}, want: RefreshFailureTerminal},
+		{name: "invalid grant", err: &HTTPStatusError{StatusCode: http.StatusBadRequest}, want: RefreshFailureTerminal},
+		{name: "forbidden", err: &HTTPStatusError{StatusCode: http.StatusForbidden}, want: RefreshFailureTerminal},
+		{name: "local persistence", err: errors.New("save refreshed token failed"), want: RefreshFailureUnknown},
+		{name: "nil error", err: nil, want: RefreshFailureUnknown},
+		{name: "dns failure", err: &net.DNSError{Err: "no such host", Name: "oauth.test"}, want: RefreshFailureTransient},
+		{name: "redirect status", err: &HTTPStatusError{StatusCode: http.StatusFound}, want: RefreshFailureUnknown},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := ClassifyRefreshFailure(tt.err); got != tt.want {
+				t.Fatalf("ClassifyRefreshFailure() = %q, want %q", got, tt.want)
+			}
+		})
+	}
+}
+
+func TestCrossPlatformCoverageHTTPStatusErrorRetainsStatusThroughWrapping(t *testing.T) {
+	want := &HTTPStatusError{StatusCode: http.StatusTooManyRequests}
+	err := errors.Join(errors.New("refresh failed"), want)
+	if got := ClassifyRefreshFailure(err); got != RefreshFailureTransient {
+		t.Fatalf("ClassifyRefreshFailure() = %q, want transient", got)
+	}
+	var statusErr *HTTPStatusError
+	if !errors.As(err, &statusErr) || statusErr.StatusCode != http.StatusTooManyRequests {
+		t.Fatalf("HTTP status error not retained: %v", err)
+	}
+	if got, want := statusErr.Error(), "HTTP 429"; got != want {
+		t.Fatalf("HTTP status error = %q, want %q", got, want)
+	}
+	var nilStatus *HTTPStatusError
+	if got, want := nilStatus.Error(), "OAuth endpoint request failed"; got != want {
+		t.Fatalf("nil HTTP status error = %q, want %q", got, want)
+	}
+}
+
+func TestCrossPlatformCoverageOAuthExchangeDisplayErrorFallsBackToPlainError(t *testing.T) {
+	if got, want := oauthExchangeDisplayError(&HTTPStatusError{StatusCode: http.StatusBadGateway}), "HTTP 502: token exchange failed"; got != want {
+		t.Fatalf("status display error = %q, want %q", got, want)
+	}
+	if got, want := oauthExchangeDisplayError(errors.New("exchange failed")), "exchange failed"; got != want {
+		t.Fatalf("plain display error = %q, want %q", got, want)
+	}
+}
+
+func TestCrossPlatformCoveragePostJSONClassifiesStatusWithoutLoggingResponseBody(t *testing.T) {
+	const secretBody = `{"refreshToken":"must-not-reach-logs"}`
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusServiceUnavailable)
+		_, _ = w.Write([]byte(secretBody))
+	}))
+	defer server.Close()
+
+	provider := &OAuthProvider{httpClient: server.Client()}
+	_, err := provider.postJSON(context.Background(), server.URL, map[string]string{"grantType": "refresh_token"})
+	if got := ClassifyRefreshFailure(err); got != RefreshFailureTransient {
+		t.Fatalf("ClassifyRefreshFailure() = %q, want transient: %v", got, err)
+	}
+	if strings.Contains(err.Error(), "must-not-reach-logs") {
+		t.Fatalf("postJSON error leaked response body: %v", err)
+	}
+	if got := httpStatusResponseBody(err); !strings.Contains(got, "must-not-reach-logs") {
+		t.Fatalf("postJSON did not retain bounded response details for internal classification: %q", got)
+	}
+}
+
+func TestCrossPlatformCoverageGetTokenSnapshotOnlyExpiresProfileForNonTransientRefreshFailures(t *testing.T) {
+	oldLoad := oauthLoadToken
+	oldLoadLocked := oauthLoadTokenLocked
+	oldAcquire := oauthAcquireLock
+	oldRefresh := oauthRefreshToken
+	oldMark := oauthMarkProfile
+	oldEdition := edition.Get()
+	t.Cleanup(func() {
+		oauthLoadToken = oldLoad
+		oauthLoadTokenLocked = oldLoadLocked
+		oauthAcquireLock = oldAcquire
+		oauthRefreshToken = oldRefresh
+		oauthMarkProfile = oldMark
+		edition.Override(oldEdition)
+	})
+	edition.Override(&edition.Hooks{})
+
+	expired := &TokenData{
+		AccessToken:  "expired-access",
+		ExpiresAt:    time.Now().Add(-time.Hour),
+		RefreshToken: "refresh",
+		RefreshExpAt: time.Now().Add(time.Hour),
+		CorpID:       "corp",
+		UserID:       "user",
+	}
+	oauthLoadToken = func(string) (*TokenData, error) { return expired, nil }
+	oauthLoadTokenLocked = func(string, string) (*TokenData, error) { return expired, nil }
+	oauthAcquireLock = func(context.Context, string) (*DualLock, error) { return &DualLock{}, nil }
+
+	markCalls := 0
+	oauthMarkProfile = func(_, _, status string) error {
+		if status != ProfileStatusExpired {
+			t.Fatalf("profile status = %q, want %q", status, ProfileStatusExpired)
+		}
+		markCalls++
+		return nil
+	}
+	provider := NewOAuthProvider(t.TempDir(), nil)
+
+	oauthRefreshToken = func(*OAuthProvider, context.Context, *TokenData) (*TokenData, error) {
+		return nil, &HTTPStatusError{StatusCode: http.StatusServiceUnavailable}
+	}
+	if _, err := provider.GetTokenSnapshot(context.Background()); ClassifyRefreshFailure(err) != RefreshFailureTransient {
+		t.Fatalf("transient refresh error = %v", err)
+	}
+	if markCalls != 0 {
+		t.Fatalf("transient refresh marked profile expired %d times", markCalls)
+	}
+
+	oauthRefreshToken = func(*OAuthProvider, context.Context, *TokenData) (*TokenData, error) {
+		return nil, &HTTPStatusError{StatusCode: http.StatusUnauthorized}
+	}
+	if _, err := provider.GetTokenSnapshot(context.Background()); ClassifyRefreshFailure(err) != RefreshFailureTerminal {
+		t.Fatalf("terminal refresh error = %v", err)
+	}
+	if markCalls != 1 {
+		t.Fatalf("terminal refresh marked profile expired %d times, want 1", markCalls)
+	}
+}

