universal-dev-standards 6.4.0 → 6.6.0

package/bundled/locales/zh-TW/core/verification-evidence.md

@@ -1,8 +1,8 @@
 ---
 source: ../../../core/verification-evidence.md
-source_version: 1.2.0
-translation_version: 1.2.0
-last_synced: 2026-07-17
+source_version: 1.3.0
+translation_version: 1.3.0
+last_synced: 2026-08-14
 status: current
 ---
 
@@ -10,8 +10,8 @@ status: current
 
 > **語言**: [English](../../../core/verification-evidence.md) | 繁體中文
 
-**版本**: 1.2.0
-**最後更新**: 2026-07-17
+**版本**: 1.3.0
+**最後更新**: 2026-08-14
 **適用性**: 所有 AI 輔助開發工作流
 **範圍**: 通用 (Universal)
 **靈感來源**: [Superpowers](https://github.com/obra/superpowers) — verification-before-completion (MIT)
@@ -40,6 +40,7 @@ status: current
 | 環境層次 (Environment Layer) | 該證據是從哪個環境蒐集而來（`local` / `uat` / `prd`） |
 | 證據有效性 (Evidence Validity) | 證據*本身*是否可信 —— 亦即驗證指令究竟有沒有真的執行、有沒有真的量到它宣稱量到的東西 |
 | 工具靜默失敗 (Silent Tool Failure) | 驗證指令根本沒跑起來，或跑了但什麼都沒量到，卻產生了與真實結果無從分辨的輸出 |
+| 證據新鮮度 (Evidence Freshness) | 證據是否由**最後一次編輯受測程式碼、提示詞或設定之後**才執行的一次執行所產生 —— 而不是沿用一次針對更早狀態的執行結果 |
 
 ---
 
@@ -169,7 +170,7 @@ status: current
 
 這不是幻覺。幻覺是虛構你沒查過的事。這是反過來：**你查了，而那個查詢工具騙了你。** `anti-hallucination` 標準涵蓋不到這一塊 —— 該標準的每一條禁令都是某種形式的「不要編造」，而這裡沒有任何東西被編造。
 
-### 四條有效性規則
+### 五條有效性規則
 
 **1. 只有在「成功時回傳 0」的工具上，`exit_code = 0` 才代表成功。**
 
@@ -187,6 +188,16 @@ status: current
 
 在 `set -o pipefail` 下，`producer | grep -q pattern` 會繼承 `producer` 的非零值，與 `grep` 有沒有匹配到無關。當決策取決於內容時，**先捕捉輸出，再評估它** —— 不要讓一個 pipeline 把兩個問題壓縮成一個數字。
 
+**5. 證據必須晚於它所驗證對象的最後一次編輯。**
+
+在受測程式碼、提示詞或設定的**最近一次變更之前**擷取的驗證執行，證明的是別的東西 —— 不是正在被主張的那件事。這個陷阱是靜默的：完整套件先跑過且全綠，之後才編輯了某個東西（提示詞、設定檔、門檻），而編輯之後只跑了範圍更窄的檢查（lint、部分套件）—— 結果卻拿更早的那次綠燈當作現狀的證據。**證據必須是最後一次編輯之後的單一次新鮮執行，不是「剛好通過的最近一次執行」。**
+
+### 過期證據 —— 一種失效樣態
+
+規則 1–4 假設證據**在正確的地方**被蒐集且被正確解讀。它們沒有回答「**什麼時候**」這個問題。一個執行了、量測正確、也回傳真結果的指令，依然可能撐不起一個完成聲明 —— 如果它跑完之後有東西又變了。
+
+**失效樣態**：完整測試套件跑過且通過。之後，某個提示詞或設定值被編輯 —— 沒有任何測試直接涵蓋那次編輯。提交前只跑了 `lint`，也通過。commit 引用「測試通過」作為證據。接著 CI 紅在一個釘住剛剛被改掉那份內容的測試上 —— 那個「通過」的套件，跑的是一個在主張被提出時已經不存在的版本。
+
 ### 此失效模式的實例證據
 
