docs(release): add v1.0.56-beta.2 notes

fix: separate Agent Product from claw-type

CHANGELOG.md

@@ -6,6 +6,44 @@ The format is inspired by [Keep a Changelog](https://keepachangelog.com/) and th
 
 ## [Unreleased]
 
+## [1.0.56-beta.2] - 2026-07-30
+
+This beta adds PRs #831 and #835 on top of v1.0.56-beta.1. It separates
+Agent Product observability and IM display identity from the stable
+edition-owned PAT and routing identity, and reduces common-path Skill context
+loading without changing the public command or Runtime Schema surface.
+
+### Changed
+
+- **Agent Product identity separation** (#831) — sends `DWS_AGENT_PRODUCT` through the new `x-dws-agent-product` observability Header and uses a valid non-empty value for the IM `clawType` display label whenever `--ai-tag` is enabled. Because `--ai-tag` defaults to `true`, callers that set `DWS_AGENT_PRODUCT` change the displayed label by default. With `--ai-tag=false`, native `chat message send` / `reply` calls preserve their existing wire shape by sending an empty IM `clawType`, while shortcut calls omit the argument. Unset or empty Product values omit the Header and preserve the active edition's IM display default.
+- **Agent Host dimension convention** (#831) — new integrations should send the runtime form (`cloud` or `desktop`) through `DWS_AGENT_HOST` and report the product separately through `DWS_AGENT_PRODUCT`. Legacy combined labels such as `qwenwork_cloud` remain syntactically valid for compatibility.
+- **Reduced common-path Skill context** (#835) — keeps the complete 97-command Chat Shortcut inventory in Runtime Catalog and leaf Schema while routing common intents through compact Skill tables and references. The generated Skill policy now detects drift, forced full-reference loading, and context-budget regressions; the common Chat plus shared activation estimate drops from 7,301 to 4,771 `o200k_base` tokens without changing the 845-tool Schema surface.
+
+### Fixed
+
+- **Stable PAT/routing identity** (#831) — restores the CLI-emitted open-source HTTP `claw-type` and PAT `hostControl.clawType` to the edition-fixed `openClaw` value. `DWS_AGENT_PRODUCT` no longer changes those wire values, and the client continues to derive PAT, authentication, routing, and Discovery behaviour from the existing independent signals.
+
+## [1.0.56-beta.1] - 2026-07-30
+
+This beta starts the v1.0.56 line on top of v1.0.55 and packages PRs #817,
+#806, and #834, together with release-validation fixes #838 and #839. It closes
+the remaining Agent-visible IM shortcut gaps, introduces reviewed
+command-scoped parameter normalization without guessing business identifiers
+or values, and prevents deterministic personal-event subscription failures
+from becoming unbounded retry storms.
+
+### Added
+
+- **Complete IM shortcut workflows** (#817) — publishes the previously excluded `+chat-messages`, `+messages-send`, `+messages-send-card`, `+search-msg`, and `+thread-replies` shortcuts in Runtime Schema. Unified send, streaming-card delivery, advanced search, thread replies, and opt-in resource downloads now share reviewed parameters, selection guidance, and runtime-aligned safety semantics.
+- **Reviewed parameter concept normalization** (#806) — adds a closed parameter-concept dictionary and generated command-level alias table, covering reviewed IM synonyms while preserving the boundaries between group, conversation, user, open-user, cursor, and paging identifiers.
+
+### Fixed
+
+- **Message delivery and resource handling** (#817) — resolves direct recipients through exact contact search, preserves rich and nested message resources, avoids same-name download overwrites, and prevents read shortcuts from silently returning empty results on non-interactive input.
+- **Parameter parsing safety** (#806) — rejects ambiguous, blocked, or conflicting aliases before dispatch, normalizes explicit boolean values such as `--dry-run false`, and keeps internal pre-parse handler details out of user-visible errors.
+- **Personal-event subscription retry safety** (#834) — adds cross-process attempt claims, deterministic backoff and jitter, `Retry-After` handling, terminal holds, compare-and-swap completion, and fail-closed state handling across all public personal-event subscriptions, preventing deterministic failures from causing unbounded callback retries.
+- **Scoped CI and release validation reliability** (#838, #839) — keeps scoped coverage aligned with intentionally skipped supporting profiles, gives focused race and Multi-profile E2E suites enough time for the current `internal/app` workload, and preserves hidden E2E diagnostics on failure.
+
 ## [1.0.55-beta.8] - 2026-07-30
 
 This beta revalidates the `v1.0.55-beta.7` product baseline through a complete

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 "1.0.55-beta.8"
+  version "1.0.56-beta.1"
   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/v1.0.55-beta.8/dws-darwin-arm64.tar.gz"
-      sha256 "07fabf720fa98f82c56027df703a3ad3f0aec16c957ea684047928f9f74fa00f"
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.56-beta.1/dws-darwin-arm64.tar.gz"
+      sha256 "26b696a7807f87cf9c6a24634e08008bac0501f2664a85a8355904309068d55e"
     else
-      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.55-beta.8/dws-darwin-amd64.tar.gz"
-      sha256 "fbec64dc5c3463a04de9720ffb1fd247196e619ea81b976b47ecbf751a17fb72"
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.56-beta.1/dws-darwin-amd64.tar.gz"
+      sha256 "28daa9fa1fc8131e3567a39042d5f34efc61d5901bf7d88a67f075a4ff366be6"
     end
   end
 
   on_linux do
     if Hardware::CPU.arm?
-      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.55-beta.8/dws-linux-arm64.tar.gz"
-      sha256 "f111cdffef0188ddf954d2174fa0d318069bcec80104775483f13f645aa8089b"
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.56-beta.1/dws-linux-arm64.tar.gz"
+      sha256 "5f9ead9b939d2db31354797cc7b7431f27aaca41f171d256d4d4e9269cc66d51"
     else
-      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.55-beta.8/dws-linux-amd64.tar.gz"
-      sha256 "d69475b7f3cec4bad4c051834df22fbfadfa7075b82347e6ca7490f67d71d0b1"
+      url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.56-beta.1/dws-linux-amd64.tar.gz"
+      sha256 "06b950676cefdd8e054c76af32cd9758c4c7786908d654281d9c65e7a3d575cf"
     end
   end
 
   resource "skills" do
-    url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.55-beta.8/dws-skills.zip"
-    sha256 "be8c9267704cfef1319fc9cd2fcba5aebe45b670b4dc37d3dae15f65523356f8"
+    url "https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases/download/v1.0.56-beta.1/dws-skills.zip"
+    sha256 "7448ccfc8e189b3d76b3e8f11f5dedc4f5a5b53380f9059e6af4c63a38e38f28"
   end
 
   def install

Makefile

@@ -8,7 +8,7 @@ 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 test-plan test-auth-legacy-compat lint format-check fmt policy edition-test interface-integrity authoritative-interface-integrity coverage-gate coverage-gate-platform update-interface-baseline reset-interface-baseline schema-compatibility skill-command-integrity cli-smoke mock-mcp-smoke test-schema-agent-examples generate-schema generate-schema-agent-metadata fetch-mcp-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 test-auth-legacy-compat lint format-check fmt policy edition-test interface-integrity authoritative-interface-integrity coverage-gate coverage-gate-platform update-interface-baseline reset-interface-baseline schema-compatibility skill-command-integrity skill-context-budget cli-smoke mock-mcp-smoke test-schema-agent-examples generate-schema generate-schema-agent-metadata fetch-mcp-metadata generate-schema-catalog package release release-pre release-stable changelog-pre changelog-stable publish-homebrew-formula setup-hooks
 
 all: setup-hooks fmt lint build test rebuild
 
@@ -30,6 +30,7 @@ help:
 	@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"
 	@printf "  make skill-command-integrity - Check dws commands referenced by skills exist\n"
+	@printf "  make skill-context-budget - Check generated Skill drift and common-path context budgets\n"
 	@printf "  make cli-smoke     - Verify help for every public top-level command\n"
 	@printf "  make mock-mcp-smoke - Verify HTTP and stdio MCP request/response transport\n"
 	@printf "  make test-schema-agent-examples - Contract-check all Agent examples and dry-run the eligible subset\n"
@@ -84,6 +85,7 @@ fmt:
 policy: test-auth-legacy-compat
 	@mkdir -p "$(POLICY_GOTMPDIR)"
 	@$(POLICY_ENV) ./scripts/policy/check-open-source-assets.sh
+	@$(POLICY_ENV) ./scripts/policy/check-skill-context-budget.sh
 	@$(POLICY_ENV) ./scripts/policy/check-schema-command-registry.sh
 	@$(POLICY_ENV) ./scripts/policy/check-command-surface.sh --strict
 	@$(POLICY_ENV) ./scripts/policy/check-generated-drift.sh
@@ -121,6 +123,9 @@ schema-compatibility:
 skill-command-integrity:
 	@./scripts/policy/check-skill-commands.sh
 
+skill-context-budget:
+	@./scripts/policy/check-skill-context-budget.sh
+
 cli-smoke:
 	@./scripts/policy/check-cli-smoke.sh
 