internal/auth/rejected_token_refresh_test.go

@@ -0,0 +1,510 @@
+// 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 auth
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
+)
+
+type rejectedTokenHookStore struct {
+	mu      sync.Mutex
+	data    TokenData
+	deletes int
+}
+
+func (s *rejectedTokenHookStore) load(string) ([]byte, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return json.Marshal(s.data)
+}
+
+func (s *rejectedTokenHookStore) save(_ string, blob []byte) error {
+	var data TokenData
+	if err := json.Unmarshal(blob, &data); err != nil {
+		return err
+	}
+	s.mu.Lock()
+	s.data = data
+	s.mu.Unlock()
+	return nil
+}
+
+func (s *rejectedTokenHookStore) delete(string) error {
+	s.mu.Lock()
+	s.data = TokenData{}
+	s.deletes++
+	s.mu.Unlock()
+	return nil
+}
+
+func (s *rejectedTokenHookStore) snapshot() (TokenData, int) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.data, s.deletes
+}
+
+func installRejectedTokenHookStore(t *testing.T, data TokenData) *rejectedTokenHookStore {
+	t.Helper()
+	store := &rejectedTokenHookStore{data: data}
+	previousHooks := edition.Get()
+	edition.Override(&edition.Hooks{
+		LoadToken:   store.load,
+		SaveToken:   store.save,
+		DeleteToken: store.delete,
+	})
+	t.Cleanup(func() { edition.Override(previousHooks) })
+	return store
+}
+
+func installOAuthRefreshStub(t *testing.T, fn func(*OAuthProvider, context.Context, *TokenData) (*TokenData, error)) {
+	t.Helper()
+	resetRejectedTokenRefreshCoordinator(t)
+	previous := oauthRefreshToken
+	oauthRefreshToken = fn
+	t.Cleanup(func() { oauthRefreshToken = previous })
+}
+
+func resetRejectedTokenRefreshCoordinator(t *testing.T) {
+	t.Helper()
+	reset := func() {
+		rejectedTokenRefreshCoordinator.Lock()
+		rejectedTokenRefreshCoordinator.inFlight = make(map[rejectedTokenRefreshKey]*rejectedTokenRefreshCall)
+		rejectedTokenRefreshCoordinator.failures = make(map[rejectedTokenRefreshKey]rejectedTokenRefreshFailure)
+		rejectedTokenRefreshCoordinator.now = time.Now
+		rejectedTokenRefreshCoordinator.Unlock()
+	}
+	reset()
+	t.Cleanup(reset)
+}
+
+func waitForRejectedTokenRefreshParticipants(t *testing.T, key rejectedTokenRefreshKey, want int) {
+	t.Helper()
+	deadline := time.Now().Add(2 * time.Second)
+	for {
+		rejectedTokenRefreshCoordinator.Lock()
+		call := rejectedTokenRefreshCoordinator.inFlight[key]
+		got := 0
+		if call != nil {
+			got = call.participants
+		}
+		rejectedTokenRefreshCoordinator.Unlock()
+		if got >= want {
+			return
+		}
+		if time.Now().After(deadline) {
+			t.Fatalf("refresh participants = %d, want %d", got, want)
+		}
+		time.Sleep(time.Millisecond)
+	}
+}
+
+func installProfilesAcquireProbe(t *testing.T) <-chan struct{} {
+	t.Helper()
+	previous := profilesAcquireDualLock
+	attempted := make(chan struct{}, 1)
+	profilesAcquireDualLock = func(ctx context.Context, configDir string) (*DualLock, error) {
+		attempted <- struct{}{}
+		return previous(ctx, configDir)
+	}
+	t.Cleanup(func() { profilesAcquireDualLock = previous })
+	return attempted
+}
+
+func waitForProfilesAcquire(t *testing.T, attempted <-chan struct{}) {
+	t.Helper()
+	select {
+	case <-attempted:
+	case <-time.After(2 * time.Second):
+		t.Fatal("public opaque token mutation did not enter the Core dual lock")
+	}
+}
+
+func validRejectedTokenData(accessToken string) TokenData {
+	return TokenData{
+		AccessToken:  accessToken,
+		RefreshToken: "refresh-token",
+		ExpiresAt:    time.Now().Add(time.Hour),
+		RefreshExpAt: time.Now().Add(24 * time.Hour),
+		Source:       "mcp",
+		ClientID:     "client-id",
+	}
+}
+
+func TestCrossPlatformCoverageForceRefreshRejectedTokenConcurrentCallersExchangeOnce(t *testing.T) {
+	store := installRejectedTokenHookStore(t, validRejectedTokenData("old-access"))
+	var refreshCalls atomic.Int32
+	started := make(chan struct{})
+	release := make(chan struct{})
+	installOAuthRefreshStub(t, func(p *OAuthProvider, _ context.Context, data *TokenData) (*TokenData, error) {
+		if refreshCalls.Add(1) == 1 {
+			close(started)
+		}
+		<-release
+		updated := *data
+		updated.AccessToken = "new-access"
+		updated.ExpiresAt = time.Now().Add(time.Hour)
+		if err := saveTokenDataLocked(p.configDir, &updated); err != nil {
+			return nil, err
+		}
+		return &updated, nil
+	})
+
+	provider := NewOAuthProvider(t.TempDir(), nil)
+	const workers = 8
+	results := make(chan string, workers)
+	errs := make(chan error, workers)
+	var wg sync.WaitGroup
+	wg.Add(workers)
+	for range workers {
+		go func() {
+			defer wg.Done()
+			token, err := provider.ForceRefreshRejectedToken(context.Background(), "old-access")
+			results <- token
+			errs <- err
+		}()
+	}
+	<-started
+	close(release)
+	wg.Wait()
+	close(results)
+	close(errs)
+
+	for err := range errs {
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+	for token := range results {
+		if token != "new-access" {
+			t.Fatalf("token = %q, want new-access", token)
+		}
+	}
+	if got := refreshCalls.Load(); got != 1 {
+		t.Fatalf("refresh calls = %d, want 1", got)
+	}
+	stored, deletes := store.snapshot()
+	if stored.AccessToken != "new-access" || deletes != 0 {
+		t.Fatalf("stored token = %q, deletes = %d", stored.AccessToken, deletes)
+	}
+}
+
+func TestCrossPlatformCoverageForceRefreshRejectedTokenFailurePreservesCredential(t *testing.T) {
+	store := installRejectedTokenHookStore(t, validRejectedTokenData("old-access"))
+	refreshErr := errors.New("temporary refresh failure")
+	installOAuthRefreshStub(t, func(*OAuthProvider, context.Context, *TokenData) (*TokenData, error) {
+		return nil, refreshErr
+	})
+
+	_, err := NewOAuthProvider(t.TempDir(), nil).ForceRefreshRejectedToken(context.Background(), "old-access")
+	if !errors.Is(err, refreshErr) {
+		t.Fatalf("error = %v, want refresh cause", err)
+	}
+	stored, deletes := store.snapshot()
+	if stored.AccessToken != "old-access" || stored.RefreshToken != "refresh-token" || deletes != 0 {
+		t.Fatalf("credential changed after transient failure: %#v, deletes=%d", stored, deletes)
+	}
+}
+
+func TestCrossPlatformCoverageForceRefreshRejectedTokenFailureIsSingleflightAndCooledDown(t *testing.T) {
+	store := installRejectedTokenHookStore(t, validRejectedTokenData("old-access"))
+	previousProfile := RuntimeProfile()
+	SetRuntimeProfile("")
+	t.Cleanup(func() { SetRuntimeProfile(previousProfile) })
+
+	refreshErr := errors.New("temporary refresh failure")
+	var refreshCalls atomic.Int32
+	started := make(chan struct{})
+	release := make(chan struct{})
+	baseNow := time.Now()
+	var nowNanos atomic.Int64
+	nowNanos.Store(baseNow.UnixNano())
+	installOAuthRefreshStub(t, func(p *OAuthProvider, _ context.Context, data *TokenData) (*TokenData, error) {
+		call := refreshCalls.Add(1)
+		if call == 1 {
+			close(started)
+			<-release
+			return nil, refreshErr
+		}
+		updated := *data
+		updated.AccessToken = "recovered-access"
+		updated.ExpiresAt = time.Now().Add(time.Hour)
+		if err := saveTokenDataLocked(p.configDir, &updated); err != nil {
+			return nil, err
+		}
+		return &updated, nil
+	})
+	rejectedTokenRefreshCoordinator.Lock()
+	rejectedTokenRefreshCoordinator.now = func() time.Time {
+		return time.Unix(0, nowNanos.Load())
+	}
+	rejectedTokenRefreshCoordinator.Unlock()
+
+	configDir := t.TempDir()
+	provider := NewOAuthProvider(configDir, nil)
+	const workers = 8
+	start := make(chan struct{})
+	ready := make(chan struct{}, workers)
+	errs := make(chan error, workers)
+	var wg sync.WaitGroup
+	wg.Add(workers)
+	for range workers {
+		go func() {
+			defer wg.Done()
+			ready <- struct{}{}
+			<-start
+			_, err := provider.ForceRefreshRejectedToken(context.Background(), "old-access")
+			errs <- err
+		}()
+	}
+	for range workers {
+		<-ready
+	}
+	close(start)
+	<-started
+	key := newRejectedTokenRefreshKey(configDir, "", "old-access")
+	waitForRejectedTokenRefreshParticipants(t, key, workers)
+	close(release)
+	wg.Wait()
+	close(errs)
+
+	for err := range errs {
+		if !errors.Is(err, refreshErr) {
+			t.Fatalf("shared refresh error = %v, want %v", err, refreshErr)
+		}
+	}
+	if got := refreshCalls.Load(); got != 1 {
+		t.Fatalf("refresh calls after concurrent failure = %d, want 1", got)
+	}
+	stored, deletes := store.snapshot()
+	if stored.AccessToken != "old-access" || stored.RefreshToken != "refresh-token" || deletes != 0 {
+		t.Fatalf("credential changed after shared failure: %#v, deletes=%d", stored, deletes)
+	}
+
+	if _, err := provider.ForceRefreshRejectedToken(context.Background(), "old-access"); !errors.Is(err, refreshErr) {
+		t.Fatalf("cooldown error = %v, want %v", err, refreshErr)
+	}
+	if got := refreshCalls.Load(); got != 1 {
+		t.Fatalf("refresh calls inside cooldown = %d, want 1", got)
+	}
+
+	nowNanos.Store(baseNow.Add(rejectedTokenRefreshFailureCooldown + time.Nanosecond).UnixNano())
+	token, err := provider.ForceRefreshRejectedToken(context.Background(), "old-access")
+	if err != nil || token != "recovered-access" {
+		t.Fatalf("refresh after cooldown = %q, %v", token, err)
+	}
+	if got := refreshCalls.Load(); got != 2 {
+		t.Fatalf("refresh calls after cooldown = %d, want 2", got)
+	}
+	stored, deletes = store.snapshot()
+	if stored.AccessToken != "recovered-access" || deletes != 0 {
+		t.Fatalf("stored token after cooldown recovery = %q, deletes=%d", stored.AccessToken, deletes)
+	}
+}
+
+func TestCrossPlatformCoverageForceRefreshRejectedTokenChangedDuringCooldownUsesNewToken(t *testing.T) {
+	store := installRejectedTokenHookStore(t, validRejectedTokenData("old-access"))
+	previousProfile := RuntimeProfile()
+	SetRuntimeProfile("")
+	t.Cleanup(func() { SetRuntimeProfile(previousProfile) })
+
+	refreshErr := errors.New("temporary refresh failure")
+	var refreshCalls atomic.Int32
+	baseNow := time.Now()
+	installOAuthRefreshStub(t, func(*OAuthProvider, context.Context, *TokenData) (*TokenData, error) {
+		refreshCalls.Add(1)
+		return nil, refreshErr
+	})
+	rejectedTokenRefreshCoordinator.Lock()
+	rejectedTokenRefreshCoordinator.now = func() time.Time { return baseNow }
+	rejectedTokenRefreshCoordinator.Unlock()
+
+	configDir := t.TempDir()
+	provider := NewOAuthProvider(configDir, nil)
+	if _, err := provider.ForceRefreshRejectedToken(context.Background(), "old-access"); !errors.Is(err, refreshErr) {
+		t.Fatalf("initial refresh error = %v, want %v", err, refreshErr)
+	}
+	if got := refreshCalls.Load(); got != 1 {
+		t.Fatalf("initial refresh calls = %d, want 1", got)
+	}
+
+	store.mu.Lock()
+	store.data = validRejectedTokenData("externally-refreshed")
+	store.mu.Unlock()
+	token, err := provider.ForceRefreshRejectedToken(context.Background(), "old-access")
+	if err != nil || token != "externally-refreshed" {
+		t.Fatalf("refresh after external publication = %q, %v", token, err)
+	}
+	if got := refreshCalls.Load(); got != 1 {
+		t.Fatalf("external publication triggered another exchange: calls=%d", got)
+	}
+	key := newRejectedTokenRefreshKey(configDir, "", "old-access")
+	rejectedTokenRefreshCoordinator.Lock()
+	_, failurePresent := rejectedTokenRefreshCoordinator.failures[key]
+	rejectedTokenRefreshCoordinator.Unlock()
+	if failurePresent {
+		t.Fatal("old-token failure cache was not cleared after external publication")
+	}
+}
+
+func TestCrossPlatformCoverageOpaquePublisherWaitsForRejectedTokenRefresh(t *testing.T) {
+	store := installRejectedTokenHookStore(t, validRejectedTokenData("old-access"))
+	started := make(chan struct{})
+	release := make(chan struct{})
+	var releaseOnce sync.Once
+	t.Cleanup(func() { releaseOnce.Do(func() { close(release) }) })
+	installOAuthRefreshStub(t, func(p *OAuthProvider, _ context.Context, data *TokenData) (*TokenData, error) {
+		close(started)
+		<-release
+		updated := *data
+		updated.AccessToken = "refreshed-from-old"
+		updated.ExpiresAt = time.Now().Add(time.Hour)
+		if err := saveTokenDataLocked(p.configDir, &updated); err != nil {
+			return nil, err
+		}
+		return &updated, nil
+	})
+
+	configDir := t.TempDir()
+	provider := NewOAuthProvider(configDir, nil)
+	refreshResult := make(chan struct {
+		token string
+		err   error
+	}, 1)
+	go func() {
+		token, err := provider.ForceRefreshRejectedToken(context.Background(), "old-access")
+		refreshResult <- struct {
+			token string
+			err   error
+		}{token: token, err: err}
+	}()
+	<-started
+
+	acquireAttempted := installProfilesAcquireProbe(t)
+	publishResult := make(chan error, 1)
+	go func() {
+		publishResult <- SaveTokenData(configDir, ptrTokenData(validRejectedTokenData("login-published")))
+	}()
+	waitForProfilesAcquire(t, acquireAttempted)
+	releaseOnce.Do(func() { close(release) })
+
+	refresh := <-refreshResult
+	if refresh.err != nil || refresh.token != "refreshed-from-old" {
+		t.Fatalf("refresh result = %q, %v", refresh.token, refresh.err)
+	}
+	if err := <-publishResult; err != nil {
+		t.Fatalf("publish token: %v", err)
+	}
+	stored, deletes := store.snapshot()
+	if stored.AccessToken != "login-published" || deletes != 0 {
+		t.Fatalf("older refresh overwrote login publication: token=%q deletes=%d", stored.AccessToken, deletes)
+	}
+}
+
+func TestCrossPlatformCoverageOpaqueLogoutWaitsForRejectedTokenRefresh(t *testing.T) {
+	for _, tc := range []struct {
+		name   string
+		logout func(string) error
+	}{
+		{name: "current profile", logout: func(configDir string) error {
+			return DeleteTokenDataForProfile(configDir, "")
+		}},
+		{name: "all profiles", logout: DeleteAllTokenData},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			store := installRejectedTokenHookStore(t, validRejectedTokenData("old-access"))
+			started := make(chan struct{})
+			release := make(chan struct{})
+			var releaseOnce sync.Once
+			t.Cleanup(func() { releaseOnce.Do(func() { close(release) }) })
+			installOAuthRefreshStub(t, func(p *OAuthProvider, _ context.Context, data *TokenData) (*TokenData, error) {
+				close(started)
+				<-release
+				updated := *data
+				updated.AccessToken = "refreshed-before-logout"
+				updated.ExpiresAt = time.Now().Add(time.Hour)
+				if err := saveTokenDataLocked(p.configDir, &updated); err != nil {
+					return nil, err
+				}
+				return &updated, nil
+			})
+
+			configDir := t.TempDir()
+			provider := NewOAuthProvider(configDir, nil)
+			refreshResult := make(chan error, 1)
+			go func() {
+				_, err := provider.ForceRefreshRejectedToken(context.Background(), "old-access")
+				refreshResult <- err
+			}()
+			<-started
+
+			acquireAttempted := installProfilesAcquireProbe(t)
+			logoutResult := make(chan error, 1)
+			go func() { logoutResult <- tc.logout(configDir) }()
+			waitForProfilesAcquire(t, acquireAttempted)
+			releaseOnce.Do(func() { close(release) })
+
+			if err := <-refreshResult; err != nil {
+				t.Fatalf("refresh: %v", err)
+			}
+			if err := <-logoutResult; err != nil {
+				t.Fatalf("logout: %v", err)
+			}
+			stored, deletes := store.snapshot()
+			if stored.AccessToken != "" || stored.RefreshToken != "" || deletes != 1 {
+				t.Fatalf("refresh resurrected logged-out credential: %#v deletes=%d", stored, deletes)
+			}
+		})
+	}
+}
+
+func ptrTokenData(data TokenData) *TokenData {
+	return &data
+}
+
+func TestCrossPlatformCoverageOAuthLockedRefreshReadsOpaqueEditionStore(t *testing.T) {
+	data := validRejectedTokenData("expired-access")
+	data.ExpiresAt = time.Now().Add(-time.Hour)
+	store := installRejectedTokenHookStore(t, data)
+	var refreshCalls atomic.Int32
+	installOAuthRefreshStub(t, func(p *OAuthProvider, _ context.Context, current *TokenData) (*TokenData, error) {
+		refreshCalls.Add(1)
+		updated := *current
+		updated.AccessToken = "proactively-refreshed"
+		updated.ExpiresAt = time.Now().Add(time.Hour)
+		if err := saveTokenDataLocked(p.configDir, &updated); err != nil {
+			return nil, err
+		}
+		return &updated, nil
+	})
+
+	token, err := NewOAuthProvider(t.TempDir(), nil).GetAccessToken(context.Background())
+	if err != nil || token != "proactively-refreshed" {
+		t.Fatalf("GetAccessToken() = %q, %v", token, err)
+	}
+	if refreshCalls.Load() != 1 {
+		t.Fatalf("refresh calls = %d, want 1", refreshCalls.Load())
+	}
+	stored, deletes := store.snapshot()
+	if stored.AccessToken != "proactively-refreshed" || deletes != 0 {
+		t.Fatalf("stored token = %q, deletes = %d", stored.AccessToken, deletes)
+	}
+}