 以下是某 AI 代理於 2026-07-17 實際執行的驗證指令，每一道都導出了自信而錯誤的結論，並被當成事實回報出去：
@@ -216,11 +227,24 @@ status: current
 | 有 `verification_evidence` 但 `exit_code ≠ 0` | 標記為**驗證失敗** —— **除非**已知該工具在受測狀態下本就會以非零退出（證據有效性規則 1），此時改以輸出判定 |
 | `exit_code = 0`，但該指令不可能量到它所宣稱的事 | 標記為**未驗證** —— 一道在錯的地方執行、或根本沒執行的通過指令，不是證據 |
 | 證據主張「不存在」（`0`、空輸出、「查無」） | 在證明查詢工具成功執行過之前，標記為**未驗證** |
+| 證據的時間戳早於它所驗證對象的最後一次編輯 | 標記為**過期（stale）** —— 須在編輯後重跑，不得引用更早那次通過 |
 | 多個驗證步驟 | **全部**步驟都必須通過 |
 | 代理提供的是錯誤指令的證據 | 標記為**未驗證** |
 
 ---
 
+## 窄涵蓋必須登記，不能只揭露
+
+Evidence Validity 與 Environment Stratification Matrix 都允許一個被記錄下來的缺口 —— 一個無法驗證某個維度的環境層次、一項量測範圍窄於它所支撐主張的檢查。單純寫下這個缺口本身並不夠：散文比真正補上缺口便宜，而一個從未被重新檢視的揭露，會悄悄變成「這份證據其實撐不起它被用來支撐的主張」的永久藉口。
+
+**要求**：任何這一類已記錄的缺口，必須同時登記到一份**帶到期日的例外清冊** —— 獨立於標準本文或 matrix 條目本身之外 —— 指名缺口、其成因、並附上覆核或到期日期。只有一則註腳而沒有對應清冊條目，不滿足本條。
+
+**可證偽條件**：一則清冊條目連續兩期審查都未變動，代表該揭露已經變成逃生口，本條對該條目失效 —— 不是「部分滿足」，是失效。清冊放在哪裡、什麼格式、多久審一次，留給採用專案自行決定；本標準只要求「有這麼一份東西存在，且條目會動」。
+
+這與[類別層修正標準](class-level-fix.md)的窄涵蓋規則是同一條要求，套用在驗證證據而非類別層檢查上：兩種情況下，一個永遠不必改變的揭露都不是揭露，是一份穿著揭露外衣的永久豁免。
+
+---
+
 ## 規則
 
 | ID | 觸發條件 | 動作 | 優先級 |
@@ -235,6 +259,8 @@ status: current
 | VE-008 | 證據主張「不存在」（`0`、空輸出、「查無」） | 在證明查詢工具成功執行過之前一律無效。以不壓抑 stderr 的方式重跑 | High |
 | VE-009 | 存在性／不存在性檢查壓抑了 stderr（`2>/dev/null` 或等價寫法） | 證據不成立。以 stderr 可見的方式重跑 | High |
 | VE-010 | 證據的 `exit_code` 來自 pipeline（尤其是在 `pipefail` 下） | 該 code 不歸屬於任何單一階段。改為捕捉輸出並評估內容 | Medium |
+| VE-011 | 證據的執行時間戳早於它所宣稱驗證之產物的最後一次編輯 | 標記為**過期（stale）** —— 對現狀不成立為證據；要求在編輯後補一次單一次新鮮執行 | High |
+| VE-012 | 文件化的證據/涵蓋缺口（environment-stratification ⚠️/❌，或任何窄於其主張的檢查）沒有登記到帶到期日的例外清冊 | 揭露本身不滿足此規則 —— 要求一則已登記、帶到期日的清冊條目，指名缺口與其覆核/到期日期 | Required |
 
 ---
 
@@ -314,6 +340,7 @@ verification_evidence:
 - [測試標準](testing-standards.md) —— 產生多數證據的那些測試是怎麼寫的
 - [代理派遣與並行協調](agent-dispatch.md) —— 委派出去的工作會回傳主張，那些主張需要套用本標準
 - [部署標準](deployment-standards.md) —— 定義 VE-005 / VE-006 所引用的「環境分層責任矩陣」
+- [類別層修正標準](class-level-fix.md) —— 與 VE-012 相同的「窄涵蓋必須登記」要求，套用在類別層檢查而非驗證證據上
 