docs/reference.md

@@ -5,8 +5,8 @@
 | Variable | Purpose / 用途 |
 |---------|---------|
 | `DWS_CONFIG_DIR` | Override default config directory / 覆盖默认配置目录 |
-| `DWS_AGENT_PRODUCT` | Optional, caller-declared Agent product sent through the existing HTTP `claw-type` header (for example `qwenwork`). Surrounding ASCII spaces/tabs are trimmed; the remaining value must be at most 64 bytes and match `^[A-Za-z0-9][A-Za-z0-9_-]*$`. Unset or empty values preserve the edition default (`openClaw` in the open-source build). / 可选、由调用方声明的 Agent 产品标识，经校验后覆盖 HTTP `claw-type` 请求头；未设置或为空时保持当前发行版默认值 |
-| `DWS_AGENT_HOST` | Optional, caller-declared Agent runtime form sent as `x-dws-agent-host` (for example `cloud` or `desktop`). Surrounding ASCII spaces/tabs are trimmed; the remaining value must be at most 64 bytes and match `^[a-z0-9][a-z0-9_-]*$`; unset values are omitted. / 可选、由调用方声明的 Agent 运行形态，经校验后作为 `x-dws-agent-host` 发送；未设置时省略 |
+| `DWS_AGENT_PRODUCT` | Optional, caller-declared Agent product sent as `x-dws-agent-product` (for example `qwenwork`) for downstream logs/BI and used as the IM `clawType` display label when `--ai-tag` is enabled. `--ai-tag` defaults to `true`, so a configured Product changes the displayed label by default. With `--ai-tag=false`, native `chat message send` / `reply` calls send an empty `clawType`, while shortcut calls omit the argument. Surrounding ASCII spaces/tabs are trimmed; the remaining value must be at most 64 bytes and match `^[A-Za-z0-9][A-Za-z0-9_-]*$`. Unset or empty values omit the Header and use the edition's IM display default. This client never uses Product to change the separate HTTP `claw-type` PAT/routing label. / 可选、由调用方声明的 Agent 产品标识，经校验后作为 `x-dws-agent-product` 发送，并用于 IM 小尾巴；`--ai-tag` 默认为 `true`，因此配置 Product 后默认会改变展示标签。使用 `--ai-tag=false` 时，原生 `chat message send` / `reply` 发送空的 `clawType`，shortcut 调用则省略该参数。未设置时省略请求头且 IM 使用发行版默认值；本客户端不会用 Product 修改独立的 HTTP `claw-type` |
+| `DWS_AGENT_HOST` | Optional, caller-declared Agent runtime form sent as `x-dws-agent-host` (for example `cloud` or `desktop`) for downstream logs/BI. Surrounding ASCII spaces/tabs are trimmed; the remaining value must be at most 64 bytes and match `^[a-z0-9][a-z0-9_-]*$`; unset values are omitted. This client does not use Host for PAT, authentication, Discovery, or MCP endpoint selection. / 可选、由调用方声明的 Agent 运行形态，经校验后作为 `x-dws-agent-host` 发送给下游日志/BI；本客户端不使用该值进行 PAT、鉴权、Discovery 或 MCP 端点选择，未设置时省略 |
 | `DWS_<PRODUCT>_MCP_URL` | Override a product MCP endpoint for local development / 本地开发时覆盖指定产品 MCP endpoint |
 | `DWS_CLIENT_ID` | OAuth client ID (DingTalk AppKey) |
 | `DWS_CLIENT_SECRET` | OAuth client secret (DingTalk AppSecret) |
@@ -14,23 +14,36 @@
 | `DWS_ALLOW_HTTP_ENDPOINTS` | Set `1` to allow HTTP for loopback during dev / 设为 `1` 允许回环地址 HTTP，仅用于开发调试 |
 | `DWS_DISABLE_KEYCHAIN` | macOS only. Set `1` to skip system Keychain for the encryption key and use file-based storage (same scheme as Linux). For sandboxed runtimes (e.g. Codex App) that block Keychain APIs. Weakens at-rest protection — DEK and ciphertext live in the same directory. / 仅 macOS。设为 `1` 时跳过系统 Keychain，密钥以文件形式存储（与 Linux 一致）。用于 Keychain API 被拦截的沙盒环境（如 Codex App）。代价是 DEK 与密文同目录，保护强度低于默认方案 |
 
-### Agent Product and Host trust model / Agent 产品与运行形态的信任模型
+### Agent Product, Host, and `claw-type` / Agent 产品、运行形态与 `claw-type`
 
-`DWS_AGENT_PRODUCT` and `DWS_AGENT_HOST` are caller-declared selection and
-observation signals. They are not credentials, attestations, or proof of the
-calling host's identity. DingTalk services may record them for logs/BI and may
-combine supported values with separately authenticated context for PAT
-compatibility, PAT identity/source derivation, or Discovery eligibility. A
-service must allowlist supported values and must never grant access, bypass
-authentication, or skip authorization solely because either Header claims a
-particular product or runtime form. They are not used to select ordinary MCP
-tool endpoints.
+`DWS_AGENT_PRODUCT` and `DWS_AGENT_HOST` are caller-declared observation
+signals. They are not credentials, attestations, or proof of the calling
+host's identity. The CLI validates and emits `x-dws-agent-product` and
+`x-dws-agent-host`, but does not use either value to derive its authentication,
+PAT mode, Discovery behaviour, or ordinary MCP endpoint selection. Downstream
+services own and must document their own contracts for these caller-declared
+Headers.
 
-`DWS_AGENT_PRODUCT` controls only the HTTP `claw-type` Header. The similarly
-named `clawType` tool argument on IM send operations is an independent
-message-display axis used for the “Send from AI” label. It remains controlled
-by the active edition's `ClawTypeValue` and `--ai-tag`; changing
-`DWS_AGENT_PRODUCT` does not change that message label.
+Service integrators should treat both Headers as untrusted input, allowlist
+expected values, and should not grant access, bypass authentication, or skip
+authorization solely because a Header claims a particular Product or Host.
+
+The HTTP `claw-type` Header is a separate, edition-fixed PAT/routing label:
+`openClaw` in the open-source build. `DWS_AGENT_PRODUCT` never changes it or
+PAT `hostControl.clawType`. On IM send/reply operations with `--ai-tag`,
+however, a valid non-empty Product value is used as the `clawType` tool
+argument so the delivered message carries the matching “Send from AI” label.
+Because `--ai-tag` defaults to `true`, this display change is enabled by
+default for callers that set Product. With `--ai-tag=false`, native
+`chat message send` / `reply` calls serialize `clawType: ""`, while shortcut
+calls omit the argument; this client does not assume downstream services treat
+an empty value and an absent key as equivalent. The display-value precedence
+when the tag is enabled is valid non-empty `DWS_AGENT_PRODUCT`, then the active
+edition's `ClawTypeValue`, then `openClaw`.
+
+Do not set arbitrary Product values that the target downstream and IM services
+have not explicitly enabled; an unknown value may be ignored or may not render
+the expected label.
 
 For QwenWork, report the dimensions separately:
 