internal/auth/testmain_test.go

@@ -14,6 +14,7 @@
 package auth
 
 import (
+	"fmt"
 	"os"
 	"testing"
 
@@ -34,7 +35,17 @@ func TestMain(m *testing.M) {
 		_ = os.RemoveAll(tmpDir)
 		panic("set " + keychain.StorageDirEnv + ": " + err.Error())
 	}
+	if err := os.Setenv(keychain.TestNamespaceEnv, tmpDir); err != nil {
+		_ = os.RemoveAll(tmpDir)
+		panic("set " + keychain.TestNamespaceEnv + ": " + err.Error())
+	}
 	code := m.Run()
+	if err := keychain.RemoveAuthTokenEntries(keychain.Service); err != nil {
+		fmt.Fprintf(os.Stderr, "internal/auth keychain cleanup: %v\n", err)
+		if code == 0 {
+			code = 1
+		}
+	}
 	_ = os.RemoveAll(tmpDir)
 	os.Exit(code)
 }

internal/auth/token.go

@@ -127,6 +127,10 @@ const tokenJSONFile = "token.json"
 type TokenMarker struct {
 	UpdatedAt   string `json:"updated_at"`
 	ManualToken bool   `json:"manual_token,omitempty"`
+	// Revision changes on every credential publication. Runtime token caches
+	// use it as a cheap cross-process invalidation signal without reading the
+	// platform keychain on every request.
+	Revision string `json:"revision,omitempty"`
 }
 
 // WriteTokenMarker writes a token.json marker containing only an updated_at
@@ -147,6 +151,7 @@ func writeTokenMarker(configDir string, manual bool) error {
 	marker := TokenMarker{
 		UpdatedAt:   time.Now().Format(time.RFC3339),
 		ManualToken: manual,
+		Revision:    uuid.NewString(),
 	}
 	data, _ := tokenJSONMarshalIndent(marker, "", "  ")
 	if err := tokenMkdirAll(configDir, 0o700); err != nil {
@@ -159,6 +164,27 @@ func writeTokenMarker(configDir string, manual bool) error {
 	return tokenRename(tmp, filepath.Join(configDir, tokenJSONFile))
 }
 
+// ReadTokenMarkerRevision returns the current credential publication revision.
+// Existing markers without a revision remain readable, but callers must avoid
+// caching them because they cannot prove that the credential is unchanged.
+func ReadTokenMarkerRevision(configDir string) (revision string, present bool, err error) {
+	data, err := tokenReadFile(filepath.Join(configDir, tokenJSONFile))
+	if err != nil {
+		if os.IsNotExist(err) {
+			return "", false, nil
+		}
+		return "", false, fmt.Errorf("read token marker: %w", err)
+	}
+	var marker TokenMarker
+	if err := json.Unmarshal(data, &marker); err != nil {
+		// The marker is only a cache-coherency hint. A malformed historical or
+		// externally modified marker must disable caching, not make an otherwise
+		// valid credential unusable.
+		return "", true, nil
+	}
+	return strings.TrimSpace(marker.Revision), true, nil
+}
+
 func manualTokenMarkerActive(configDir string) (bool, error) {
 	data, err := tokenReadFile(filepath.Join(configDir, tokenJSONFile))
 	if err != nil {
@@ -184,13 +210,10 @@ func DeleteTokenMarker(configDir string) error {
 	return nil
 }
 
-// SaveTokenData persists TokenData. When an edition hook (SaveToken) is
-// registered, it delegates entirely to the hook; otherwise it falls back
-// to the default keychain-based storage.
+// SaveTokenData persists TokenData under the auth dual lock. When an edition
+// hook (SaveToken) is registered, the locked write delegates to that hook;
+// otherwise it falls back to the default keychain-based storage.
 func SaveTokenData(configDir string, data *TokenData) error {
-	if h := edition.Get(); h.SaveToken != nil {
-		return saveTokenViaHook(h, configDir, data)
-	}
 	return withProfilesLock(configDir, func() error {
 		return saveTokenDataLocked(configDir, data)
 	})
@@ -464,9 +487,8 @@ func tokenLoadProfileIdentity(profile Profile) (*TokenData, error) {
 	return orgData, nil
 }
 
-// DeleteTokenData removes token data. When an edition hook (DeleteToken) is
-// registered, it delegates entirely to the hook; otherwise it falls back
-// to keychain + legacy cleanup.
+// DeleteTokenData removes token data. Edition hooks and the default keychain
+// path are both serialized with refresh through the auth dual lock.
 func DeleteTokenData(configDir string) error {
 	return DeleteTokenDataForProfile(configDir, RuntimeProfile())
 }
@@ -478,7 +500,9 @@ func DeleteTokenDataForProfile(configDir, profile string) error {
 		if strings.TrimSpace(profile) != "" {
 			return fmt.Errorf("profile selection is not supported by the current auth backend")
 		}
-		return h.DeleteToken(configDir)
+		return withProfilesLock(configDir, func() error {
+			return h.DeleteToken(configDir)
+		})
 	}
 	return withProfilesLock(configDir, func() error {
 		return deleteTokenDataForProfileLocked(configDir, profile)
@@ -913,7 +937,9 @@ func restoreTokenMarker(configDir string, marker tokenMarkerSnapshot) error {
 // DeleteAllTokenData removes all profile-scoped and legacy token data.
 func DeleteAllTokenData(configDir string) error {
 	if h := edition.Get(); h.DeleteToken != nil {
-		return h.DeleteToken(configDir)
+		return withProfilesLock(configDir, func() error {
+			return h.DeleteToken(configDir)
+		})
 	}
 	return withProfilesLock(configDir, func() error {
 		var firstErr error