 ---
 

package/bundled/locales/zh-TW/docs/CHEATSHEET.md

@@ -1,6 +1,6 @@
 # UDS 速查表
 
-> Quick reference for all UDS features | Last updated: 2026-08-10
+> Quick reference for all UDS features | Last updated: 2026-08-12
 
 **Language**: [English](../../../docs/user/CHEATSHEET.md) | 繁體中文 | [简体中文](../../zh-CN/docs/CHEATSHEET.md)
 
@@ -240,7 +240,7 @@
 | `logging-standards` | Logging Standards |
 | `mock-boundary` | This document defines rules for what can and canno |
 | `model-provenance` | Model Provenance Policy Standards |
-| `model-selection` | Define a cost-effective strategy for selecting AI  |
+| `model-selection` | Define how to choose **which model** and **how dee |
 | `multi-environment-e2e-testing` | Multi-Environment E2E Testing Standards |
 | `mutation-testing` | Mutation testing evaluates test suite effectivenes |
 | `no-cicd-deployment` | No-CI/CD Deployment Strategy |
@@ -319,27 +319,22 @@
 | `aggregate-effectiveness.mjs` | Aggregate Standards Effectiveness Reports |
 | `analyze-hook-stats.mjs` | Hook Statistics Analyzer (SPEC-SELFDIAG-001 REQ-7, |
 | `bump-version.mjs` | Build a platform-aware shell command for a .sh scr |
-| `bump-version.sh` | DEPRECATED: Use 'node scripts/bump-version.mjs <ve |
+| `bump-version.sh` | Thin wrapper — scripts/bump-version.mjs is the onl |
 | `check-ai-agent-sync.ps1` | Check Ai Agent Sync |
 | `check-ai-agent-sync.sh` | AI Agent Sync Checker |
-| `check-ai-behavior-sync.sh` | DEPRECATED: Use 'npx tsx scripts/check-ai-behavior |
 | `check-ai-yaml-parses.mjs` | Every shipped .ai.yaml must parse, and must parse  |
 | `check-cli-docs-sync.ps1` | Check Cli Docs Sync |
 | `check-cli-docs-sync.sh` | CLI-to-Documentation Sync Checker |
 | `check-commands-sync.ps1` | Check Commands Sync |
 | `check-commands-sync.sh` | Commands Sync Checker |
-| `check-commit-spec-reference.sh` | DEPRECATED: Use 'npx tsx scripts/check-commit-spec |
+| `check-commit-spec-reference.sh` | Thin wrapper — scripts/check-commit-spec-reference |
 | `check-docs-integrity.ps1` | Check Docs Integrity |
 | `check-docs-integrity.sh` | Documentation Integrity Checker |
 | `check-docs-sync.ps1` | Check Docs Sync |
 | `check-docs-sync.sh` | Documentation Sync Checker |
 | `check-external-references.mjs` | External Reference Checker (SPEC-SELFDIAG-001 REQ- |
-| `check-flow-gate-report.sh` | DEPRECATED: Use 'npx tsx scripts/check-flow-gate-r |
-| `check-integration-commands-sync.sh` | DEPRECATED: Use 'npx tsx scripts/check-integration |
 | `check-orphan-specs.ps1` | Check Orphan Specs |
 | `check-orphan-specs.sh` | Orphan Spec Detection Script |
-| `check-registry-completeness.sh` | DEPRECATED: Use 'npx tsx scripts/check-registry-co |
-| `check-release-readiness-signoff.sh` | DEPRECATED: Use 'npx tsx scripts/check-release-rea |
 | `check-scope-sync.ps1` | Check Scope Sync |
 | `check-scope-sync.sh` | Scope Consistency Check Script |
 | `check-skill-next-steps-sync.ps1` | Check Skill Next Steps Sync |
@@ -356,16 +351,16 @@
 | `check-usage-docs-sync.sh` | check-usage-docs-sync.sh |
 | `check-version-sync.ps1` | Check Version Sync |
 | `check-version-sync.sh` | Version Sync Checker |
-| `check-workflow-compliance.sh` | DEPRECATED: Use 'npx tsx scripts/check-workflow-co |
+| `check-workflow-compliance.sh` | Thin wrapper — scripts/check-workflow-compliance.t |
 | `commitlint-bilingual-rule.mjs` | commitlint-bilingual-rule.mjs — custom commitlint  |
 | `convert-md-to-yaml.mjs` | Markdown to AI-YAML Conversion Script |
 | `fix-manifest-paths.ps1` | Fix Manifest Paths |
 | `fix-manifest-paths.sh` | Manifest Path Fixer |
-| `generate-docs.mjs` | Sync the "AI Tool Support" table's Skills/Slash Co |
+| `generate-docs.mjs` | Look up the release date for `version` from CHANGE |
 | `generate-locale-coverage.mjs` | Locale Coverage Generator |
 | `generate-version-manifest.mjs` | Generate Version Manifest (SPEC-SELFDIAG-001 REQ-9 |
 | `install-hooks.mjs` | Install Hooks |
-| `install-hooks.sh` | DEPRECATED: Use 'node scripts/install-hooks.mjs' i |
+| `install-hooks.sh` | Thin wrapper — scripts/install-hooks.mjs is the on |
 | `pre-commit.mjs` | Build a platform-aware shell command for a .sh scr |
 | `pre-release-check.ps1` | Pre Release Check |
 | `pre-release-check.sh` | Pre-release Check Script |