@@ -39,16 +52,25 @@ DWS_AGENT_PRODUCT=qwenwork
 DWS_AGENT_HOST=cloud       # or desktop
 ```
 
-Do not set arbitrary product values that the target service has not explicitly
-enabled. Older combined Host labels such as `qwenwork_cloud` still satisfy the
-generic syntax for compatibility, but new integrations should use the
-two-dimensional convention above.
+Older combined Host labels such as `qwenwork_cloud` still satisfy the generic
+syntax for compatibility, but new integrations should use the two-dimensional
+convention above.
 
 `DWS_AGENT_PRODUCT` 和 `DWS_AGENT_HOST` 均由调用方声明，不是认证凭据，也不能证明
-真实宿主身份。服务端可以在独立认证上下文中将受支持值用于日志/BI、PAT 兼容策略、
-PAT 身份/来源派生或 Discovery 准入，但不得仅凭这两个 Header 放权、绕过认证或跳过
-授权。HTTP `claw-type` 与 IM 消息发送参数 `clawType` 是两个独立维度；后者仅控制
-“Send from AI”展示，仍由发行版 `ClawTypeValue` 和 `--ai-tag` 决定。
+真实宿主身份。CLI 只负责校验并发送 `x-dws-agent-product` 与 `x-dws-agent-host`，
+不会用它们派生本客户端的鉴权、PAT 模式、Discovery 行为或 MCP 端点；下游服务的
+使用契约由对应服务自行定义和说明。HTTP `claw-type` 是发行版固定的 PAT/路由标签，
+开源版固定为 `openClaw`，不受 `DWS_AGENT_PRODUCT` 影响。
+
+服务集成方应将这两个请求头视为不可信输入并对白名单值做校验，不应仅因请求头声明了
+某个 Product 或 Host 就授予访问、绕过认证或跳过鉴权。
+
+`--ai-tag` 默认为 `true`，因此配置合法非空 Product 后，默认发送的 IM 工具参数
+`clawType` 及小尾巴会随之改变。传入 `--ai-tag=false` 时，原生
+`chat message send` / `reply` 会发送 `clawType: ""`，shortcut 调用则省略该参数；
+本客户端不假定下游会将空值与键缺失等价处理。启用小尾巴时，展示值优先级依次为
+`DWS_AGENT_PRODUCT`、当前发行版的 `ClawTypeValue`、`openClaw`。不要传入目标下游及
+IM 服务未明确支持的 Product 值，否则可能被忽略或无法展示预期标签。
 
 ## Exit Codes / 退出码
 

internal/app/agent_host.go

@@ -33,7 +33,7 @@ func init() {
 	configmeta.Register(configmeta.ConfigItem{
 		Name:        envDWSAgentHost,
 		Category:    configmeta.CategoryExternal,
-		Description: "调用 DWS 的 Agent 运行形态标识；服务端可结合产品用于观测和 PAT 兼容策略",
+		Description: "调用 DWS 的 Agent 运行形态标识；作为 x-dws-agent-host 发送供下游观测，本客户端不使用该值改变 PAT、鉴权或路由",
 		Example:     "cloud",
 	})
 }

internal/app/agent_product.go

@@ -24,8 +24,8 @@ func init() {
 	configmeta.Register(configmeta.ConfigItem{
 		Name:         agentproduct.EnvName,
 		Category:     configmeta.CategoryExternal,
-		Description:  "调用方声明的 Agent 产品标识；覆盖 HTTP claw-type，但不是认证凭据",
-		DefaultValue: "由当前发行版决定",
+		Description:  "调用方声明的 Agent 产品标识；作为 x-dws-agent-product 发送并用于 IM 小尾巴，本客户端不使用该值改变 HTTP claw-type/PAT",
+		DefaultValue: "未设置（请求头省略，IM 使用当前发行版默认值）",
 		Example:      "qwenwork",
 	})
 }
@@ -47,25 +47,26 @@ func invalidAgentProductError() error {
 	)
 }
 
-// resolveEffectiveAgentProduct resolves the request-header identity with one
-// shared precedence rule: a valid non-empty runtime override wins, otherwise
-// the edition's MergeHeaders value wins, otherwise the OSS default is used.
-// Invalid runtime input falls back here for library callers that bypass root
-// validation; normal CLI execution rejects it before network access.
-func resolveEffectiveAgentProduct(headers map[string]string) string {
-	fallback := edition.DefaultOSSClawType
-	if value := headers[agentproduct.HeaderName]; value != "" {
-		fallback = value
+// resolveEditionClawType resolves the fixed routing/PAT identity supplied by
+// the active edition. DWS_AGENT_PRODUCT is deliberately not consulted.
+func resolveEditionClawType(headers map[string]string) string {
+	if value := headers["claw-type"]; value != "" {
+		return value
 	}
-	value, err := agentproduct.ResolveFromEnv(fallback)
-	if err != nil {
-		return fallback
-	}
-	return value
+	return edition.DefaultOSSClawType
 }
 
-func applyAgentProductOverride(headers map[string]string) map[string]string {
-	value := resolveEffectiveAgentProduct(headers)
+// applyAgentProductHeader injects only a valid, non-empty caller-declared
+// product. Invalid values are omitted on library paths that bypass root
+// validation; normal CLI execution rejects them before network access.
+func applyAgentProductHeader(headers map[string]string) map[string]string {
+	value, err := agentproduct.ResolveFromEnv("")
+	if err != nil || value == "" {
+		if headers != nil {
+			delete(headers, agentproduct.HeaderName)
+		}
+		return headers
+	}
 	if headers == nil {
 		headers = make(map[string]string)
 	}

internal/app/agent_product_test.go

@@ -26,13 +26,16 @@ import (
 	"github.com/spf13/cobra"
 )
 
-func TestUnsetAgentProductKeepsOpenSourceDefault(t *testing.T) {
+func TestUnsetAgentProductOmitsHeaderAndKeepsOpenSourceClawType(t *testing.T) {
 	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
 	t.Setenv(agentproduct.EnvName, "")
 
 	headers := resolveIdentityHeaders()
-	if got := headers[agentproduct.HeaderName]; got != edition.DefaultOSSClawType {
-		t.Fatalf("%s = %q, want %q", agentproduct.HeaderName, got, edition.DefaultOSSClawType)
+	if got := headers["claw-type"]; got != edition.DefaultOSSClawType {
+		t.Fatalf("claw-type = %q, want %q", got, edition.DefaultOSSClawType)
+	}
+	if _, ok := headers[agentproduct.HeaderName]; ok {
+		t.Fatalf("unset Product must omit %s", agentproduct.HeaderName)
 	}
 }
 
@@ -59,38 +62,46 @@ func TestParseAgentProductReturnsStableValidationError(t *testing.T) {
 	}
 }
 
-func TestResolveIdentityHeadersAgentProductPrecedence(t *testing.T) {
+func TestResolveIdentityHeadersSeparatesAgentProductFromClawType(t *testing.T) {
 	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
 
 	oldEdition := edition.Get()
 	t.Cleanup(func() { edition.Override(oldEdition) })
 	edition.Override(&edition.Hooks{
 		MergeHeaders: func(headers map[string]string) map[string]string {
-			headers[agentproduct.HeaderName] = "wukong"
+			headers["claw-type"] = "wukong"
+			headers[agentproduct.HeaderName] = "merge-product-must-not-win"
 			headers["x-edition-header"] = "preserved"
 			return headers
 		},
 		EnterpriseCredentialHeaders: func(headers map[string]string) map[string]string {
-			headers[agentproduct.HeaderName] = "enterprise-default"
+			headers["claw-type"] = "credential-must-not-win"
+			headers[agentproduct.HeaderName] = "credential-product-must-not-win"
 			headers["x-enterprise-header"] = "preserved"
 			return headers
 		},
 	})
 
-	t.Run("unset keeps edition default", func(t *testing.T) {
+	t.Run("unset omits Product and keeps edition claw-type", func(t *testing.T) {
 		t.Setenv(agentproduct.EnvName, "")
 		headers := resolveIdentityHeaders()
-		if got := headers[agentproduct.HeaderName]; got != "wukong" {
-			t.Fatalf("%s = %q, want wukong", agentproduct.HeaderName, got)
+		if got := headers["claw-type"]; got != "wukong" {
+			t.Fatalf("claw-type = %q, want wukong", got)
+		}
+		if _, ok := headers[agentproduct.HeaderName]; ok {
+			t.Fatalf("unset Product must omit %s", agentproduct.HeaderName)
 		}
 	})
 
-	t.Run("valid override is final", func(t *testing.T) {
+	t.Run("valid Product is final without changing claw-type", func(t *testing.T) {
 		t.Setenv(agentproduct.EnvName, " qwenwork ")
 		headers := resolveIdentityHeaders()
 		if got := headers[agentproduct.HeaderName]; got != "qwenwork" {
 			t.Fatalf("%s = %q, want qwenwork", agentproduct.HeaderName, got)
 		}
+		if got := headers["claw-type"]; got != "wukong" {
+			t.Fatalf("claw-type = %q, want wukong", got)
+		}
 		if got := headers["x-edition-header"]; got != "preserved" {
 			t.Fatalf("edition header = %q, want preserved", got)
 		}
@@ -102,21 +113,48 @@ func TestResolveIdentityHeadersAgentProductPrecedence(t *testing.T) {
 		}
 	})
 
-	t.Run("invalid library input falls back to edition", func(t *testing.T) {
+	t.Run("invalid library input omits Product and keeps edition claw-type", func(t *testing.T) {
 		t.Setenv(agentproduct.EnvName, "qwen work")
 		headers := resolveIdentityHeaders()
-		if got := headers[agentproduct.HeaderName]; got != "wukong" {
-			t.Fatalf("%s = %q, want wukong", agentproduct.HeaderName, got)
+		if got := headers["claw-type"]; got != "wukong" {
+			t.Fatalf("claw-type = %q, want wukong", got)
+		}
+		if _, ok := headers[agentproduct.HeaderName]; ok {
+			t.Fatalf("invalid Product must omit %s", agentproduct.HeaderName)
 		}
 	})
 }
 
-func TestApplyAgentProductOverrideAllocatesHeaders(t *testing.T) {
-	t.Setenv(agentproduct.EnvName, "qwenwork")
+func TestApplyAgentProductHeader(t *testing.T) {
+	t.Run("valid value allocates headers", func(t *testing.T) {
+		t.Setenv(agentproduct.EnvName, "qwenwork")
 
-	headers := applyAgentProductOverride(nil)
-	if got := headers[agentproduct.HeaderName]; got != "qwenwork" {
-		t.Fatalf("%s = %q, want qwenwork", agentproduct.HeaderName, got)
+		headers := applyAgentProductHeader(nil)
+		if got := headers[agentproduct.HeaderName]; got != "qwenwork" {
+			t.Fatalf("%s = %q, want qwenwork", agentproduct.HeaderName, got)
+		}
+	})
+
+	for _, tc := range []struct {
+		name  string
+		value string
+	}{
+		{name: "empty value", value: ""},
+		{name: "invalid value", value: "qwen work"},
+	} {
+		t.Run(tc.name+" removes inherited header", func(t *testing.T) {
+			t.Setenv(agentproduct.EnvName, tc.value)
+			headers := applyAgentProductHeader(map[string]string{
+				agentproduct.HeaderName: "must-not-leak",
+				"x-preserved":           "yes",
+			})
+			if _, ok := headers[agentproduct.HeaderName]; ok {
+				t.Fatalf("%s must be omitted", agentproduct.HeaderName)
+			}
+			if got := headers["x-preserved"]; got != "yes" {
+				t.Fatalf("x-preserved = %q, want yes", got)
+			}
+		})
 	}
 }
 
@@ -167,17 +205,17 @@ func TestEffectiveClawTypeDoesNotInvokeEnterpriseCredentialHeaders(t *testing.T)
 	hookCalled := false
 	edition.Override(&edition.Hooks{
 		MergeHeaders: func(headers map[string]string) map[string]string {
-			headers[agentproduct.HeaderName] = "wukong"
+			headers["claw-type"] = "wukong"
 			return headers
 		},
 		EnterpriseCredentialHeaders: func(headers map[string]string) map[string]string {
 			hookCalled = true
-			headers[agentproduct.HeaderName] = "enterprise-default"
+			headers["claw-type"] = "enterprise-default"
 			return headers
 		},
 	})
 
-	t.Setenv(agentproduct.EnvName, "")
+	t.Setenv(agentproduct.EnvName, "qwenwork")
 	if got := effectiveClawType(); got != "wukong" {
 		t.Fatalf("effectiveClawType() = %q, want wukong", got)
 	}
@@ -186,7 +224,7 @@ func TestEffectiveClawTypeDoesNotInvokeEnterpriseCredentialHeaders(t *testing.T)
 	}
 }
 
-func TestAgentProductHeaderIsSeparateFromMessageClawType(t *testing.T) {
+func TestAgentProductControlsObservabilityHeaderAndMessageClawTypeOnly(t *testing.T) {
 	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
 	t.Setenv(agentproduct.EnvName, "qwenwork")
 
@@ -195,20 +233,24 @@ func TestAgentProductHeaderIsSeparateFromMessageClawType(t *testing.T) {
 	edition.Override(&edition.Hooks{
 		ClawTypeValue: "message-brand",
 		MergeHeaders: func(headers map[string]string) map[string]string {
-			headers[agentproduct.HeaderName] = "wukong"
+			headers["claw-type"] = "wukong"
 			return headers
 		},
 	})
 
-	if got := resolveIdentityHeaders()[agentproduct.HeaderName]; got != "qwenwork" {
+	headers := resolveIdentityHeaders()
+	if got := headers[agentproduct.HeaderName]; got != "qwenwork" {