internal/cli/AGENTS.md

@@ -0,0 +1,42 @@
+# Schema Runtime and Publication Agent Guide
+
+This file applies to `internal/cli/`. Read
+[`docs/schema-contributor-guide.md`](../../docs/schema-contributor-guide.md)
+before editing.
+
+## Owning inputs
+
+| Change | Edit |
+|---|---|
+| Canonical identity, primary path, aliases, navigation | `schema_command_registry.json` |
+| Parameter/property mapping | `schema_parameter_bindings.json` or reviewed metadata overlay |
+| Safety, interface, runtime gate | `schema_hints/metadata/<product>.json` |
+| Agent selection and examples | `schema_hints/selection/<product>.json` |
+| Exact reviewed omission | `schema_command_exclusions.json` |
+| Runtime query/projection behavior | Go implementation and focused tests in this package |
+
+The executable Cobra tree outside this package owns whether a command exists
+and which flags it accepts. Do not create a command or flag in Schema inputs.
+
+## Generated boundary
+
+- `schema_catalog.json` and `schema_agent_metadata/` are generated outputs.
+- Never hand-edit, merge from, or use a previous generated output as an input.
+- Change the owning reviewed input or generator, run `make generate-schema`,
+  and inspect authored and generated diffs separately.
+- A broad unrelated generated diff is a failure signal, not acceptable churn.
+
+## Self-check
+
+- Every public runnable Cobra leaf is bound or has one exact reviewed
+  exclusion; every delivered tool binds back to a runnable leaf.
+- Registry identity and native annotations agree; aliases do not mutate the
+  resolved contract.
+- Parameters reference real flags and retain Cobra-required floors.
+- Selection examples use executable paths/flags and never include `--yes`.
+- Runtime confirmation gates and published safety metadata agree.
+- Full, compact, summary, alias, and Catalog projections derive from the same
+  typed `ToolSpec` and preserve provenance winners.
+
+Run the focused package/generator tests and the complete command list in
+`docs/schema-contributor-guide.md`, beginning with `make generate-schema`.

internal/cli/canonical.go

@@ -304,3 +304,14 @@ func splitSchemaPathTokens(raw string) []string {
 	}
 	return out
 }
+
+// normalizeSchemaQueryCLIPath accepts the historical query spellings while
+// keeping authored Registry CLI paths strict and space-separated. Canonical
+// identity lookup still runs before this compatibility normalization.
+func normalizeSchemaQueryCLIPath(path string) string {
+	parts := splitSchemaPathTokens(strings.TrimSpace(path))
+	if len(parts) > 0 && parts[0] == "dws" {
+		parts = parts[1:]
+	}
+	return strings.Join(parts, " ")
+}

internal/cli/schema_agent_metadata/event.json

@@ -2,7 +2,7 @@
   "product_id": "event",
   "tools": {
     "event consume": {
-      "agent_summary": "订阅并持续消费指定个人事件，输出 NDJSON 事件流",
+      "agent_summary": "订阅并持续消费指定个人事件；Agent 使用 --flatten 输出顶层业务 NDJSON",
       "agent_summary_source": "dws-agent-selection/event",
       "availability": "available",
       "avoid_when": [
@@ -13,19 +13,19 @@
       "effect": "write",
       "effect_source": "agent-hint",
       "examples": [
-        "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --max-events 1 --format ndjson",
-        "dws event consume user_im_message_reaction_group --group cid-example --max-events 1 --format ndjson"
+        "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --flatten --max-events 1 --format ndjson",
+        "dws event consume user_im_message_reaction_group --group cid-example --flatten --max-events 1 --format ndjson"
       ],
       "field_provenance": {
         "agent_summary": {
-          "value": "订阅并持续消费指定个人事件，输出 NDJSON 事件流",
+          "value": "订阅并持续消费指定个人事件；Agent 使用 --flatten 输出顶层业务 NDJSON",
           "source": "internal/cli/schema_hints/selection/event.json",
           "precedence": "reviewed_explicit",
           "resolution": "highest_precedence",
           "review_reason": "人工依据实时 dws schema（或 Skill/Cobra/pinned MCP 对照）决策化选型文案与门禁；不改写命令身份与参数契约；示例不含 --yes。",
           "candidates": [
             {
-              "value": "订阅并持续消费指定个人事件，输出 NDJSON 事件流",
+              "value": "订阅并持续消费指定个人事件；Agent 使用 --flatten 输出顶层业务 NDJSON",
               "source": "internal/cli/schema_hints/selection/event.json",
               "precedence": "reviewed_explicit",
               "selected": true,
@@ -105,8 +105,8 @@
         },
         "examples": {
           "value": [
-            "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --max-events 1 --format ndjson",
-            "dws event consume user_im_message_reaction_group --group cid-example --max-events 1 --format ndjson"
+            "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --flatten --max-events 1 --format ndjson",
+            "dws event consume user_im_message_reaction_group --group cid-example --flatten --max-events 1 --format ndjson"
           ],
           "source": "internal/cli/schema_hints/selection/event.json",
           "precedence": "reviewed_explicit",
@@ -115,8 +115,8 @@
           "candidates": [
             {
               "value": [
-                "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --max-events 1 --format ndjson",
-                "dws event consume user_im_message_reaction_group --group cid-example --max-events 1 --format ndjson"
+                "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --flatten --max-events 1 --format ndjson",
+                "dws event consume user_im_message_reaction_group --group cid-example --flatten --max-events 1 --format ndjson"
               ],
               "source": "internal/cli/schema_hints/selection/event.json",
               "precedence": "reviewed_explicit",
@@ -538,7 +538,7 @@
       ]
     },
     "event schema": {
-      "agent_summary": "查询指定个人事件码的 payload 字段结构",
+      "agent_summary": "查询指定个人事件码的输出字段结构；Agent 应查询 --flatten 模式",
       "agent_summary_source": "dws-agent-selection/event",
       "availability": "available",
       "avoid_when": [
@@ -549,18 +549,18 @@
       "effect": "read",
       "effect_source": "agent-hint",
       "examples": [
-        "dws event schema user_im_message_receive_at --format json"
+        "dws event schema user_im_message_receive_at --flatten --format json"
       ],
       "field_provenance": {
         "agent_summary": {
-          "value": "查询指定个人事件码的 payload 字段结构",
+          "value": "查询指定个人事件码的输出字段结构；Agent 应查询 --flatten 模式",
           "source": "internal/cli/schema_hints/selection/event.json",
           "precedence": "reviewed_explicit",
           "resolution": "highest_precedence",
           "review_reason": "人工依据实时 dws schema（或 Skill/Cobra/pinned MCP 对照）决策化选型文案与门禁；不改写命令身份与参数契约；示例不含 --yes。",
           "candidates": [
             {
-              "value": "查询指定个人事件码的 payload 字段结构",
+              "value": "查询指定个人事件码的输出字段结构；Agent 应查询 --flatten 模式",
               "source": "internal/cli/schema_hints/selection/event.json",
               "precedence": "reviewed_explicit",
               "selected": true,
@@ -640,7 +640,7 @@
         },
         "examples": {
           "value": [
-            "dws event schema user_im_message_receive_at --format json"
+            "dws event schema user_im_message_receive_at --flatten --format json"
           ],
           "source": "internal/cli/schema_hints/selection/event.json",
           "precedence": "reviewed_explicit",
@@ -649,7 +649,7 @@
           "candidates": [
             {
               "value": [
-                "dws event schema user_im_message_receive_at --format json"
+                "dws event schema user_im_message_receive_at --flatten --format json"
               ],
               "source": "internal/cli/schema_hints/selection/event.json",
               "precedence": "reviewed_explicit",

internal/cli/schema_agent_metadata/index.json

@@ -1,6 +1,6 @@
 {
   "version": 1,
-  "source_hash": "sha256:de587cba5051dd4c2715353012d8d9f2a7c5208a0c6dee4ea703cfc97b7351f0",
+  "source_hash": "sha256:5df496973c41b3b4f7ae8c856ff0e9e99bcabd4455419b7d41bb23c44df6f1af",
   "surface_hash": "sha256:7ef588f38052f0104e027c8daff5fefd68698781f2c87715461183d4058288ed",
   "coverage": {
     "surface_products": 22,

internal/cli/schema_agent_metadata_audit.json

@@ -1,6 +1,6 @@
 {
   "version": 1,
-  "source_hash": "sha256:de587cba5051dd4c2715353012d8d9f2a7c5208a0c6dee4ea703cfc97b7351f0",
+  "source_hash": "sha256:5df496973c41b3b4f7ae8c856ff0e9e99bcabd4455419b7d41bb23c44df6f1af",
   "surface_hash": "sha256:7ef588f38052f0104e027c8daff5fefd68698781f2c87715461183d4058288ed",
   "source_files": 150,
   "hint_files": 46,

internal/cli/schema_catalog.go

@@ -290,7 +290,7 @@ func schemaPayloadFromLoadedCatalog(loaded loadedSchemaCatalog, args []string) (
 		return payload, nil
 	}
 	raw := strings.TrimSpace(args[0])
-	if tool, ok := loaded.Index.Resolve(raw); ok {
+	if tool, ok := loaded.Index.ResolveQuery(raw); ok {
 		return schemaToolForResolvedPath(tool, raw).ToPayload()
 	}
 	tokens := splitSchemaPathTokens(raw)