package/bundled/locales/zh-TW/docs/FEATURE-REFERENCE.md

@@ -1,7 +1,7 @@
 # UDS 功能參考手冊
 
 > Universal Development Standards - 完整功能文件
-> Auto-generated | Last updated: 2026-08-10
+> Auto-generated | Last updated: 2026-08-12
 
 **Language**: [English](../../../docs/reference/FEATURE-REFERENCE.md) | 繁體中文 | [简体中文](../../zh-CN/docs/FEATURE-REFERENCE.md)
 
@@ -15,9 +15,9 @@
 4. [代理](#agents) (5)
 5. [工作流程](#workflows) (5)
 6. [核心規範](#core-standards) (150)
-7. [腳本](#scripts) (59)
+7. [腳本](#scripts) (54)
 
-**Total Features: 334**
+**Total Features: 329**
 
 ---
 
@@ -367,7 +367,7 @@
 | `logging-standards` | 1.4.0 |  |
 | `mock-boundary` | 1.1.0 | This document defines rules for what can and cannot be mocked in tests. Its goal |
 | `model-provenance` | 1.0.0 | **Status**: Active | **Updated**: 2026-06-17 |  |
-| `model-selection` | 1.0.1 | Define a cost-effective strategy for selecting AI model tiers based on task comp |
+| `model-selection` | 2.1.0 | Define how to choose **which model** and **how deeply it should think** — two in |
 | `multi-environment-e2e-testing` | 1.0.0 | **Status**: Active | **Updated**: 2026-05-13 |  |
 | `mutation-testing` | 1.0.0 | Mutation testing evaluates test suite effectiveness by injecting artificial bugs |
 | `no-cicd-deployment` | - |  |
@@ -428,7 +428,7 @@
 | `testing-standards` | 3.2.0 | This standard defines actionable testing rules and conventions for AI agents and |
 | `timeout-standards` | - |  |
 | `token-budget` | - |  |
-| `translation-lifecycle-standards` | 1.0.0 | Translation lifecycle standards: MISSING vs OUTDATED distinction, semver-aware s |
+| `translation-lifecycle-standards` | 1.0.1 | Translation lifecycle standards: MISSING vs OUTDATED distinction, semver-aware s |
 | `user-journey-testing` | - |  |
 | `user-story-mapping` | 1.0.0 | **Status**: Active | **Updated**: 2026-06-17 |  |
 | `verification-evidence` | 1.2.0 | Establish an "Iron Law" that no task can be claimed as complete without verifica |
@@ -448,27 +448,22 @@
 | `aggregate-effectiveness.mjs` | Aggregate Standards Effectiveness Reports |
 | `analyze-hook-stats.mjs` | Hook Statistics Analyzer (SPEC-SELFDIAG-001 REQ-7, AC-11) |
 | `bump-version.mjs` | Build a platform-aware shell command for a .sh script. |
-| `bump-version.sh` | DEPRECATED: Use 'node scripts/bump-version.mjs <version>' instead (cross-platform). |
+| `bump-version.sh` | Thin wrapper — scripts/bump-version.mjs is the only copy of the bump logic. |
 | `check-ai-agent-sync.ps1` | Check Ai Agent Sync |
 | `check-ai-agent-sync.sh` | AI Agent Sync Checker |
-| `check-ai-behavior-sync.sh` | DEPRECATED: Use 'npx tsx scripts/check-ai-behavior-sync.ts' instead (cross-platform). |
 | `check-ai-yaml-parses.mjs` | Every shipped .ai.yaml must parse, and must parse into what it says. |
 | `check-cli-docs-sync.ps1` | Check Cli Docs Sync |
 | `check-cli-docs-sync.sh` | CLI-to-Documentation Sync Checker |
 | `check-commands-sync.ps1` | Check Commands Sync |