 		t.Fatalf("HTTP %s = %q, want qwenwork", agentproduct.HeaderName, got)
 	}
-	if got := edition.ClawType(); got != "message-brand" {
-		t.Fatalf("message clawType = %q, want message-brand", got)
+	if got := headers["claw-type"]; got != "wukong" {
+		t.Fatalf("HTTP claw-type = %q, want wukong", got)
+	}
+	if got := edition.ClawType(); got != "qwenwork" {
+		t.Fatalf("message clawType = %q, want qwenwork", got)
 	}
 }
 
-func TestResolveIdentityHeadersRestoresAgentProductAfterNilCredentialHeaders(t *testing.T) {
+func TestResolveIdentityHeadersRestoresIdentityAfterNilCredentialHeaders(t *testing.T) {
 	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
 	t.Setenv(agentproduct.EnvName, "qwenwork")
 
@@ -218,7 +260,7 @@ func TestResolveIdentityHeadersRestoresAgentProductAfterNilCredentialHeaders(t *
 	credentialHookCalled := false
 	edition.Override(&edition.Hooks{
 		MergeHeaders: func(headers map[string]string) map[string]string {
-			headers[agentproduct.HeaderName] = "wukong"
+			headers["claw-type"] = "wukong"
 			return headers
 		},
 		EnterpriseCredentialHeaders: func(map[string]string) map[string]string {
@@ -234,25 +276,49 @@ func TestResolveIdentityHeadersRestoresAgentProductAfterNilCredentialHeaders(t *
 	if got := headers[agentproduct.HeaderName]; got != "qwenwork" {
 		t.Fatalf("%s = %q, want qwenwork", agentproduct.HeaderName, got)
 	}
+	if got := headers["claw-type"]; got != "wukong" {
+		t.Fatalf("claw-type = %q, want wukong", got)
+	}
 }
 
-func TestEffectiveClawTypeUsesAgentProductOverride(t *testing.T) {
+func TestResolveIdentityHeadersRestoresDefaultsAfterNilMergeHeaders(t *testing.T) {
+	t.Setenv("DWS_CONFIG_DIR", t.TempDir())
+	t.Setenv(agentproduct.EnvName, "")
+
+	oldEdition := edition.Get()
+	t.Cleanup(func() { edition.Override(oldEdition) })
+	edition.Override(&edition.Hooks{
+		MergeHeaders: func(map[string]string) map[string]string {
+			return nil
+		},
+	})
+
+	headers := resolveIdentityHeaders()
+	if got := headers["claw-type"]; got != edition.DefaultOSSClawType {
+		t.Fatalf("claw-type = %q, want %q", got, edition.DefaultOSSClawType)
+	}
+	if _, ok := headers[agentproduct.HeaderName]; ok {
+		t.Fatalf("unset Product must omit %s", agentproduct.HeaderName)
+	}
+}
+
+func TestEffectiveClawTypeIgnoresAgentProduct(t *testing.T) {
 	oldEdition := edition.Get()
 	t.Cleanup(func() { edition.Override(oldEdition) })
 	edition.Override(&edition.Hooks{
 		MergeHeaders: func(headers map[string]string) map[string]string {
-			headers[agentproduct.HeaderName] = "wukong"
+			headers["claw-type"] = "wukong"
 			return headers
 		},
 	})
 
 	t.Setenv(agentproduct.EnvName, "qwenwork")
-	if got := effectiveClawType(); got != "qwenwork" {
-		t.Fatalf("effectiveClawType() = %q, want qwenwork", got)
+	if got := effectiveClawType(); got != "wukong" {
+		t.Fatalf("effectiveClawType() = %q, want wukong", got)
 	}
 	t.Setenv(authpkg.AgentCodeEnv, "agent-code")
-	if got := apperrors.HostControlBlock()["clawType"]; got != "qwenwork" {
-		t.Fatalf("hostControl.clawType = %q, want qwenwork", got)
+	if got := apperrors.HostControlBlock()["clawType"]; got != "wukong" {
+		t.Fatalf("hostControl.clawType = %q, want wukong", got)
 	}
 
 	t.Setenv(agentproduct.EnvName, "")

internal/app/pat_auth_retry.go

@@ -569,9 +569,9 @@ func handlePatAuthCheck(
 	// In host-controlled PAT mode (driven solely by DINGTALK_DWS_AGENTCODE),
 	// or when flowId is absent, the CLI returns machine-readable JSON to
 	// stderr and leaves UI/polling/retry to the host. `claw-type` is NOT
-	// used for this decision — it is only forwarded on the wire via
-	// the edition default / DWS_AGENT_PRODUCT override and surfaced in
-	// hostControl for traceability.
+	// used for this decision — its edition-fixed value is forwarded on the
+	// wire and surfaced in hostControl for traceability. DWS_AGENT_PRODUCT
+	// does not affect this PAT contract.
 	if hostOwnedPAT || patData.Data.FlowID == "" {
 		if hostOwnedPAT {
 			return executor.Result{}, &apperrors.PATError{RawJSON: enrichPATErrorForHostControl(patErr.RawJSON)}

internal/app/pat_auth_retry_test.go

@@ -1007,11 +1007,11 @@ func TestHandlePatAuthCheck_HostControlledFlowIDPassthrough(t *testing.T) {
 	t.Setenv("DWS_CONFIG_DIR", tmpDir)
 	// Host-owned decision: driven ONLY by DINGTALK_DWS_AGENTCODE.
 	// DINGTALK_AGENT is set to demonstrate it does NOT leak into
-	// hostControl.clawType. With no DWS_AGENT_PRODUCT override the
-	// open-source edition default remains "openClaw".
+	// hostControl.clawType. DWS_AGENT_PRODUCT is also set to demonstrate
+	// that Product does not change the open-source fixed "openClaw" value.
 	t.Setenv(authpkg.AgentCodeEnv, "agt-sales")
 	t.Setenv("DINGTALK_AGENT", "sales-copilot")
-	t.Setenv(agentproduct.EnvName, "")
+	t.Setenv(agentproduct.EnvName, "qwenwork")
 
 	mock := &mockRunner{
 		runFunc: func(ctx context.Context, inv executor.Invocation) (executor.Result, error) {

internal/app/pat_hostcontrol_wire.go

@@ -29,9 +29,8 @@ import (
 //   - Host-owned is triggered iff DINGTALK_DWS_AGENTCODE is non-empty.
 //   - When triggered, `clawType` in the emitted hostControl block MUST be the
 //     exact value the CLI actually injects on the wire. Each edition supplies
-//     its existing default and an optional valid DWS_AGENT_PRODUCT overrides
-//     it. Invalid input falls back here for library compatibility; root command
-//     execution rejects it before network access.
+//     its fixed value; DWS_AGENT_PRODUCT is a separate observability and IM
+//     message-display signal and never affects this PAT value.
 //   - When DINGTALK_DWS_AGENTCODE is empty the provider returns "" so
 //     HostControlBlock yields nil and no hostControl block is emitted.
 func init() {
@@ -59,5 +58,5 @@ func effectiveClawType() string {
 			headers = h.MergeHeaders(headers)
 		}
 	}
-	return resolveEffectiveAgentProduct(headers)
+	return resolveEditionClawType(headers)
 }

internal/app/runner.go

@@ -1005,9 +1005,8 @@ func resolveIdentityHeaders() map[string]string {
 
 	// Inject environment variable based headers for MCP gateway tracking.
 	// DINGTALK_AGENT, if set by the caller, is forwarded verbatim as the
-	// x-dingtalk-agent header. It does NOT influence claw-type (which comes
-	// from the edition default plus the explicit DWS_AGENT_PRODUCT override)
-	// and it does NOT influence the host-owned PAT decision (driven solely by
+	// x-dingtalk-agent header. It does NOT influence the edition-fixed
+	// claw-type or the host-owned PAT decision (driven solely by
 	// DINGTALK_DWS_AGENTCODE).
 	sessionID := os.Getenv(envDingtalkSessionID)
 	if sessionID == "" {
@@ -1068,22 +1067,30 @@ func resolveIdentityHeaders() map[string]string {
 	if fn := edition.Get().MergeHeaders; fn != nil {
 		headers = fn(headers)
 	}
-	// Resolve the Agent Product before credential injection. The credential
-	// hook has a separate contract and must not be able to replace the
-	// request identity used by PAT hostControl serialization.
-	headers = applyAgentProductOverride(headers)
-	agentProduct := headers[agentproduct.HeaderName]
+	if headers == nil {
+		headers = make(map[string]string)
+	}
+
+	// claw-type is the edition-fixed routing/PAT identity. Agent Product is a
+	// separate caller-declared observability and IM-display dimension.
+	clawType := resolveEditionClawType(headers)
+	headers["claw-type"] = clawType
+	headers = applyAgentProductHeader(headers)
+	agentProduct, hasAgentProduct := headers[agentproduct.HeaderName]
 	if fn := edition.Get().EnterpriseCredentialHeaders; fn != nil {
 		headers = fn(headers)
 	}
 	if headers == nil {
 		headers = make(map[string]string)
 	}
-	// DWS_AGENT_PRODUCT is the explicit caller override for the existing
-	// claw-type wire header. Reassert the resolved product after credential
-	// injection so that hook cannot alter identity. Invalid values are ignored
-	// on this best-effort library path; root execution rejects them earlier.
-	headers[agentproduct.HeaderName] = agentProduct
+	// Credential hooks cannot alter either identity dimension. Restore the
+	// fixed claw-type and the validated Product Header (or its absence).
+	headers["claw-type"] = clawType
+	if hasAgentProduct {
+		headers[agentproduct.HeaderName] = agentProduct
+	} else {
+		delete(headers, agentproduct.HeaderName)
+	}
 	return headers
 }
 

internal/cli/schema_agent_metadata/index.json

@@ -1,6 +1,6 @@
 {
   "version": 1,
-  "source_hash": "sha256:7484b82d1ca793a6129acfaa192ff08fc0864d47a6e75ecd16d8e7fa32857e16",
+  "source_hash": "sha256:670ca810a83bf2aa6f387a18c3af746393d5ea9de24570cbcebbcf994eb7b613",
   "surface_hash": "sha256:60eee8e2f37d6d9d60689efce85082798eb9ad38b7ba7c0b471c3de676a85a16",
   "coverage": {
     "surface_products": 26,

internal/cli/schema_agent_metadata_audit.json

@@ -1,6 +1,6 @@
 {
   "version": 1,
-  "source_hash": "sha256:7484b82d1ca793a6129acfaa192ff08fc0864d47a6e75ecd16d8e7fa32857e16",
+  "source_hash": "sha256:670ca810a83bf2aa6f387a18c3af746393d5ea9de24570cbcebbcf994eb7b613",
   "surface_hash": "sha256:60eee8e2f37d6d9d60689efce85082798eb9ad38b7ba7c0b471c3de676a85a16",
   "source_files": 160,
   "hint_files": 54,

internal/cli/schema_catalog/catalog.json

@@ -1,12 +1,12 @@
 {
   "version": 1,
   "surface_hash": "sha256:60eee8e2f37d6d9d60689efce85082798eb9ad38b7ba7c0b471c3de676a85a16",
-  "source_hash": "sha256:ba691e70f1c232fc3378a743c3fcd34113960d30b46bffa7d1fc8b524115052c",
+  "source_hash": "sha256:862734fb976a4c9c78d010b7391c74822f37805f5acc650551f4ac60f701c13a",
   "catalog": {
     "agent_metadata": {
       "products_with_metadata": 26,
       "source": "embedded-skill-metadata",
-      "source_hash": "sha256:7484b82d1ca793a6129acfaa192ff08fc0864d47a6e75ecd16d8e7fa32857e16",
+      "source_hash": "sha256:670ca810a83bf2aa6f387a18c3af746393d5ea9de24570cbcebbcf994eb7b613",
       "surface_hash": "sha256:60eee8e2f37d6d9d60689efce85082798eb9ad38b7ba7c0b471c3de676a85a16",
       "surface_products": 26,
       "surface_tools": 845,

internal/helpers/chat_message_search_test.go

@@ -22,6 +22,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/agentproduct"
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
 )
 
@@ -217,6 +218,73 @@ func TestCrossPlatformCoverageChatSendResolvesUserBeforeDispatch(t *testing.T) {
 	}
 }
 
+func TestChatSendAndReplyDefaultToAgentProductForIMClawType(t *testing.T) {
+	t.Setenv(agentproduct.EnvName, "qwenwork")
+
+	tests := []struct {
+		name string
+		args []string
+	}{
+		{
+			name: "send",
+			args: []string{"message", "send", "--open-dingtalk-id", "D1", "--text", "hello"},
+		},
+		{
+			name: "reply",
+			args: []string{"message", "reply", "--conversation-id", "cid", "--ref-msg-id", "mid", "--ref-sender", "D1", "--text", "hello"},
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			caller := &chatChangedContractCaller{}