internal/cli/schema_catalog.json

@@ -1,12 +1,12 @@
 {
   "version": 1,
-  "source_hash": "sha256:522d4b43cf13f07430a407319a772cd11e9b1f268f8d98b4d68bdb385e4e585b",
+  "source_hash": "sha256:14398788381c822f208e8cae059d98cd60da6625023869c74015735650d67e6f",
   "surface_hash": "sha256:7ef588f38052f0104e027c8daff5fefd68698781f2c87715461183d4058288ed",
   "catalog": {
     "agent_metadata": {
       "products_with_metadata": 22,
       "source": "embedded-skill-metadata",
-      "source_hash": "sha256:de587cba5051dd4c2715353012d8d9f2a7c5208a0c6dee4ea703cfc97b7351f0",
+      "source_hash": "sha256:5df496973c41b3b4f7ae8c856ff0e9e99bcabd4455419b7d41bb23c44df6f1af",
       "surface_hash": "sha256:7ef588f38052f0104e027c8daff5fefd68698781f2c87715461183d4058288ed",
       "surface_products": 22,
       "surface_tools": 572,
@@ -12178,7 +12178,7 @@
         "tools": [
           {
             "agent_metadata_source": "embedded-skill-metadata",
-            "agent_summary": "订阅并持续消费指定个人事件，输出 NDJSON 事件流",
+            "agent_summary": "订阅并持续消费指定个人事件；Agent 使用 --flatten 输出顶层业务 NDJSON",
             "agent_summary_source": "dws-agent-selection/event",
             "availability": "available",
             "avoid_when": [
@@ -12189,7 +12189,7 @@
             "cli_name": "consume",
             "cli_path": "event consume",
             "confirmation": "not_required",
-            "description": "订阅 DingTalk 个人事件并将每条事件以 NDJSON 输出到 stdout。\n\n输出格式（事件流默认 ndjson；显式 -f json/pretty/raw 可覆盖；-f table/csv 对\n事件流无意义会 fallback 到 ndjson）：\n  ndjson  (默认)  一行一对象，适合 jq / 管道处理\n  json            每事件多行美化 JSON（必须配 --max-events 或 --duration）\n  pretty          同 json，未来加颜色\n  raw             仅 SDK 原始 payload，无外层封装\n  compact         扁平化 + 解析嵌套 + 抽取语义字段（Agent 友好）\n\n默认使用当前 OAuth 登录态自动创建/复用个人订阅并建立个人长连接；非默认组织加\n--profile。连上后 stderr 打就绪行 [event] ready，等它出现再读 stdout；停机用\nSIGTERM、关 stdin，或先用 dws event stop \u003csubscribe_id\u003e --dry-run 预览、确认后加\n--yes，绝不要 kill -9。\n--event-types/--filter 只影响本地 bus → consume 这一段投递；普通个人事件消费\n通常不需要设置。",
+            "description": "订阅 DingTalk 个人事件并将每条事件以 NDJSON 输出到 stdout。\n\n输出格式（事件流默认 ndjson；显式 -f json/pretty/raw 可覆盖；-f table/csv 对\n事件流无意义会 fallback 到 ndjson）：\n  ndjson  (默认)  一行一对象，适合 jq / 管道处理\n  json            每事件多行美化 JSON（必须配 --max-events 或 --duration）\n  pretty          同 json，未来加颜色\n  raw             仅 SDK 原始 payload，无外层封装\n  compact         单行紧凑 JSON；不传 --flatten 时沿用原 compact processor\n\n数据结构：\n  ndjson/json/pretty 默认保持 transport envelope（type/event_type/data/headers）\n  --flatten          结构化格式输出稳定的顶层业务字段，适合 Agent / 脚本直接消费\n\n默认使用当前 OAuth 登录态自动创建/复用个人订阅并建立个人长连接；非默认组织加\n--profile。连上后 stderr 打就绪行 [event] ready，等它出现再读 stdout；停机用\nSIGTERM、关 stdin，或先用 dws event stop \u003csubscribe_id\u003e --dry-run 预览、确认后加\n--yes，绝不要 kill -9。\n--event-types/--filter 只影响本地 bus → consume 这一段投递；普通个人事件消费\n通常不需要设置。",
             "effect": "write",
             "idempotency": "non_idempotent",
             "interface_mode": "composite",
@@ -12234,7 +12234,7 @@
           },
           {
             "agent_metadata_source": "embedded-skill-metadata",
-            "agent_summary": "查询指定个人事件码的 payload 字段结构",
+            "agent_summary": "查询指定个人事件码的输出字段结构；Agent 应查询 --flatten 模式",
             "agent_summary_source": "dws-agent-selection/event",
             "availability": "available",
             "avoid_when": [
@@ -290130,7 +290130,7 @@
         "skills/mono/SKILL.md",
         "skills/mono/references/products/event.md"
       ],
-      "agent_summary": "订阅并持续消费指定个人事件，输出 NDJSON 事件流",
+      "agent_summary": "订阅并持续消费指定个人事件；Agent 使用 --flatten 输出顶层业务 NDJSON",
       "agent_summary_source": "dws-agent-selection/event",
       "availability": "available",
       "avoid_when": [
@@ -290149,13 +290149,13 @@
           ]
         ]
       },
-      "description": "订阅 DingTalk 个人事件并将每条事件以 NDJSON 输出到 stdout。\n\n输出格式（事件流默认 ndjson；显式 -f json/pretty/raw 可覆盖；-f table/csv 对\n事件流无意义会 fallback 到 ndjson）：\n  ndjson  (默认)  一行一对象，适合 jq / 管道处理\n  json            每事件多行美化 JSON（必须配 --max-events 或 --duration）\n  pretty          同 json，未来加颜色\n  raw             仅 SDK 原始 payload，无外层封装\n  compact         扁平化 + 解析嵌套 + 抽取语义字段（Agent 友好）\n\n默认使用当前 OAuth 登录态自动创建/复用个人订阅并建立个人长连接；非默认组织加\n--profile。连上后 stderr 打就绪行 [event] ready，等它出现再读 stdout；停机用\nSIGTERM、关 stdin，或先用 dws event stop \u003csubscribe_id\u003e --dry-run 预览、确认后加\n--yes，绝不要 kill -9。\n--event-types/--filter 只影响本地 bus → consume 这一段投递；普通个人事件消费\n通常不需要设置。",
+      "description": "订阅 DingTalk 个人事件并将每条事件以 NDJSON 输出到 stdout。\n\n输出格式（事件流默认 ndjson；显式 -f json/pretty/raw 可覆盖；-f table/csv 对\n事件流无意义会 fallback 到 ndjson）：\n  ndjson  (默认)  一行一对象，适合 jq / 管道处理\n  json            每事件多行美化 JSON（必须配 --max-events 或 --duration）\n  pretty          同 json，未来加颜色\n  raw             仅 SDK 原始 payload，无外层封装\n  compact         单行紧凑 JSON；不传 --flatten 时沿用原 compact processor\n\n数据结构：\n  ndjson/json/pretty 默认保持 transport envelope（type/event_type/data/headers）\n  --flatten          结构化格式输出稳定的顶层业务字段，适合 Agent / 脚本直接消费\n\n默认使用当前 OAuth 登录态自动创建/复用个人订阅并建立个人长连接；非默认组织加\n--profile。连上后 stderr 打就绪行 [event] ready，等它出现再读 stdout；停机用\nSIGTERM、关 stdin，或先用 dws event stop \u003csubscribe_id\u003e --dry-run 预览、确认后加\n--yes，绝不要 kill -9。\n--event-types/--filter 只影响本地 bus → consume 这一段投递；普通个人事件消费\n通常不需要设置。",
       "display": "事件订阅 (DingTalk Stream 长连接)",
       "effect": "write",
       "effect_source": "agent-hint",
       "examples": [
-        "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --max-events 1 --format ndjson",
-        "dws event consume user_im_message_reaction_group --group cid-example --max-events 1 --format ndjson"
+        "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --flatten --max-events 1 --format ndjson",
+        "dws event consume user_im_message_reaction_group --group cid-example --flatten --max-events 1 --format ndjson"
       ],
       "field_provenance": {
         "agent_summary": {
@@ -290165,14 +290165,14 @@
               "review_reason": "人工依据实时 dws schema（或 Skill/Cobra/pinned MCP 对照）决策化选型文案与门禁；不改写命令身份与参数契约；示例不含 --yes。",
               "selected": true,
               "source": "internal/cli/schema_hints/selection/event.json",
-              "value": "订阅并持续消费指定个人事件，输出 NDJSON 事件流"
+              "value": "订阅并持续消费指定个人事件；Agent 使用 --flatten 输出顶层业务 NDJSON"
             }
           ],
           "precedence": "reviewed_explicit",
           "resolution": "highest_precedence",
           "review_reason": "人工依据实时 dws schema（或 Skill/Cobra/pinned MCP 对照）决策化选型文案与门禁；不改写命令身份与参数契约；示例不含 --yes。",
           "source": "internal/cli/schema_hints/selection/event.json",
-          "value": "订阅并持续消费指定个人事件，输出 NDJSON 事件流"
+          "value": "订阅并持续消费指定个人事件；Agent 使用 --flatten 输出顶层业务 NDJSON"
         },
         "availability": {
           "candidates": [
@@ -290250,13 +290250,13 @@
               "precedence": "cobra_help",
               "selected": true,
               "source": "cobra_help",
-              "value": "订阅 DingTalk 个人事件并将每条事件以 NDJSON 输出到 stdout。\n\n输出格式（事件流默认 ndjson；显式 -f json/pretty/raw 可覆盖；-f table/csv 对\n事件流无意义会 fallback 到 ndjson）：\n  ndjson  (默认)  一行一对象，适合 jq / 管道处理\n  json            每事件多行美化 JSON（必须配 --max-events 或 --duration）\n  pretty          同 json，未来加颜色\n  raw             仅 SDK 原始 payload，无外层封装\n  compact         扁平化 + 解析嵌套 + 抽取语义字段（Agent 友好）\n\n默认使用当前 OAuth 登录态自动创建/复用个人订阅并建立个人长连接；非默认组织加\n--profile。连上后 stderr 打就绪行 [event] ready，等它出现再读 stdout；停机用\nSIGTERM、关 stdin，或先用 dws event stop \u003csubscribe_id\u003e --dry-run 预览、确认后加\n--yes，绝不要 kill -9。\n--event-types/--filter 只影响本地 bus → consume 这一段投递；普通个人事件消费\n通常不需要设置。"
+              "value": "订阅 DingTalk 个人事件并将每条事件以 NDJSON 输出到 stdout。\n\n输出格式（事件流默认 ndjson；显式 -f json/pretty/raw 可覆盖；-f table/csv 对\n事件流无意义会 fallback 到 ndjson）：\n  ndjson  (默认)  一行一对象，适合 jq / 管道处理\n  json            每事件多行美化 JSON（必须配 --max-events 或 --duration）\n  pretty          同 json，未来加颜色\n  raw             仅 SDK 原始 payload，无外层封装\n  compact         单行紧凑 JSON；不传 --flatten 时沿用原 compact processor\n\n数据结构：\n  ndjson/json/pretty 默认保持 transport envelope（type/event_type/data/headers）\n  --flatten          结构化格式输出稳定的顶层业务字段，适合 Agent / 脚本直接消费\n\n默认使用当前 OAuth 登录态自动创建/复用个人订阅并建立个人长连接；非默认组织加\n--profile。连上后 stderr 打就绪行 [event] ready，等它出现再读 stdout；停机用\nSIGTERM、关 stdin，或先用 dws event stop \u003csubscribe_id\u003e --dry-run 预览、确认后加\n--yes，绝不要 kill -9。\n--event-types/--filter 只影响本地 bus → consume 这一段投递；普通个人事件消费\n通常不需要设置。"
             }
           ],
           "precedence": "cobra_help",
           "resolution": "highest_precedence",
           "source": "cobra_help",
-          "value": "订阅 DingTalk 个人事件并将每条事件以 NDJSON 输出到 stdout。\n\n输出格式（事件流默认 ndjson；显式 -f json/pretty/raw 可覆盖；-f table/csv 对\n事件流无意义会 fallback 到 ndjson）：\n  ndjson  (默认)  一行一对象，适合 jq / 管道处理\n  json            每事件多行美化 JSON（必须配 --max-events 或 --duration）\n  pretty          同 json，未来加颜色\n  raw             仅 SDK 原始 payload，无外层封装\n  compact         扁平化 + 解析嵌套 + 抽取语义字段（Agent 友好）\n\n默认使用当前 OAuth 登录态自动创建/复用个人订阅并建立个人长连接；非默认组织加\n--profile。连上后 stderr 打就绪行 [event] ready，等它出现再读 stdout；停机用\nSIGTERM、关 stdin，或先用 dws event stop \u003csubscribe_id\u003e --dry-run 预览、确认后加\n--yes，绝不要 kill -9。\n--event-types/--filter 只影响本地 bus → consume 这一段投递；普通个人事件消费\n通常不需要设置。"
+          "value": "订阅 DingTalk 个人事件并将每条事件以 NDJSON 输出到 stdout。\n\n输出格式（事件流默认 ndjson；显式 -f json/pretty/raw 可覆盖；-f table/csv 对\n事件流无意义会 fallback 到 ndjson）：\n  ndjson  (默认)  一行一对象，适合 jq / 管道处理\n  json            每事件多行美化 JSON（必须配 --max-events 或 --duration）\n  pretty          同 json，未来加颜色\n  raw             仅 SDK 原始 payload，无外层封装\n  compact         单行紧凑 JSON；不传 --flatten 时沿用原 compact processor\n\n数据结构：\n  ndjson/json/pretty 默认保持 transport envelope（type/event_type/data/headers）\n  --flatten          结构化格式输出稳定的顶层业务字段，适合 Agent / 脚本直接消费\n\n默认使用当前 OAuth 登录态自动创建/复用个人订阅并建立个人长连接；非默认组织加\n--profile。连上后 stderr 打就绪行 [event] ready，等它出现再读 stdout；停机用\nSIGTERM、关 stdin，或先用 dws event stop \u003csubscribe_id\u003e --dry-run 预览、确认后加\n--yes，绝不要 kill -9。\n--event-types/--filter 只影响本地 bus → consume 这一段投递；普通个人事件消费\n通常不需要设置。"
         },
         "effect": {
           "candidates": [
@@ -290282,8 +290282,8 @@
               "selected": true,
               "source": "internal/cli/schema_hints/selection/event.json",
               "value": [