 | `check-commands-sync.sh` | Commands Sync Checker |
-| `check-commit-spec-reference.sh` | DEPRECATED: Use 'npx tsx scripts/check-commit-spec-reference.ts' instead (cross-platform). |
+| `check-commit-spec-reference.sh` | Thin wrapper — scripts/check-commit-spec-reference.ts is the only copy of |
 | `check-docs-integrity.ps1` | Check Docs Integrity |
 | `check-docs-integrity.sh` | Documentation Integrity Checker |
 | `check-docs-sync.ps1` | Check Docs Sync |
 | `check-docs-sync.sh` | Documentation Sync Checker |
 | `check-external-references.mjs` | External Reference Checker (SPEC-SELFDIAG-001 REQ-5, AC-7) |
-| `check-flow-gate-report.sh` | DEPRECATED: Use 'npx tsx scripts/check-flow-gate-report.ts' instead (cross-platform). |
-| `check-integration-commands-sync.sh` | DEPRECATED: Use 'npx tsx scripts/check-integration-commands-sync.ts' instead (cross-platform). |
 | `check-orphan-specs.ps1` | Check Orphan Specs |
 | `check-orphan-specs.sh` | Orphan Spec Detection Script |
-| `check-registry-completeness.sh` | DEPRECATED: Use 'npx tsx scripts/check-registry-completeness.ts' instead (cross-platform). |
-| `check-release-readiness-signoff.sh` | DEPRECATED: Use 'npx tsx scripts/check-release-readiness-signoff.ts' instead (cross-platform). |
 | `check-scope-sync.ps1` | Check Scope Sync |
 | `check-scope-sync.sh` | Scope Consistency Check Script |
 | `check-skill-next-steps-sync.ps1` | Check Skill Next Steps Sync |
@@ -485,16 +480,16 @@
 | `check-usage-docs-sync.sh` | check-usage-docs-sync.sh |
 | `check-version-sync.ps1` | Check Version Sync |
 | `check-version-sync.sh` | Version Sync Checker |
-| `check-workflow-compliance.sh` | DEPRECATED: Use 'npx tsx scripts/check-workflow-compliance.ts' instead (cross-platform). |
+| `check-workflow-compliance.sh` | Thin wrapper — scripts/check-workflow-compliance.ts is the only copy of the |
 | `commitlint-bilingual-rule.mjs` | commitlint-bilingual-rule.mjs — custom commitlint rules enforcing the |
 | `convert-md-to-yaml.mjs` | Markdown to AI-YAML Conversion Script |
 | `fix-manifest-paths.ps1` | Fix Manifest Paths |
 | `fix-manifest-paths.sh` | Manifest Path Fixer |
-| `generate-docs.mjs` | Sync the "AI Tool Support" table's Skills/Slash Commands numeric columns |
+| `generate-docs.mjs` | Look up the release date for `version` from CHANGELOG.md's own |
 | `generate-locale-coverage.mjs` | Locale Coverage Generator |
 | `generate-version-manifest.mjs` | Generate Version Manifest (SPEC-SELFDIAG-001 REQ-9, AC-14) |
 | `install-hooks.mjs` | Install Hooks |
-| `install-hooks.sh` | DEPRECATED: Use 'node scripts/install-hooks.mjs' instead (cross-platform). |
+| `install-hooks.sh` | Thin wrapper — scripts/install-hooks.mjs is the only copy of the installer |
 | `pre-commit.mjs` | Build a platform-aware shell command for a .sh script. |
 | `pre-release-check.ps1` | Pre Release Check |
 | `pre-release-check.sh` | Pre-release Check Script |