+			if err := executeChatChangedContract(t, caller, tc.args...); err != nil {
+				t.Fatal(err)
+			}
+			if len(caller.calls) != 1 || caller.calls[0].toolName != "send_personal_message" {
+				t.Fatalf("calls = %#v", caller.calls)
+			}
+			if got := caller.calls[0].args["clawType"]; got != "qwenwork" {
+				t.Fatalf("clawType = %#v, want qwenwork", got)
+			}
+		})
+	}
+}
+
+func TestChatSendAndReplyDisableAITagWithEmptyClawType(t *testing.T) {
+	t.Setenv(agentproduct.EnvName, "qwenwork")
+
+	tests := []struct {
+		name string
+		args []string
+	}{
+		{
+			name: "send",
+			args: []string{"message", "send", "--open-dingtalk-id", "D1", "--text", "hello", "--ai-tag=false"},
+		},
+		{
+			name: "reply",
+			args: []string{"message", "reply", "--conversation-id", "cid", "--ref-msg-id", "mid", "--ref-sender", "D1", "--text", "hello", "--ai-tag=false"},
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			caller := &chatChangedContractCaller{}
+			if err := executeChatChangedContract(t, caller, tc.args...); err != nil {
+				t.Fatal(err)
+			}
+			if len(caller.calls) != 1 || caller.calls[0].toolName != "send_personal_message" {
+				t.Fatalf("calls = %#v", caller.calls)
+			}
+			got, present := caller.calls[0].args["clawType"]
+			if !present || got != "" {
+				t.Fatalf("clawType = %#v, present = %v; want present empty string", got, present)
+			}
+		})
+	}
+}
+
 func TestCrossPlatformCoverageChatSendFailsClosedWhenUserCannotResolve(t *testing.T) {
 	caller := &chatChangedContractCaller{}
 	err := executeChatChangedContract(t, caller, "message", "send", "--user", "123", "--text", "hello")

internal/pat/pat.go

@@ -55,12 +55,15 @@ Host-owned PAT 开关：
   由宿主处理全部 UI / 交互 / 回调节奏 / 重试逻辑，
   CLI 侧不再拉起任何本地浏览器 / 轮询。
 
-服务端 Agent 产品标签 claw-type：
-  开源构建默认在出站 MCP 请求中注入 claw-type: openClaw。
-  如设置 DWS_AGENT_PRODUCT，则使用经校验的环境变量值覆盖该默认值。
-  hostControl.clawType 会回填请求实际使用的值，避免 PAT 与请求标识漂移。
-  该值由调用方声明，不是认证凭据；服务端不得仅凭它放权或跳过授权。
-  它也不会修改 IM 消息展示使用的 clawType 参数或 --ai-tag 行为。
+服务端 PAT / 路由标签 claw-type：
+  开源构建固定在出站 MCP 请求中注入 claw-type: openClaw，
+  hostControl.clawType 会回填相同值。DWS_AGENT_PRODUCT 不会修改它。
+
+Agent 产品标识 DWS_AGENT_PRODUCT：
+  合法非空值作为 x-dws-agent-product 请求头发送，供下游日志 / BI 使用；
+  本客户端不使用该值派生或改变 PAT、鉴权或路由，下游使用契约由对应
+  服务自行定义。同时该值作为启用 --ai-tag 时 IM 消息小尾巴的 clawType
+  参数。未设置或为空时请求头省略，小尾巴回退发行版默认值。
 
 DINGTALK_AGENT（可选，仅供 x-dingtalk-agent 使用）：
   如设置，将原样注入 HTTP 请求头 x-dingtalk-agent，

internal/shortcut/smart/compatibility_coverage_test.go

@@ -24,6 +24,7 @@ import (
 
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/helpers"
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/shortcut"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/agentproduct"
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
 	"github.com/spf13/cobra"
 )
@@ -99,6 +100,8 @@ func newPlatformCoverageRoot() *cobra.Command {
 }
 
 func TestCrossPlatformCoverageAIMessageTag(t *testing.T) {
+	t.Setenv(agentproduct.EnvName, "qwenwork")
+
 	tests := []struct {
 		name string
 		argv []string
@@ -125,8 +128,8 @@ func TestCrossPlatformCoverageAIMessageTag(t *testing.T) {
 			if send.product != "chat" || send.tool != "send_personal_message" {
 				t.Fatalf("last call = %s/%s, want chat/send_personal_message", send.product, send.tool)
 			}
-			if got := send.args["clawType"]; got != edition.ClawType() {
-				t.Fatalf("clawType = %#v, want %q", got, edition.ClawType())
+			if got := send.args["clawType"]; got != "qwenwork" {
+				t.Fatalf("clawType = %#v, want qwenwork", got)
 			}
 		})
 	}

pkg/agentproduct/agent_product.go

@@ -12,7 +12,7 @@
 // limitations under the License.
 
 // Package agentproduct defines the caller-provided Agent product identity
-// carried on outbound DWS requests.
+// used for outbound observability and IM message display.
 package agentproduct
 
 import (
@@ -25,8 +25,9 @@ import (
 const (
 	// EnvName is the runtime override for the Agent product identity.
 	EnvName = "DWS_AGENT_PRODUCT"
-	// HeaderName is the existing wire header used for the Agent product.
-	HeaderName = "claw-type"
+	// HeaderName is the observability header used for the Agent product.
+	// It is intentionally separate from the routing/PAT claw-type header.
+	HeaderName = "x-dws-agent-product"
 	// MaxValueBytes bounds the value because it is attached to every outbound
 	// MCP request. Supported values are ASCII, so bytes and characters match.
 	MaxValueBytes = 64

pkg/agentproduct/agent_product_test.go

@@ -103,3 +103,12 @@ func TestResolveFromEnv(t *testing.T) {
 		}
 	})
 }
+
+func TestHeaderNameIsSeparateFromClawType(t *testing.T) {
+	if HeaderName != "x-dws-agent-product" {
+		t.Fatalf("HeaderName = %q, want x-dws-agent-product", HeaderName)
+	}
+	if HeaderName == "claw-type" {
+		t.Fatal("Agent Product observability Header must not reuse claw-type")
+	}
+}

pkg/edition/default.go

@@ -15,18 +15,18 @@ package edition
 
 import "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/syncdata"
 
-// DefaultOSSClawType is the default wire value for request header claw-type
-// in the open-source build. DWS_AGENT_PRODUCT may explicitly override the
-// request identity; unrelated caller inputs such as DINGTALK_AGENT do not.
+// DefaultOSSClawType is the fixed wire value for request header claw-type in
+// the open-source build. It is intentionally independent of caller-provided
+// environment variables so PAT and routing behaviour stays predictable.
 const DefaultOSSClawType = "openClaw"
 
 // defaultHooks returns the open-source edition defaults.
 //
 // MergeHeaders is the only hook that ships with behaviour: it supplies
 // DefaultOSSClawType so every open-source MCP request has a stable default.
-// The core applies an optional DWS_AGENT_PRODUCT override after edition hooks.
-// All other fields are nil — the internal code interprets nil as "use
-// standard open-source behaviour".
+// DWS_AGENT_PRODUCT is sent through a separate observability Header and never
+// changes claw-type. All other fields are nil — the internal code interprets
+// nil as "use standard open-source behaviour".
 func defaultHooks() *Hooks {
 	return &Hooks{
 		Name: "open",

pkg/edition/edition.go

@@ -21,6 +21,7 @@ import (
 	"context"
 	"sync"
 
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/agentproduct"
 	"github.com/spf13/cobra"
 )
 
@@ -87,10 +88,10 @@ type Hooks struct {
 
 	// ClawTypeValue is the display identity carried only in message-send tool
 	// arguments (parameter clawType) so the IM server can render the
-	// "Send from AI" indicator on delivered messages. It is intentionally
-	// separate from the HTTP claw-type Agent Product header and is not
-	// overridden by DWS_AGENT_PRODUCT. Empty → DefaultOSSClawType; overlays
-	// set their own message-display value (e.g. "wukong").
+	// "Send from AI" indicator on delivered messages. A valid non-empty
+	// DWS_AGENT_PRODUCT overrides this display default, but never the separate
+	// HTTP claw-type routing/PAT header. Empty → DefaultOSSClawType; overlays
+	// set their own message-display default (e.g. "wukong").
 	ClawTypeValue string
 
 	// PersonalEventSourceID identifies the personal-event source channel
@@ -115,8 +116,8 @@ type Hooks struct {
 	MergeHeaders func(base map[string]string) map[string]string
 
 	// --- EnterpriseCredential HTTP headers ---
-	// This hook is only for credential material. It must not set claw-type;
-	// the core reasserts that Header after the hook returns.
+	// This hook is only for credential material. It must not set claw-type or
+	// x-dws-agent-product; the core reasserts both after the hook returns.
 	EnterpriseCredentialHeaders func(base map[string]string) map[string]string
 
 	// --- auth ---
@@ -206,16 +207,22 @@ func Override(h *Hooks) {
 	current = h
 }
 
-// ClawType returns the message-display identity for the active edition,
-// falling back to DefaultOSSClawType when the overlay does not set one.
-// Message-send helpers attach this value as the clawType tool argument so the
-// IM server can label delivered messages as sent via AI. It is a separate axis
-// from the HTTP claw-type header and is not affected by DWS_AGENT_PRODUCT.
+// ClawType returns the message-display identity for the active edition.
+// A valid non-empty DWS_AGENT_PRODUCT wins; otherwise the active edition's
+// ClawTypeValue (or DefaultOSSClawType) is used. Message-send helpers attach
+// this value as the clawType tool argument so the IM server can label delivered
+// messages as sent via AI. It never changes the HTTP claw-type routing/PAT
+// header.
 func ClawType() string {
-	if v := Get().ClawTypeValue; v != "" {
-		return v
+	fallback := Get().ClawTypeValue
+	if fallback == "" {
+		fallback = DefaultOSSClawType
 	}
-	return DefaultOSSClawType
+	value, err := agentproduct.ResolveFromEnv(fallback)
+	if err != nil {
+		return fallback
+	}
+	return value
 }
 
 // PersonalEventSourceID returns the source channel for user-level events.

pkg/edition/edition_test.go

@@ -13,9 +13,14 @@
 
 package edition
 
-import "testing"
+import (
+	"testing"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/agentproduct"
+)
 
 func TestClawTypeDefaultsToOSSValue(t *testing.T) {
+	t.Setenv(agentproduct.EnvName, "")
 	prev := Get()
 	defer Override(prev)
 
@@ -26,6 +31,7 @@ func TestClawTypeDefaultsToOSSValue(t *testing.T) {
 }
 
 func TestClawTypeUsesOverlayValue(t *testing.T) {
+	t.Setenv(agentproduct.EnvName, "")
 	prev := Get()
 	defer Override(prev)
 
@@ -35,6 +41,28 @@ func TestClawTypeUsesOverlayValue(t *testing.T) {
 	}
 }
 
+func TestClawTypeUsesValidAgentProduct(t *testing.T) {
+	t.Setenv(agentproduct.EnvName, " qwenwork ")
+	prev := Get()
+	defer Override(prev)
+
+	Override(&Hooks{Name: "overlay", ClawTypeValue: "wukong"})
+	if got := ClawType(); got != "qwenwork" {
+		t.Fatalf("ClawType() = %q, want qwenwork", got)
+	}
+}
+
+func TestClawTypeInvalidAgentProductFallsBackToOverlay(t *testing.T) {
+	t.Setenv(agentproduct.EnvName, "qwen work")
+	prev := Get()
+	defer Override(prev)
+
+	Override(&Hooks{Name: "overlay", ClawTypeValue: "wukong"})
+	if got := ClawType(); got != "wukong" {
+		t.Fatalf("ClawType() = %q, want overlay fallback wukong", got)
+	}
+}
+
 func TestOpenStaticServersIncludesCoreProducts(t *testing.T) {
 	servers := openStaticServers()
 	byID := make(map[string]ServerInfo, len(servers))

scripts/gen_skill_shortcut_sections.py

@@ -1,13 +1,15 @@
 #!/usr/bin/env python3
-"""Generate shortcut sections for DWS skills.
+"""Generate shortcut discovery sections for DWS skills.
 
 The skill should teach agents which high-level shortcut entries are available
-in the public catalog and Runtime Schema. Leaf Schema publishes the Agent
-contract, while leaf `--help` remains the source of truth for accepted flags.
+without forcing large product catalogs into every common task. Leaf Schema
+publishes the Agent contract, while leaf `--help` remains the source of truth