-                "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --max-events 1 --format ndjson",
-                "dws event consume user_im_message_reaction_group --group cid-example --max-events 1 --format ndjson"
+                "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --flatten --max-events 1 --format ndjson",
+                "dws event consume user_im_message_reaction_group --group cid-example --flatten --max-events 1 --format ndjson"
               ]
             }
           ],
@@ -290292,8 +290292,8 @@
           "review_reason": "人工依据实时 dws schema（或 Skill/Cobra/pinned MCP 对照）决策化选型文案与门禁；不改写命令身份与参数契约；示例不含 --yes。",
           "source": "internal/cli/schema_hints/selection/event.json",
           "value": [
-            "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --max-events 1 --format ndjson",
-            "dws event consume user_im_message_reaction_group --group cid-example --max-events 1 --format ndjson"
+            "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --flatten --max-events 1 --format ndjson",
+            "dws event consume user_im_message_reaction_group --group cid-example --flatten --max-events 1 --format ndjson"
           ]
         },
         "idempotency": {
@@ -290444,7 +290444,7 @@
       "interface_reason": "Reviewed composite workflow: the command creates or reuses a remote personal-event subscription and coordinates the local event bus and Stream consumer; no single pinned RPC represents the workflow.",
       "is_alias": false,
       "name": "consume",
-      "parameter_count": 27,
+      "parameter_count": 28,
       "parameters": {
         "compact": {
           "description": "提示 bus 客户端期望 compact 渲染（语义透传，bus 仍按原 payload 投递）",
@@ -291124,6 +291124,90 @@
           "required": false,
           "type": "string"
         },
+        "flatten": {
+          "description": "将个人事件 transport envelope 投影为稳定的顶层业务字段",
+          "field_provenance": {
+            "description": {
+              "candidates": [
+                {
+                  "precedence": "cobra_contract",
+                  "selected": true,
+                  "source": "cobra_usage",
+                  "value": "将个人事件 transport envelope 投影为稳定的顶层业务字段"
+                },
+                {
+                  "precedence": "default",
+                  "selected": false,
+                  "source": "default",
+                  "value": ""
+                }
+              ],
+              "precedence": "cobra_contract",
+              "resolution": "highest_precedence",
+              "source": "cobra_usage",
+              "value": "将个人事件 transport envelope 投影为稳定的顶层业务字段"
+            },
+            "property": {
+              "candidates": [
+                {
+                  "precedence": "inference",
+                  "selected": true,
+                  "source": "flag_name_inference",
+                  "value": "flatten"
+                }
+              ],
+              "precedence": "inference",
+              "resolution": "highest_precedence",
+              "source": "flag_name_inference",
+              "value": "flatten"
+            },
+            "required": {
+              "candidates": [
+                {
+                  "precedence": "default",
+                  "selected": true,
+                  "source": "default",
+                  "value": false
+                }
+              ],
+              "precedence": "default",
+              "resolution": "fallback",
+              "source": "default",
+              "value": false
+            },
+            "required_when": {
+              "candidates": [
+                {
+                  "precedence": "default",
+                  "selected": true,
+                  "source": "default",
+                  "value": ""
+                }
+              ],
+              "precedence": "default",
+              "resolution": "highest_precedence",
+              "source": "default",
+              "value": ""
+            },
+            "type": {
+              "candidates": [
+                {
+                  "precedence": "cobra_contract",
+                  "selected": true,
+                  "source": "cobra_flag_type",
+                  "value": "boolean"
+                }
+              ],
+              "precedence": "cobra_contract",
+              "resolution": "highest_precedence",
+              "source": "cobra_flag_type",
+              "value": "boolean"
+            }
+          },
+          "property": "flatten",
+          "required": false,
+          "type": "boolean"
+        },
         "force": {
           "description": "仅 --foreground 模式生效：跳过单实例锁 (慎用：会让云事件被随机切分)",
           "field_provenance": {
@@ -293464,7 +293548,7 @@
         "skills/mono/SKILL.md",
         "skills/mono/references/products/event.md"
       ],
-      "agent_summary": "查询指定个人事件码的 payload 字段结构",
+      "agent_summary": "查询指定个人事件码的输出字段结构；Agent 应查询 --flatten 模式",
       "agent_summary_source": "dws-agent-selection/event",
       "availability": "available",
       "avoid_when": [
@@ -293480,7 +293564,7 @@
       "effect": "read",
       "effect_source": "agent-hint",
       "examples": [
-        "dws event schema user_im_message_receive_at --format json"
+        "dws event schema user_im_message_receive_at --flatten --format json"
       ],
       "field_provenance": {
         "agent_summary": {
@@ -293490,14 +293574,14 @@
               "review_reason": "人工依据实时 dws schema（或 Skill/Cobra/pinned MCP 对照）决策化选型文案与门禁；不改写命令身份与参数契约；示例不含 --yes。",
               "selected": true,
               "source": "internal/cli/schema_hints/selection/event.json",
-              "value": "查询指定个人事件码的 payload 字段结构"
+              "value": "查询指定个人事件码的输出字段结构；Agent 应查询 --flatten 模式"
             }
           ],
           "precedence": "reviewed_explicit",
           "resolution": "highest_precedence",
           "review_reason": "人工依据实时 dws schema（或 Skill/Cobra/pinned MCP 对照）决策化选型文案与门禁；不改写命令身份与参数契约；示例不含 --yes。",
           "source": "internal/cli/schema_hints/selection/event.json",
-          "value": "查询指定个人事件码的 payload 字段结构"
+          "value": "查询指定个人事件码的输出字段结构；Agent 应查询 --flatten 模式"
         },
         "availability": {
           "candidates": [
@@ -293607,7 +293691,7 @@
               "selected": true,
               "source": "internal/cli/schema_hints/selection/event.json",
               "value": [
-                "dws event schema user_im_message_receive_at --format json"
+                "dws event schema user_im_message_receive_at --flatten --format json"
               ]
             }
           ],
@@ -293616,7 +293700,7 @@
           "review_reason": "人工依据实时 dws schema（或 Skill/Cobra/pinned MCP 对照）决策化选型文案与门禁；不改写命令身份与参数契约；示例不含 --yes。",
           "source": "internal/cli/schema_hints/selection/event.json",
           "value": [
-            "dws event schema user_im_message_receive_at --format json"
+            "dws event schema user_im_message_receive_at --flatten --format json"
           ]
         },
         "idempotency": {
@@ -293763,8 +293847,92 @@
       "interface_reason": "命令读取 CLI 内置的个人事件 payload 定义，不绑定 pinned MCP RPC",
       "is_alias": false,
       "name": "schema",
-      "parameter_count": 1,
+      "parameter_count": 2,
       "parameters": {
+        "flatten": {
+          "description": "显示 --flatten 消费模式对应的顶层业务字段 schema",
+          "field_provenance": {
+            "description": {
+              "candidates": [
+                {
+                  "precedence": "cobra_contract",
+                  "selected": true,
+                  "source": "cobra_usage",
+                  "value": "显示 --flatten 消费模式对应的顶层业务字段 schema"
+                },
+                {
+                  "precedence": "default",
+                  "selected": false,
+                  "source": "default",
+                  "value": ""
+                }
+              ],
+              "precedence": "cobra_contract",
+              "resolution": "highest_precedence",
+              "source": "cobra_usage",
+              "value": "显示 --flatten 消费模式对应的顶层业务字段 schema"
+            },
+            "property": {
+              "candidates": [
+                {
+                  "precedence": "inference",
+                  "selected": true,
+                  "source": "flag_name_inference",
+                  "value": "flatten"
+                }
+              ],
+              "precedence": "inference",
+              "resolution": "highest_precedence",
+              "source": "flag_name_inference",
+              "value": "flatten"
+            },
+            "required": {
+              "candidates": [
+                {
+                  "precedence": "default",
+                  "selected": true,
+                  "source": "default",
+                  "value": false
+                }
+              ],
+              "precedence": "default",
+              "resolution": "fallback",
+              "source": "default",
+              "value": false
+            },
+            "required_when": {
+              "candidates": [
+                {
+                  "precedence": "default",
+                  "selected": true,
+                  "source": "default",
+                  "value": ""
+                }
+              ],
+              "precedence": "default",
+              "resolution": "highest_precedence",
+              "source": "default",
+              "value": ""
+            },
+            "type": {
+              "candidates": [
+                {
+                  "precedence": "cobra_contract",
+                  "selected": true,
+                  "source": "cobra_flag_type",
+                  "value": "boolean"
+                }
+              ],
+              "precedence": "cobra_contract",
+              "resolution": "highest_precedence",
+              "source": "cobra_flag_type",
+              "value": "boolean"
+            }
+          },
+          "property": "flatten",
+          "required": false,
+          "type": "boolean"
+        },
         "format": {
           "default": "json",
           "description": "输出格式: json",