package/bundled/locales/zh-TW/integrations/claude-code/README.md

@@ -1,8 +1,8 @@
 ---
 source: ../../../../integrations/claude-code/README.md
-source_version: 1.0.0
-translation_version: 1.0.0
-last_synced: 2026-04-22
+source_version: 1.1.0
+translation_version: 1.1.0
+last_synced: 2026-08-12
 status: current
 ---
 
@@ -10,8 +10,8 @@ status: current
 
 > **語言**: English | [繁體中文](README.md)
 
-**版本**: 1.0.0
-**最後更新**: 2026-01-29
+**版本**: 1.1.0
+**最後更新**: 2026-08-12
 
 本目錄包含將通用開發標準 (Universal Development Standards) 與 [Claude Code](https://docs.anthropic.com/claude-code) 整合的資源。
 
@@ -21,6 +21,32 @@ Claude Code 是一個先進的 AI 編碼代理，可以直接與您的程式碼
 
 1.  **專案上下文 (`CLAUDE.md`)**：定義專案特定的規則、風格指南和指令。
 2.  **技能 (`skills/`)**：針對 TDD、SDD、程式碼審查等的專業能力。
+3.  **模型層級 × effort 映射**：`core/model-selection.md` 的宿主層那一半。
+4.  **參考 subagent 定義**：`.claude/agents/*.md`，每個模型層級各一份。
+5.  **跨 repo 派工模板**：當目標 repo 不是你的 session 所在的那一個時，subagent 的 prompt 必須自帶什麼。
+
+## 本宿主的模型選擇（XSPEC-362 R5）
+
+`core/model-selection.md` 依規則保持 vendor-neutral——它把 tier（`fast` / `standard` / `capable`）
+與 effort 級距（`low` … `max`）定義為標籤，並明白表示它無法斷言某個模型接受哪些 effort 級距。
+下列檔案為 Claude Code 回答這件事，且是本 repo **唯一**應該出現具體模型識別字的地方。
+
+| 檔案 | 內容 |
+|---|---|
+| [`model-selection-mapping.md`](../../../../integrations/claude-code/model-selection-mapping.md) | tier → `model`、effort → `effort`、已解析的 tier × effort 表格，以及硬邊界登記表 |
+| [`model-selection-mapping.ai.yaml`](../../../../integrations/claude-code/model-selection-mapping.ai.yaml) | 同一份映射的機器可讀版 |
+| [`.claude/agents/`](../../../../integrations/claude-code/.claude/agents/) | 四份參考 subagent 定義：三層各一，外加一個長時程變體 |
+| [`dispatch-template.md`](../../../../integrations/claude-code/dispatch-template.md) | 跨 repo 派工模板，以及支撐它的實測 |
+
+該映射有兩件事值得先知道：
+
+- **UDS 的 `very-high` 在 Claude Code 寫作 `xhigh`。** 這是唯一名稱不同的一級。標準不改名去遷就工具，
+  改名由宿主層吸收。
+- **`fast` 層在本宿主沒有 effort 軸。** 該層的模型不接受任何 effort 級距，
+  所以標準的「先調 effort 再升級模型層」在那裡沒有東西可調。
+
+派工機制——並行安全、獨立域、狀態協定——**不**由這些檔案涵蓋。
+那屬於 [`core/agent-dispatch.md`](../../../../core/agent-dispatch.md)，這些檔案引用它而不重寫它。
 