+for accepted flags.
 """
 
 from __future__ import annotations
 
+import argparse
 import json
 import os
 import sys
@@ -47,6 +49,13 @@ MONO_END = "<!-- VISIBLE_SHORTCUTS_OVERVIEW_END -->"
 PRODUCT_START = "<!-- VISIBLE_SHORTCUTS_START -->"
 PRODUCT_END = "<!-- VISIBLE_SHORTCUTS_END -->"
 
+# Large, high-frequency product skills should route known intents directly and
+# keep their full shortcut inventory in Runtime Catalog/Schema. Add services
+# here only after verifying that the product skill has its own reviewed routing
+# section and intent table; compacting a sparse skill without an alternative
+# route would make its shortcuts harder to discover.
+COMPACT_PRODUCT_SERVICES = {"chat"}
+
 
 def md_escape(value: Any) -> str:
     text = str(value or "")
@@ -90,21 +99,24 @@ def mono_overview(items: list[dict[str, Any]]) -> str:
         path = SERVICE_TO_SKILL.get(service)
         skill = "—"
         if path:
-            skill = next((part for part in path.parts if part.startswith("dingtalk-")), path.parent.name)
-        rows.append(f"| `{md_escape(service)}` | {count} | `{md_escape(skill)}` | `dws shortcut list --service {md_escape(service)} --format json` |")
+            skill = next((part for part in reversed(path.parts) if part.startswith("dingtalk-")), path.parent.name)
+        rows.append(f"| `{md_escape(service)}` | {count} | `{md_escape(skill)}` |")
     body = "\n".join(rows)
     return f"""{MONO_START}
 ## Shortcut 总览
 
-下面统计当前公开 catalog 中的 shortcut。mono 模式不展开 200+ 行明细，避免 skill 过重；需要执行时先按产品路由，再用 `dws shortcut list --service <service> --format json` 读取参数、约束、风险和示例，最后用 `dws <service> +<shortcut> --help` 核对当前 Cobra flags。multi 模式的各产品 skill 会展开该产品的 shortcut 表。
+下面只统计当前公开 catalog 中的 shortcut，不展开完整明细。已知意图应先按产品 Skill、意图表或任务 reference 选择唯一命令；命令已选中时直接执行，只在参数或安全语义不确定时读取 leaf Schema，在当前 Cobra flags 不确定时读取 leaf Help。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service <service> --format json` 做最后回退；不要为已知高频意图加载完整产品 Catalog。
 
-| 服务 | shortcut 数 | multi skill | 发现命令 |
-|---|---:|---|---|
+| 服务 | shortcut 数 | multi skill |
+|---|---:|---|
 {body}
 {MONO_END}"""
 
 
 def product_section(service: str, rows: list[dict[str, Any]]) -> str:
+    if service in COMPACT_PRODUCT_SERVICES:
+        return compact_product_section(service, rows)
+
     table = []
     for item in rows:
         table.append(
@@ -114,7 +126,7 @@ def product_section(service: str, rows: list[dict[str, Any]]) -> str:
     return f"""{PRODUCT_START}
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "{service} +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service {service} --format json` 批量发现；最后以 `dws {service} <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "{service} +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws {service} <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service {service} --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|
@@ -122,17 +134,38 @@ def product_section(service: str, rows: list[dict[str, Any]]) -> str:
 {PRODUCT_END}"""
 
 
-def update_mono(items: list[dict[str, Any]]) -> None:
+def compact_product_section(service: str, rows: list[dict[str, Any]]) -> str:
+    return f"""{PRODUCT_START}
+## Shortcut 发现（按需）
+
+`{md_escape(service)}` 当前有 {len(rows)} 条公开 shortcut，完整清单保留在 Runtime Catalog 与 Schema，不在高频产品根 Skill 中重复展开。已知意图直接使用下方的优先路由、意图表或任务 reference；命令已选中时直接执行，只在参数/安全语义不确定时读取 leaf Schema，在当前 Cobra flags 不确定时读取 leaf Help。
+
+仅当现有路由和 reference 都无法定位低频能力时，才执行 `dws shortcut list --service {md_escape(service)} --format json` 做最后回退；不要为已知高频意图加载完整 Shortcut Catalog 或产品级 Schema。
+{PRODUCT_END}"""
+
+
+def apply_update(path: Path, text: str, updated: str, check: bool) -> bool:
+    if updated == text:
+        return False
+    if check:
+        print(f"generated skill drift: {path.relative_to(ROOT)}", file=sys.stderr)
+    else:
+        path.write_text(updated, encoding="utf-8")
+    return True
+
+
+def update_mono(items: list[dict[str, Any]], check: bool) -> list[Path]:
     text = MONO_SKILL.read_text(encoding="utf-8")
     block = mono_overview(items)
     updated = replace_block(text, MONO_START, MONO_END, block, "## 产品总览")
-    MONO_SKILL.write_text(updated, encoding="utf-8")
+    return [MONO_SKILL] if apply_update(MONO_SKILL, text, updated, check) else []
 
 
-def update_product_skills(items: list[dict[str, Any]]) -> None:
+def update_product_skills(items: list[dict[str, Any]], check: bool) -> list[Path]:
     by_service: dict[str, list[dict[str, Any]]] = defaultdict(list)
     for item in items:
         by_service[item["service"]].append(item)
+    changed = []
     for service, path in SERVICE_TO_SKILL.items():
         if service not in by_service:
             continue
@@ -142,13 +175,26 @@ def update_product_skills(items: list[dict[str, Any]]) -> None:
         block = product_section(service, by_service[service])
         anchor = "## 概念地图" if service == "devapp" else "## 意图表"
         updated = replace_block(text, PRODUCT_START, PRODUCT_END, block, anchor)
-        path.write_text(updated, encoding="utf-8")
+        if apply_update(path, text, updated, check):
+            changed.append(path)
+    return changed
 
 
 def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--check",
+        action="store_true",
+        help="verify generated skill sections are current without rewriting files",
+    )
+    args = parser.parse_args()
+
     items = collect_visible()
-    update_mono(items)
-    update_product_skills(items)
+    changed = update_mono(items, args.check)
+    changed.extend(update_product_skills(items, args.check))
+    if args.check and changed:
+        print("run: python3 scripts/gen_skill_shortcut_sections.py", file=sys.stderr)
+        return 1
     print(f"visible_shortcuts={len(items)} services={len(set(item['service'] for item in items))}")
     return 0
 

scripts/policy/check-skill-context-budget.sh

@@ -0,0 +1,41 @@
+#!/bin/sh
+set -eu
+
+ROOT="$(CDPATH= cd -- "$(dirname -- "$0")/../.." && pwd)"
+cd "$ROOT"
+
+python3 scripts/gen_skill_shortcut_sections.py --check
+
+chat_skill="skills/multi/dingtalk-chat/SKILL.md"
+mono_skill="skills/mono/SKILL.md"
+chat_max_bytes=14000
+
+chat_bytes="$(wc -c < "$chat_skill" | tr -d ' ')"
+if [ "$chat_bytes" -gt "$chat_max_bytes" ]; then
+	printf '%s\n' \
+		"skill context budget exceeded: $chat_skill is ${chat_bytes} bytes (max ${chat_max_bytes})" >&2
+	exit 1
+fi
+
+shortcut_rows="$(
+	awk '
+		/<!-- VISIBLE_SHORTCUTS_START -->/ { in_block = 1; next }
+		/<!-- VISIBLE_SHORTCUTS_END -->/ { in_block = 0 }
+		in_block && /^\| `dws chat \+/ { count++ }
+		END { print count + 0 }
+	' "$chat_skill"
+)"
+if [ "$shortcut_rows" -ne 0 ]; then
+	printf '%s\n' \
+		"skill context budget exceeded: $chat_skill re-expanded $shortcut_rows shortcut rows" >&2
+	exit 1
+fi
+
+if grep -Fq "充分阅读产品参考文件" "$mono_skill"; then
+	printf '%s\n' \
+		"skill context budget regression: $mono_skill requires full product-reference loading" >&2
+	exit 1
+fi
+
+printf '%s\n' \
+	"skill context budget: ok (chat_bytes=$chat_bytes max=$chat_max_bytes shortcut_rows=$shortcut_rows)"

skills/mono/SKILL.md

@@ -39,26 +39,26 @@ cli_version: ">=1.0.15"
 <!-- VISIBLE_SHORTCUTS_OVERVIEW_START -->
 ## Shortcut 总览
 
-下面统计当前公开 catalog 中的 shortcut。mono 模式不展开 200+ 行明细，避免 skill 过重；需要执行时先按产品路由，再用 `dws shortcut list --service <service> --format json` 读取参数、约束、风险和示例，最后用 `dws <service> +<shortcut> --help` 核对当前 Cobra flags。multi 模式的各产品 skill 会展开该产品的 shortcut 表。
+下面只统计当前公开 catalog 中的 shortcut，不展开完整明细。已知意图应先按产品 Skill、意图表或任务 reference 选择唯一命令；命令已选中时直接执行，只在参数或安全语义不确定时读取 leaf Schema，在当前 Cobra flags 不确定时读取 leaf Help。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service <service> --format json` 做最后回退；不要为已知高频意图加载完整产品 Catalog。
 
-| 服务 | shortcut 数 | multi skill | 发现命令 |
-|---|---:|---|---|
-| `aitable` | 29 | `dingtalk-workspace` | `dws shortcut list --service aitable --format json` |
-| `attendance` | 19 | `dingtalk-workspace` | `dws shortcut list --service attendance --format json` |
-| `calendar` | 20 | `dingtalk-workspace` | `dws shortcut list --service calendar --format json` |
-| `chat` | 97 | `dingtalk-workspace` | `dws shortcut list --service chat --format json` |
-| `contact` | 14 | `dingtalk-workspace` | `dws shortcut list --service contact --format json` |
-| `devapp` | 19 | `dingtalk-workspace` | `dws shortcut list --service devapp --format json` |
-| `ding` | 4 | `dingtalk-workspace` | `dws shortcut list --service ding --format json` |
-| `doc` | 17 | `dingtalk-workspace` | `dws shortcut list --service doc --format json` |
-| `drive` | 7 | `dingtalk-workspace` | `dws shortcut list --service drive --format json` |
-| `mail` | 10 | `dingtalk-workspace` | `dws shortcut list --service mail --format json` |
-| `minutes` | 6 | `dingtalk-workspace` | `dws shortcut list --service minutes --format json` |
-| `oa` | 7 | `dingtalk-workspace` | `dws shortcut list --service oa --format json` |
-| `report` | 2 | `dingtalk-workspace` | `dws shortcut list --service report --format json` |
-| `sheet` | 2 | `dingtalk-workspace` | `dws shortcut list --service sheet --format json` |
-| `todo` | 11 | `dingtalk-workspace` | `dws shortcut list --service todo --format json` |
-| `wiki` | 1 | `dingtalk-workspace` | `dws shortcut list --service wiki --format json` |
+| 服务 | shortcut 数 | multi skill |
+|---|---:|---|
+| `aitable` | 29 | `dingtalk-aitable` |
+| `attendance` | 19 | `dingtalk-misc` |
+| `calendar` | 20 | `dingtalk-calendar` |
+| `chat` | 97 | `dingtalk-chat` |
+| `contact` | 14 | `dingtalk-contact` |
+| `devapp` | 19 | `dingtalk-dev` |
+| `ding` | 4 | `dingtalk-misc` |
+| `doc` | 17 | `dingtalk-doc` |
+| `drive` | 7 | `dingtalk-drive` |
+| `mail` | 10 | `dingtalk-mail` |
+| `minutes` | 6 | `dingtalk-minutes` |
+| `oa` | 7 | `dingtalk-misc` |
+| `report` | 2 | `dingtalk-misc` |
+| `sheet` | 2 | `dingtalk-misc` |
+| `todo` | 11 | `dingtalk-todo` |
+| `wiki` | 1 | `dingtalk-wiki` |
 <!-- VISIBLE_SHORTCUTS_OVERVIEW_END -->
 
 ## 多组织 / 多账号
@@ -174,7 +174,7 @@ Step 3 → 加 --yes 执行命令
 1. 意图分类：首先，判断用户指令的核心 动词/动作 属于哪一类。这比关注名词更重要。
 2. 歧义处理与信息追问：如果用户指令模糊或包含多个产品的关键字，严禁猜测。必须主动向用户追问以澄清意图。这是你作为智能助手而非命令执行器的核心价值。
 3. 精准产品映射：在完成前两步，意图已经清晰后，参考产品总览和意图判断决策树 来选择产品。
-4. 充分阅读产品参考文件，通过编写代码或直接调用指令实现用户意图。
+4. 按任务最小化读取：已知高频意图直接使用本 Skill 或产品 reference 已给出的唯一命令，不预加载完整产品参考文件；只有路由、参数或异常恢复确实需要时，才读取对应产品或任务 reference。
 
 ## 命令发现（Schema 渐进查询 + --help 互为补充）
 
@@ -184,6 +184,8 @@ Step 3 → 加 --yes 执行命令
 
 本节同时适用于基础/原子命令与公开内建 `+` shortcut。用户自定义或未公开 shortcut 不进入发布 Schema；其是否可执行仍以当前 Cobra help 为准。
 
+**已知命令路径例外**：当本 Skill、产品意图表或任务 reference 已经给出精确 CLI path 时，不要再查询产品级/分组级 Schema，也不要调用完整 Shortcut Catalog；可直接执行。只有参数、约束或安全语义不确定时才读取该命令的 leaf Schema，只有当前 Cobra flags 不确定时才补读 leaf Help。
+
 稳定 command identity、主 CLI path 和 alias 已在构建时由 reviewed registry 与真实 Cobra tree 精确绑定。Agent 不应读取 Catalog 文件、native annotation 或其他生成 JSON 来重新推断命令；所有运行时查询都以当前二进制交付的 Schema 投影为准。
 