internal/cli/schema_catalog_coverage_more_test.go

@@ -194,6 +194,23 @@ func TestCrossPlatformCoverageSchemaCatalogLookupAndConversionEdges(t *testing.T
 	}
 }
 
+func TestEmbeddedSchemaLookupAcceptsCompatibleCLIPathSeparators(t *testing.T) {
+	loaded := embeddedSchemaCatalog()
+	for _, path := range []string{
+		"dev app list",
+		"dev.app.list",
+		"dev/app/list",
+	} {
+		payload, err := schemaPayloadFromLoadedCatalog(loaded, []string{path})
+		if err != nil {
+			t.Fatalf("schemaPayloadFromLoadedCatalog(%q) error = %v", path, err)
+		}
+		if got := schemaString(payload["canonical_path"]); got != "dev.list_dev_app" {
+			t.Fatalf("schemaPayloadFromLoadedCatalog(%q) canonical_path = %q, want %q", path, got, "dev.list_dev_app")
+		}
+	}
+}
+
 type catalogHookSnapshot struct {
 	parameterBindings func(BoundCommandRegistry, SchemaRegistry) error
 	dryRun            func(SchemaRegistry) error

internal/cli/schema_contract_model.go

@@ -617,6 +617,22 @@ func (i SchemaIndex) Resolve(path string) (ToolSpec, bool) {
 	return i.registry.Products[location.product].Tools[location.tool], true
 }
 
+// ResolveQuery adds compatibility for dotted and slash-separated CLI paths at
+// the user-facing query boundary. Resolve remains strict because Registry
+// validation uses it to detect missing canonical identities without falling
+// through to a similarly spelled CLI path.
+func (i SchemaIndex) ResolveQuery(path string) (ToolSpec, bool) {
+	if tool, ok := i.Resolve(path); ok {
+		return tool, true
+	}
+	canonical, ok := i.byCLIPath[normalizeSchemaQueryCLIPath(path)]
+	if !ok {
+		return ToolSpec{}, false
+	}
+	location := i.byCanonical[canonical]
+	return i.registry.Products[location.product].Tools[location.tool], true
+}
+
 // CanonicalPaths returns the complete tool identity set in stable order.
 func (i SchemaIndex) CanonicalPaths() []string {
 	paths := make([]string, 0, len(i.byCanonical))

internal/cli/schema_contract_model_test.go

@@ -514,6 +514,19 @@ func TestSchemaRegistryIndexResolvesCanonicalCLIAndAlias(t *testing.T) {
 			t.Fatalf("Resolve(%q) = %#v, %v", path, resolved.Identity, ok)
 		}
 	}
+	for _, path := range []string{
+		"calendar.attendee.delete",
+		"calendar/attendee/delete",
+		"dws.calendar.attendee.delete",
+	} {
+		resolved, ok := index.ResolveQuery(path)
+		if !ok || resolved.Identity.CanonicalPath != "calendar.attendee_delete" {
+			t.Fatalf("ResolveQuery(%q) = %#v, %v", path, resolved.Identity, ok)
+		}
+	}
+	if _, ok := index.ResolveQuery("calendar.attendee.unknown"); ok {
+		t.Fatal("unknown dotted CLI path unexpectedly resolved")
+	}
 	if got := index.CanonicalPaths(); !reflect.DeepEqual(got, []string{"calendar.attendee_delete"}) {
 		t.Fatalf("CanonicalPaths() = %#v", got)
 	}

internal/cli/schema_delivery_invariants_test.go

@@ -154,6 +154,35 @@ func TestSelectionExplicitEmptyListSurvivesFinalDelivery(t *testing.T) {
 	}
 }
 
+func TestCompatibleSchemaAliasSeparatorsSurviveFinalDelivery(t *testing.T) {
+	snapshot := schemaDeliveryTestSnapshot(schemaDeliveryTestTool{
+		Canonical: "sample.run",
+		CLIPath:   "sample category run",
+		Aliases:   []string{"sample legacy execute"},
+	})
+	encoded, err := json.Marshal(snapshot)
+	if err != nil {
+		t.Fatal(err)
+	}
+	loaded, err := decodeSchemaCatalogSnapshot(encoded)
+	if err != nil {
+		t.Fatalf("decodeSchemaCatalogSnapshot(): %v", err)
+	}
+	canonical, err := schemaPayloadFromLoadedCatalog(loaded, []string{"sample.run"})
+	if err != nil {
+		t.Fatalf("canonical query: %v", err)
+	}
+	for _, path := range []string{"sample legacy execute", "sample.legacy.execute", "sample/legacy/execute"} {
+		alias, aliasErr := schemaPayloadFromLoadedCatalog(loaded, []string{path})
+		if aliasErr != nil {
+			t.Fatalf("alias query %q: %v", path, aliasErr)
+		}
+		if problem := schemaAliasViewProblem(canonical, alias, "sample legacy execute"); problem != "" {
+			t.Fatalf("alias projection for %q: %s", path, problem)
+		}
+	}
+}
+
 func TestValidateSchemaDeliveryInvariantsAllowsOnlyEnvelopeHashes(t *testing.T) {
 	snapshot := schemaDeliveryTestSnapshot(schemaDeliveryTestTool{Canonical: "sample.run", CLIPath: "sample run"})
 	snapshot.SurfaceHash = "sha256:reviewed-command-registry"

internal/cli/schema_full_pipeline_external_test.go

@@ -16,7 +16,7 @@ import (
 // Go's normal per-package coverage accounting attributes the exercised Schema
 // assembly code to internal/cli.
 func TestCrossPlatformCoverageProductionSchemaSourcePipeline(t *testing.T) {
-	root := app.NewRootCommand()
+	root := app.NewSchemaSourceRootCommand()
 	resolved, err := cli.ResolveSchemaBuild(root)
 	if err != nil {
 		t.Fatalf("ResolveSchemaBuild() error = %v", err)
@@ -33,17 +33,17 @@ func TestCrossPlatformCoverageProductionSchemaSourcePipeline(t *testing.T) {
 	if len(snapshot.Tools) == 0 {
 		t.Fatal("production Schema snapshot contains no tools")
 	}
-	registry, err := cli.AssembleSchemaRegistry(app.NewRootCommand())
+	registry, err := cli.AssembleSchemaRegistry(app.NewSchemaSourceRootCommand())
 	if err != nil {
 		t.Fatalf("AssembleSchemaRegistry() error = %v", err)
 	}
 	if len(registry.Products) == 0 {
 		t.Fatal("assembled production Schema registry contains no products")
 	}
-	if err := cli.ValidateEmbeddedRuntimeSchemaCompleteness(app.NewRootCommand()); err != nil {
+	if err := cli.ValidateEmbeddedRuntimeSchemaCompleteness(app.NewSchemaSourceRootCommand()); err != nil {
 		t.Fatalf("ValidateEmbeddedRuntimeSchemaCompleteness() error = %v", err)
 	}
-	root = app.NewRootCommand()
+	root = app.NewSchemaSourceRootCommand()
 	if _, err := cli.ApplyEmbeddedManualSchemaHints(root); err != nil {
 		t.Fatal(err)
 	}

internal/cli/schema_hints/index.json

@@ -634,6 +634,11 @@
       "target": "event consume",
       "reason": "已审查的带 event_key 和输出参数的 Skill 引用，固定映射到当前公开 event consume leaf"
     },
+    "event consume user_im_message_receive_at": {
+      "status": "alias",
+      "target": "event consume",
+      "reason": "已审查的带 event_key 参数的 Skill 引用，固定映射到当前公开 event consume leaf"
+    },
     "event consume user_im_message_receive_group": {
       "status": "alias",
       "target": "event consume",

internal/cli/schema_hints/reference-review.json

@@ -585,6 +585,11 @@
       "target": "event consume",
       "reason": "已审查的带 event_key 和输出参数的 Skill 引用，固定映射到当前公开 event consume leaf"
     },
+    "event consume user_im_message_receive_at": {
+      "status": "alias",
+      "target": "event consume",
+      "reason": "已审查的带 event_key 参数的 Skill 引用，固定映射到当前公开 event consume leaf"
+    },
     "event consume user_im_message_receive_group": {
       "status": "alias",
       "target": "event consume",

internal/cli/schema_hints/selection/event.json

@@ -9,7 +9,7 @@
   },
   "tools": {
     "event.consume": {
-      "agent_summary": "订阅并持续消费指定个人事件，输出 NDJSON 事件流",
+      "agent_summary": "订阅并持续消费指定个人事件；Agent 使用 --flatten 输出顶层业务 NDJSON",
       "use_when": [
         "需要实时监听 @我、指定单聊、指定群或指定发送人的后续消息事件",
         "需要监听指定单聊或群聊中的消息已读、撤回或表情回应事件",
@@ -20,8 +20,8 @@
         "只看事件目录/字段时用 event list / event schema"
       ],
       "examples": [
-        "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --max-events 1 --format ndjson",
-        "dws event consume user_im_message_reaction_group --group cid-example --max-events 1 --format ndjson"
+        "dws event consume user_im_message_receive_user --open-dingtalk-id open-example --flatten --max-events 1 --format ndjson",
+        "dws event consume user_im_message_reaction_group --group cid-example --flatten --max-events 1 --format ndjson"
       ],
       "reviewed": true,
       "review_reason": "人工依据实时 dws schema（或 Skill/Cobra/pinned MCP 对照）决策化选型文案与门禁；不改写命令身份与参数契约；示例不含 --yes。",
@@ -57,7 +57,7 @@
       ]
     },
     "event.schema": {
-      "agent_summary": "查询指定个人事件码的 payload 字段结构",
+      "agent_summary": "查询指定个人事件码的输出字段结构；Agent 应查询 --flatten 模式",
       "use_when": [
         "已知任一公开个人消息 event_key，消费前需要理解扁平输出字段"
       ],
@@ -66,7 +66,7 @@
         "要实际收事件时用 event consume"
       ],
       "examples": [
-        "dws event schema user_im_message_receive_at --format json"
+        "dws event schema user_im_message_receive_at --flatten --format json"
       ],
       "reviewed": true,
       "review_reason": "人工依据实时 dws schema（或 Skill/Cobra/pinned MCP 对照）决策化选型文案与门禁；不改写命令身份与参数契约；示例不含 --yes。",