 ## 設定
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "universal-dev-standards",
-  "version": "6.4.0",
+  "version": "6.6.0",
   "description": "CLI tool for adopting Universal Development Standards",
   "keywords": [
     "documentation",

package/src/utils/reference-sync.js

@@ -75,18 +75,35 @@ export const STANDARD_TO_CATEGORY = {
  * Matches patterns like:
  * - Reference: .standards/anti-hallucination.md
  * - 參考: .standards/anti-hallucination.md (Chinese)
+ * - Reference: .standards/a.ai.yaml, .standards/options/b.ai.yaml (more than one)
+ * - Reference: `.standards/a.ai.yaml` (quoted, as markdown renders it)
+ *
+ * The first version anchored `.standards/` directly after the label and ran to
+ * the first space, which made it see EXACTLY ONE path per line, keep whatever
+ * punctuation followed it, and miss any path that was quoted. Measured on the
+ * two telemetry repos: of three references it reported, one was a false
+ * positive (`commit-message.ai.yaml,` — the file exists, the comma did not),
+ * and it silently skipped two more, one of which
+ * (`workflow-enforcement.ai.yaml`) does not exist in any repo or upstream.
  *
  * @param {string} content - Integration file content
  * @returns {string[]} - Array of referenced standard filenames (e.g., ['anti-hallucination.md'])
  */
 export function parseReferences(content) {
-  // Match both English and Chinese reference patterns
-  const referencePattern = /(?:Reference|參考):\s*\.standards\/([^\s\n)]+)/gi;
+  // Both English and Chinese labels, and both colon widths.
+  const referenceLinePattern = /(?:Reference|參考)[:：][^\n]*/gi;
+  // Every `.standards/` path on that line, stopping at anything that quotes or
+  // separates rather than names: whitespace, brackets, backticks, quotes,
+  // commas and semicolons.
+  const pathPattern = /\.standards\/([^\s\n)\]`,;'"]+)/g;
   const references = new Set();
-  let match;
 
-  while ((match = referencePattern.exec(content)) !== null) {
-    references.add(match[1]);
+  for (const line of content.match(referenceLinePattern) ?? []) {
+    for (const [, path] of line.matchAll(pathPattern)) {
+      // Sentence punctuation is not part of a filename.
+      const cleaned = path.replace(/[.,;:!?]+$/, '');
+      if (cleaned) references.add(cleaned);
+    }
   }
 
   return Array.from(references);
@@ -113,6 +130,42 @@ export function getStandardCategory(sourcePath) {
   return STANDARD_TO_CATEGORY[fileName] || null;
 }
 