 ```bash
@@ -250,7 +252,7 @@ dws schema --all --format json
 |------|--------|
 | 命令是否存在、当前 Cobra 接受哪些 flags | `dws <cli_path> --help` |
 | Agent 选择、参数映射/required/组合约束、risk/confirmation（原子/基础命令） | `dws schema "<cli_path>"`（按需加 `--compact`） |
-| shortcut 的参数、组合约束、risk/confirmation、示例 | `dws shortcut list --service <service> --format json` |
+| shortcut 的参数、组合约束、risk/confirmation、示例 | 已知路径优先 `dws schema --cli-path "<service> +<shortcut>" --compact --format json`；完整 `shortcut list` 仅用于无法定位低频能力时的最后回退 |
 | 人类可读用法 | `dws <cli_path> --help` |
 | 钉钉中的文档、文件、日程、消息等实际数据 | 真正执行对应的 `read` / `search` / `list` 命令 |
 

skills/multi/dingtalk-aitable/SKILL.md

@@ -20,7 +20,7 @@ metadata:
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "aitable +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service aitable --format json` 批量发现；最后以 `dws aitable <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "aitable +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws aitable <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service aitable --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-calendar/SKILL.md

@@ -20,7 +20,7 @@ metadata:
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "calendar +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service calendar --format json` 批量发现；最后以 `dws calendar <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "calendar +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws calendar <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service calendar --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-chat/SKILL.md

@@ -18,109 +18,11 @@ metadata:
 > 命令参考：[chat.md](references/chat.md)；表情：[chat-emoji-list.md](references/chat-emoji-list.md)；剧本：[01-messaging.md](references/01-messaging.md)。
 
 <!-- VISIBLE_SHORTCUTS_START -->
-## Shortcuts（无专用脚本/recipe 时优先）
+## Shortcut 发现（按需）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "chat +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service chat --format json` 批量发现；最后以 `dws chat <shortcut> --help` 核对当前 Cobra flags。
+`chat` 当前有 97 条公开 shortcut，完整清单保留在 Runtime Catalog 与 Schema，不在高频产品根 Skill 中重复展开。已知意图直接使用下方的优先路由、意图表或任务 reference；命令已选中时直接执行，只在参数/安全语义不确定时读取 leaf Schema，在当前 Cobra flags 不确定时读取 leaf Help。
 