internal/cli/schema_runtime_registry.go

@@ -284,7 +284,7 @@ func runtimeSchemaPayloadFromRegistry(registry SchemaRegistry, args []string) (m
 	}
 
 	raw := strings.TrimSpace(args[0])
-	if tool, ok := index.Resolve(raw); ok {
+	if tool, ok := index.ResolveQuery(raw); ok {
 		tool = schemaToolForResolvedPath(tool, raw)
 		return renderRegistryToolPayload(tool)
 	}
@@ -338,7 +338,7 @@ func runtimeSchemaAllPayloadFromRegistry(registry SchemaRegistry) (map[string]an
 }
 
 func schemaToolForResolvedPath(tool ToolSpec, raw string) ToolSpec {
-	normalized := normalizeSchemaCLIPath(raw)
+	normalized := normalizeSchemaQueryCLIPath(raw)
 	if normalized == "" || normalized == tool.Identity.CLIPath || normalized == tool.Identity.PrimaryCLIPath {
 		return tool
 	}

internal/event/bus/coverage_edges_test.go

@@ -377,9 +377,11 @@ func TestCrossPlatformCoverageDaemonMethodEdges(t *testing.T) {
 	d.conns.Store("not-a-conn", struct{}{})
 	d.conns.Store(&queryConn{}, struct{}{})
 	d.consumerWG.Add(1)
-	d.shutdown()
+	acceptDone := make(chan struct{})
+	close(acceptDone)
+	d.shutdown(acceptDone)
 	d.consumerWG.Done()
-	d.shutdown()
+	d.shutdown(acceptDone)
 
 	pr, pw, err := os.Pipe()
 	if err != nil {

internal/event/bus/daemon.go

@@ -246,10 +246,11 @@ func Run(ctx context.Context, cfg Config) error {
 	}
 
 	// 7. Graceful shutdown — cancel runCtx first so all background
-	// goroutines wake up, then close listener / drain consumers.
+	// goroutines wake up, then stop accepting connections before draining
+	// consumers. The accept-loop barrier is required before WaitGroup.Wait:
+	// sync.WaitGroup forbids a positive Add racing with Wait.
 	cancelRun()
-	d.shutdown()
-	<-acceptDone
+	d.shutdown(acceptDone)
 	<-idleDone
 	<-dropWarnDone
 
@@ -291,6 +292,10 @@ func (d *daemon) acceptLoop(ctx context.Context) {
 			d.log.Warn("bus: accept error", "err", err)
 			continue
 		}
+		// Track the connection before publishing the handler goroutine. Once
+		// acceptLoop returns, shutdown can therefore close every accepted
+		// connection before waiting for handlers to drain.
+		d.conns.Store(conn, struct{}{})
 		d.consumerWG.Add(1)
 		go func() {
 			defer d.consumerWG.Done()
@@ -303,7 +308,6 @@ func (d *daemon) acceptLoop(ctx context.Context) {
 // Hello → register with Hub → spawn writer goroutine → read until EOF/Bye.
 // Always Unregisters and Closes on exit (plan invariant #5).
 func (d *daemon) handleConnection(ctx context.Context, conn net.Conn) {
-	d.conns.Store(conn, struct{}{})
 	defer func() {
 		d.conns.Delete(conn)
 		conn.Close()
@@ -478,9 +482,10 @@ func (d *daemon) triggerShutdown(reason string) {
 //  1. mark shuttingDown so acceptLoop exits cleanly
 //  2. broadcast Bye to all consumers
 //  3. close listener (interrupts pending Accept)
-//  4. wait for all per-connection goroutines to drain
-//  5. lock + meta cleanup via Run's defers
-func (d *daemon) shutdown() {
+//  4. wait for acceptLoop to return so no future consumerWG.Add can occur
+//  5. close all accepted connections and wait for handlers to drain
+//  6. lock + meta cleanup via Run's defers
+func (d *daemon) shutdown(acceptDone <-chan struct{}) {
 	d.shutdownMu.Lock()
 	defer d.shutdownMu.Unlock()
 	if !d.shuttingDown.CompareAndSwap(false, true) {
@@ -488,6 +493,7 @@ func (d *daemon) shutdown() {
 	}
 	d.hub.Broadcast(transport.Bye{Type: transport.FrameTypeBye, Reason: "shutdown"})
 	_ = d.listener.Close()
+	<-acceptDone
 	// Force-close all open IPC connections so any reader goroutine blocked
 	// on Read() returns with a network error and exits cleanly. Without
 	// this the consumerWG never drains and Run hangs forever.

internal/event/consume/integration_test.go

@@ -31,6 +31,31 @@ import (
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/event/transport"
 )
 
+type consumeReadyWriter struct {
+	ready chan struct{}
+	once  sync.Once
+}
+
+func newConsumeReadyWriter() *consumeReadyWriter {
+	return &consumeReadyWriter{ready: make(chan struct{})}
+}
+
+func (w *consumeReadyWriter) Write(p []byte) (int, error) {
+	if bytes.Contains(p, []byte("[event] ready ")) {
+		w.once.Do(func() { close(w.ready) })
+	}
+	return len(p), nil
+}
+
+func waitForConsumerReady(t *testing.T, ready <-chan struct{}) {
+	t.Helper()
+	select {
+	case <-ready:
+	case <-time.After(3 * time.Second):
+		t.Fatal("consumer did not report ready")
+	}
+}
+
 // TestIntegration_HelloPushdownFiltersAtBus verifies the Hello-time
 // event_types pushdown contract (plan §4 unsung superpower): a consumer
 // subscribing to "im.*" must NOT receive "approval.*" events even when
@@ -49,40 +74,50 @@ func TestIntegration_HelloPushdownFiltersAtBus(t *testing.T) {
 	defer func() { cancel(); <-runDone }()
 
 	var imBuf, approvalBuf bytes.Buffer
-	var wg sync.WaitGroup
-	wg.Add(2)
+	imReady := newConsumeReadyWriter()
+	approvalReady := newConsumeReadyWriter()
+	consumeCtx, cancelConsumers := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancelConsumers()
+	consumeDone := make(chan error, 2)
 
 	// Consumer A: im.* only
 	go func() {
-		defer wg.Done()
-		_ = Run(context.Background(), Config{
+		consumeDone <- Run(consumeCtx, Config{
 			WorkDir:     dir,
 			IPCEndpoint: sock,
 			ClientID:    "ding_test",
 			Stdout:      &imBuf,
-			Stderr:      io.Discard,
+			Stderr:      imReady,
 			EventTypes:  []string{"im.*"},
 			MaxEvents:   3, // 3 im events expected
 		})
 	}()
 	// Consumer B: approval.* only
 	go func() {
-		defer wg.Done()
-		_ = Run(context.Background(), Config{
+		consumeDone <- Run(consumeCtx, Config{
 			WorkDir:     dir,
 			IPCEndpoint: sock,
 			ClientID:    "ding_test",
 			Stdout:      &approvalBuf,
-			Stderr:      io.Discard,
+			Stderr:      approvalReady,
 			EventTypes:  []string{"approval.*"},
 			MaxEvents:   2, // 2 approval events expected
 		})
 	}()
 
-	// Give both consumers time to Hello + register.
-	time.Sleep(200 * time.Millisecond)
+	waitForConsumerReady(t, imReady.ready)
+	waitForConsumerReady(t, approvalReady.ready)
 	close(trigger)
-	wg.Wait()
+	for range 2 {
+		select {
+		case err := <-consumeDone:
+			if err != nil {
+				t.Fatalf("consume failed: %v", err)
+			}
+		case <-consumeCtx.Done():
+			t.Fatalf("consumers did not finish: %v", consumeCtx.Err())
+		}
+	}
 
 	// Verify consumer A got exactly the 3 im.* events.
 	imLines := nonEmptyLines(imBuf.String())
@@ -130,23 +165,32 @@ func TestIntegration_FilterRegexNarrowsFurther(t *testing.T) {
 	defer func() { cancel(); <-runDone }()
 
 	var buf bytes.Buffer
-	consumeDone := make(chan struct{})
+	ready := newConsumeReadyWriter()
+	consumeCtx, cancelConsumer := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancelConsumer()
+	consumeDone := make(chan error, 1)
 	go func() {
-		defer close(consumeDone)
-		_ = Run(context.Background(), Config{
+		consumeDone <- Run(consumeCtx, Config{
 			WorkDir:     dir,
 			IPCEndpoint: sock,
 			ClientID:    "ding_test",
 			Stdout:      &buf,
-			Stderr:      io.Discard,
+			Stderr:      ready,
 			EventTypes:  []string{"im.*"},
 			Filter:      `\.at_v1$`, // only at_v1 events
 			MaxEvents:   1,
 		})
 	}()
-	time.Sleep(150 * time.Millisecond)
+	waitForConsumerReady(t, ready.ready)
 	close(trigger)
-	<-consumeDone
+	select {
+	case err := <-consumeDone:
+		if err != nil {
+			t.Fatalf("consume failed: %v", err)
+		}
+	case <-consumeCtx.Done():
+		t.Fatalf("consumer did not finish: %v", consumeCtx.Err())
+	}
 
 	lines := nonEmptyLines(buf.String())
 	if len(lines) != 1 {

internal/event/consume/run.go

@@ -90,6 +90,10 @@ type Config struct {
 	// NormalizeFormat; an empty Format here defaults to NDJSON inside
 	// BuildPipeline.
 	Format Format
+	// Flatten records whether the caller explicitly selected a structured
+	// business projection. Projector remains the executable behavior; this
+	// field is surfaced in dry-run output so users can verify the final mode.
+	Flatten bool
 	// OutputDir, if non-empty, switches the fallback sink from stdout to
 	// "file per event" under this directory.
 	OutputDir string

internal/event/consume/validate.go

@@ -132,6 +132,7 @@ func PrintDryRun(w io.Writer, cfg Config) {
 		fmt.Fprintf(w, "  filter           : %s\n", cfg.Filter)
 	}
 	fmt.Fprintf(w, "  format           : %s\n", cfg.Format)
+	fmt.Fprintf(w, "  flatten          : %v\n", cfg.Flatten)
 	if cfg.OutputDir != "" {
 		fmt.Fprintf(w, "  output_dir       : %s\n", cfg.OutputDir)
 	}