-| Shortcut | 风险 | 适用场景 |
-|---|---|---|
-| `dws chat +at-me` | read | 查最近 @我 的消息（自动算时间窗，投影发送人/时间/内容/会话） |
-| `dws chat +bot-find` | read | 搜索全部可用机器人（含他人/官方，返回 openDingTalkId 可发单聊） |
-| `dws chat +bot-search` | read | 搜索当前用户自己创建的机器人 |
-| `dws chat +broadcast` | write | 按姓名逐一给多个人群发同一条单聊消息（自动解析 userId、逐个发送） |
-| `dws chat +category-add-conversation` | write | 将会话移动到指定的自定义分组中 |
-| `dws chat +category-create` | write | 创建用户自定义会话分组 |
-| `dws chat +category-delete` | high-risk-write | 删除用户自定义会话分组 |
-| `dws chat +category-list` | read | 获取用户自定义会话分组 |
-| `dws chat +category-list-conversations` | read | 拉取指定自定义会话分组下的会话 |
-| `dws chat +category-remove-conversation` | write | 将会话从指定的自定义分组中移出 |
-| `dws chat +category-rename` | write | 更新用户自定义会话分组的名称 |
-| `dws chat +chat-add-bot` | write | 将机器人添加到群中 |
-| `dws chat +chat-audit-join` | write | 审批入群验证（通过/拒绝/删除/忽略/拉黑） |
-| `dws chat +chat-bots` | read | 查看群内所有机器人 |
-| `dws chat +chat-create` | write | 以当前用户身份创建钉钉群聊 |
-| `dws chat +chat-dismiss` | high-risk-write | 解散群聊（不可逆，需群主权限） |
-| `dws chat +chat-get-by-id` | read | 根据群号获取群聊信息 |
-| `dws chat +chat-invite-url` | read | 获取群邀请链接 |
-| `dws chat +chat-list-all` | read | 分页拉取我加入的所有群列表 |
-| `dws chat +chat-list-join-requests` | read | 分页拉取入群验证记录 |
-| `dws chat +chat-list-mine` | read | 拉取我创建/管理的群 |
-| `dws chat +chat-members-get` | read | 根据成员 openDingTalkId 批量查询群成员详情 |
-| `dws chat +chat-members-list` | read | 列出群成员并把用户与机器人分桶（支持群名语义解析） |
-| `dws chat +chat-messages` | read | 拉取某个会话（群聊或单聊）的消息列表并投影出发言人/文本/时间 |
-| `dws chat +chat-mute` | write | 全员禁言 / 取消全员禁言 |
-| `dws chat +chat-mute-member` | write | 指定群成员禁言 / 取消禁言 |
-| `dws chat +chat-quit` | write | 退出群聊 |
-| `dws chat +chat-remove-bot` | high-risk-write | 从群内移除机器人 |
-| `dws chat +chat-role-add` | write | 添加群身份 |
-| `dws chat +chat-role-list` | read | 拉取会话的群身份列表 |
-| `dws chat +chat-role-query-user` | read | 查询群成员的群身份 |
-| `dws chat +chat-role-remove` | high-risk-write | 删除群身份 |
-| `dws chat +chat-role-remove-user` | write | 移除用户的指定群身份 |
-| `dws chat +chat-role-set-user` | write | 设置用户的群身份（覆盖该用户的全部群身份） |
-| `dws chat +chat-role-update` | write | 更新群身份名称 |
-| `dws chat +chat-search` | read | 按关键词搜索群聊 |
-| `dws chat +chat-set-admin` | write | 设置 / 取消群管理员 |
-| `dws chat +chat-set-history` | write | 设置新成员入群可查看历史消息范围 |
-| `dws chat +chat-transfer-owner` | write | 转让群主 |
-| `dws chat +chat-update` | write | 更新群名称（仅名称，不支持 description） |
-| `dws chat +chat-update-alias` | write | 设置群备注（仅自己可见） |
-| `dws chat +chat-update-icon` | write | 更新群头像 |
-| `dws chat +chat-update-nick` | write | 设置当前用户在群内的群昵称 |
-| `dws chat +chat-update-settings` | write | 更新群设置（settingKey + status） |
-| `dws chat +conversation-clear-all-red-point` | write | 清除所有会话红点（全部已读） |
-| `dws chat +conversation-clear-messages` | high-risk-write | 清空当前用户指定会话的聊天记录（仅本人视角，不可逆） |
-| `dws chat +conversation-clear-red-point` | write | 清除会话红点 |
-| `dws chat +conversation-hide` | write | 会话列表中隐藏会话（收到新消息会重新出现） |
-| `dws chat +conversation-info` | read | 获取会话信息（群聊传 --group，单聊传 --open-dingtalk-id） |
-| `dws chat +conversation-list` | read | 分页获取当前用户的全部会话列表（单聊+群聊） |
-| `dws chat +conversation-list-top` | read | 拉取置顶会话列表，可只看群聊或单聊 |
-| `dws chat +conversation-mark-read` | write | 标记消息已读（该消息及之前的消息都标记为已读） |
-| `dws chat +conversation-mark-unread` | write | 标记会话为未读 |
-| `dws chat +conversation-mute` | write | 会话消息免打扰（支持单聊/群聊） |
-| `dws chat +conversation-set-top` | write | 批量会话置顶 / 取消置顶（最多 10 个） |
-| `dws chat +dm` | write | 按姓名直接给某人发单聊消息（自动解析 userId） |
-| `dws chat +feed-group-query-item` | read | 在会话分组结果中按会话 ID 精确查询多项 |
-| `dws chat +flag-cancel` | write | 取消收藏一条或多条消息（最多 10 条） |
-| `dws chat +flag-create` | write | 收藏一条或多条消息（最多 10 条） |
-| `dws chat +flag-list` | read | 分页查询当前用户收藏的消息 |
-| `dws chat +group-members` | read | 按群名列出群成员（自动搜群解析 openConversationId） |
-| `dws chat +messages-add-emoji` | write | 对消息添加 emoji 表情回应 |
-| `dws chat +messages-add-text-emotion` | write | 对消息添加文字表情回应 |
-| `dws chat +messages-batch-recall-by-bot` | write | 机器人撤回单聊消息 |
-| `dws chat +messages-batch-send-by-bot` | write | 机器人批量向用户发送单聊 Markdown 消息 |
-| `dws chat +messages-combine-forward` | write | 合并转发多条消息 |
-| `dws chat +messages-create-text-emotion` | write | 创建文字表情（获取 emotionId） |
-| `dws chat +messages-forward` | write | 转发单条消息 |
-| `dws chat +messages-forward-topic` | write | 转发话题消息到目标会话 |
-| `dws chat +messages-list` | read | 拉取群聊会话消息 |
-| `dws chat +messages-list-direct` | read | 拉取单聊会话消息 |
-| `dws chat +messages-list-pin` | read | 拉取会话中钉住的消息列表 |
-| `dws chat +messages-list-unread-conversations` | read | 获取有未读消息的会话列表 |
-| `dws chat +messages-mget` | read | 根据消息 ID 批量查询消息（最多 50 条） |
-| `dws chat +messages-query-send-status` | read | 查询消息发送状态 |
-| `dws chat +messages-read-status` | read | 查询消息的已读/未读状态 |
-| `dws chat +messages-recall` | write | 撤回当前用户发送的消息 |
-| `dws chat +messages-recall-by-bot` | write | 机器人撤回群消息 |
-| `dws chat +messages-remove-emoji` | write | 移除消息的 emoji 表情回应 |
-| `dws chat +messages-remove-text-emotion` | write | 移除消息的文字表情回应 |
-| `dws chat +messages-reply` | write | 以当前用户身份引用回复消息（自动补全原发送者） |
-| `dws chat +messages-resource-download` | read | 安全下载消息资源（图片/视频/语音/文件）到工作目录内，无需交互确认 |
-| `dws chat +messages-resource-url` | read | 获取消息资源（图片/视频/语音）下载链接 |
-| `dws chat +messages-send` | write | 统一发送文本、Markdown、当前用户文件或已有 mediaId 图片 |
-| `dws chat +messages-send-by-bot` | write | 机器人向群聊发送 Markdown 消息 |
-| `dws chat +messages-send-by-webhook` | write | 自定义机器人 Webhook 发送群消息 |
-| `dws chat +messages-send-card` | write | 创建流式卡片，可在同一次调用中写入内容并结束 |
-| `dws chat +messages-set-pin` | write | 钉住消息（Pin） |
-| `dws chat +messages-set-top` | write | 置顶消息 |
-| `dws chat +messages-unset-pin` | write | 取消钉住消息（Unpin） |
-| `dws chat +messages-unset-top` | write | 取消置顶消息 |
-| `dws chat +messages-update-card` | write | 流式更新卡片内容（最后一次 --flow-status 应为 3） |
-| `dws chat +my-groups` | read | 列出我加入的群，可按类型过滤并投影关键字段 |
-| `dws chat +search-msg` | read | 多维搜索消息，可全量翻页并批量富化详情 |
-| `dws chat +send-to-group` | write | 按群名直接给群发消息（自动搜群解析 openConversationId） |
-| `dws chat +thread-replies` | read | 拉取某条话题消息的全部回复并投影出发言人/文本/时间 |
-| `dws chat +unread-chats` | read | 列出我有未读消息的会话（投影会话名/未读数/会话ID） |
+仅当现有路由和 reference 都无法定位低频能力时，才执行 `dws shortcut list --service chat --format json` 做最后回退；不要为已知高频意图加载完整 Shortcut Catalog 或产品级 Schema。
 <!-- VISIBLE_SHORTCUTS_END -->
 
 ## IM Shortcut 优先路由

skills/multi/dingtalk-contact/SKILL.md

@@ -20,7 +20,7 @@ metadata:
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "contact +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service contact --format json` 批量发现；最后以 `dws contact <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "contact +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws contact <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service contact --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-dev/SKILL.md

@@ -29,7 +29,7 @@ metadata:
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "devapp +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service devapp --format json` 批量发现；最后以 `dws devapp <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "devapp +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws devapp <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service devapp --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-doc/SKILL.md

@@ -29,7 +29,7 @@ metadata:
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "doc +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service doc --format json` 批量发现；最后以 `dws doc <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "doc +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws doc <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service doc --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-drive/SKILL.md

@@ -20,7 +20,7 @@ metadata:
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "drive +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service drive --format json` 批量发现；最后以 `dws drive <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "drive +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws drive <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service drive --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-mail/SKILL.md

@@ -20,7 +20,7 @@ metadata:
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "mail +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service mail --format json` 批量发现；最后以 `dws mail <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "mail +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws mail <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service mail --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-minutes/SKILL.md

@@ -20,7 +20,7 @@ metadata:
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "minutes +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service minutes --format json` 批量发现；最后以 `dws minutes <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "minutes +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws minutes <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service minutes --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-misc/references/attendance.md

@@ -1411,7 +1411,7 @@ dws attendance boss-check --plan-id 948964045503 --time "2026-05-13 18:00" --res
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "attendance +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service attendance --format json` 批量发现；最后以 `dws attendance <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "attendance +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws attendance <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service attendance --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-misc/references/ding.md

@@ -169,7 +169,7 @@ dws ding message recall-personal --id <OPEN_DING_ID> --format json
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "ding +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service ding --format json` 批量发现；最后以 `dws ding <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "ding +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws ding <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service ding --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-misc/references/oa.md

@@ -473,7 +473,7 @@ dws oa approval revert-task --instance-id <processInstanceId> --task-id <taskId>
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "oa +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service oa --format json` 批量发现；最后以 `dws oa <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "oa +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws oa <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service oa --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-misc/references/report.md

@@ -445,7 +445,7 @@ dws report outbox list --cursor 0 --size 20 --format json
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "report +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service report --format json` 批量发现；最后以 `dws report <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "report +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws report <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service report --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-misc/references/sheet.md

@@ -210,7 +210,7 @@ Flags:
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "sheet +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service sheet --format json` 批量发现；最后以 `dws sheet <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "sheet +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws sheet <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service sheet --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-todo/SKILL.md

@@ -20,7 +20,7 @@ metadata:
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "todo +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service todo --format json` 批量发现；最后以 `dws todo <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "todo +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws todo <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service todo --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|

skills/multi/dingtalk-wiki/SKILL.md

@@ -20,7 +20,7 @@ metadata:
 <!-- VISIBLE_SHORTCUTS_START -->
 ## Shortcuts（无专用脚本/recipe 时优先）
 
-以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。用 leaf Schema（例如 `dws schema --cli-path "wiki +<shortcut>" --format json`）读取 Agent 选择、参数、约束、风险和确认语义；用 `dws shortcut list --service wiki --format json` 批量发现；最后以 `dws wiki <shortcut> --help` 核对当前 Cobra flags。
+以下 shortcut 同时进入公开 catalog 与 Runtime Schema。先按本 skill 的意图表、脚本和 recipe 路由：存在精确覆盖该场景的专用脚本/recipe 时按其执行；否则用户意图命中时，shortcut 优先于手写原子命令。命令已选中时直接执行；只在参数或安全语义不确定时读取 leaf Schema（例如 `dws schema --cli-path "wiki +<shortcut>" --format json`），在当前 Cobra flags 不确定时读取 `dws wiki <shortcut> --help`。仅当现有路由和 reference 都无法定位低频能力时，才用 `dws shortcut list --service wiki --format json` 批量发现。
 
 | Shortcut | 风险 | 适用场景 |
 |---|---|---|