+/**
+ * Reduce a standard to the identity all three of its spellings share.
+ *
+ * The same standard appears as a manifest stem (`commit-message`), a shipped
+ * file (`commit-message.ai.yaml`) and a human-readable reference
+ * (`.standards/commit-message.md`). Comparing any two of those literally
+ * fails, and it fails as "this standard is not adopted".
+ *
+ * @param {string} pathOrName - Any of the three spellings
+ * @returns {string} - Bare stem, e.g. 'commit-message'
+ */
+export function standardStem(pathOrName) {
+  return pathOrName
+    .split('/')
+    .pop()
+    .replace(/\.(ai\.yaml|yaml|md)$/i, '');
+}
+
+/**
+ * Category for a standard, accepting a manifest stem as well as a filename.
+ *
+ * {@link STANDARD_TO_CATEGORY} is keyed by filename; manifests store stems.
+ * Looking a stem up directly always missed, which is what emptied
+ * `manifestCategories`.
+ *
+ * @param {string} pathOrName - Manifest stem, filename or reference path
+ * @returns {string|null} - Category ID, or null when the table does not cover it
+ */
+function categoryForStandard(pathOrName) {
+  const fileName = pathOrName.split('/').pop();
+  if (STANDARD_TO_CATEGORY[fileName]) return STANDARD_TO_CATEGORY[fileName];
+
+  const stem = standardStem(pathOrName);
+  return STANDARD_TO_CATEGORY[`${stem}.md`] || STANDARD_TO_CATEGORY[`${stem}.ai.yaml`] || null;
+}
+
 /**
  * Compare manifest standards with integration file references
  *
@@ -124,27 +177,44 @@ export function getStandardCategory(sourcePath) {
  *   - syncedRefs: Standards that are properly synced
  */
 export function compareStandardsWithReferences(manifestStandards, integrationReferences) {
-  // Compare at category level to handle .md vs .ai.yaml format differences
-  // e.g., manifest has 'anti-hallucination.ai.yaml' but integration references 'anti-hallucination.md'
+  // The manifest is the source of truth for what this project adopted, so ask
+  // it directly. The category map below is a hand-written table covering a
+  // small fraction of the standards UDS ships (compare its size against
+  // `cli/standards-registry.json` — a count written here would go stale the
+  // next time a standard lands); using it as the primary test reported every
+  // standard outside the table as "not in manifest", which is a different
+  // claim, and a false one.
+  //
+  // It was worse than partial coverage: manifest entries are bare stems
+  // (`commit-message`) while the table is keyed by filename
+  // (`commit-message.md`), so `manifestCategories` came out EMPTY on the
+  // current manifest format and every reference fell through to orphaned.
+  // Both halves were wrong at once, which is why the output still looked
+  // plausible — three orphans reported on the telemetry repos, one of which
+  // (`commit-message.ai.yaml`) is adopted, and the only genuinely dead one
+  // (`workflow-enforcement.ai.yaml`) was not among them.
+  const manifestStems = new Set(manifestStandards.map(standardStem));
+
   const manifestCategories = new Set();
   for (const std of manifestStandards) {
-    const fileName = std.split('/').pop();
-    const category = STANDARD_TO_CATEGORY[fileName];
+    const category = categoryForStandard(std);
     if (category) {
       manifestCategories.add(category);
     }
   }
 
-  // References in integration file but not backed by any manifest standard (by category)
+  // A reference is orphaned when nothing in the manifest answers to it, by
+  // stem (extension- and directory-insensitive) or by legacy category name.
   const orphanedRefs = integrationReferences.filter(ref => {
-    const category = STANDARD_TO_CATEGORY[ref];
+    if (manifestStems.has(standardStem(ref))) return false;
+    const category = categoryForStandard(ref);
     return !category || !manifestCategories.has(category);
   });
 
   // Categories in manifest but not referenced in integration file
   const refCategories = new Set();
   for (const ref of integrationReferences) {
-    const category = STANDARD_TO_CATEGORY[ref];
+    const category = categoryForStandard(ref);
     if (category) {
       refCategories.add(category);
     }
@@ -157,10 +227,7 @@ export function compareStandardsWithReferences(manifestStandards, integrationRef
   });
 
   // Properly synced references
-  const syncedRefs = integrationReferences.filter(ref => {
-    const category = STANDARD_TO_CATEGORY[ref];
-    return category && manifestCategories.has(category);
-  });
+  const syncedRefs = integrationReferences.filter(ref => !orphanedRefs.includes(ref));
 
   return { orphanedRefs, missingRefs, syncedRefs };
 }

package/standards-registry.json

@@ -1,6 +1,6 @@
 {
   "$schema": "https://json-schema.org/draft/2020-12/schema",
-  "version": "6.4.0",
+  "version": "6.6.0",
   "lastUpdated": "2026-05-13",
   "description": "Standards registry for universal-dev-standards with integrated skills and AI-optimized formats",
   "formats": {
@@ -58,14 +58,14 @@
     "standards": {
       "name": "universal-dev-standards",
       "url": "https://github.com/AsiaOstrich/universal-dev-standards",
-      "version": "6.4.0"
+      "version": "6.6.0"
     },
     "skills": {
       "name": "universal-dev-standards",
       "url": "https://github.com/AsiaOstrich/universal-dev-standards",
       "localPath": "skills",
       "rawUrl": "https://raw.githubusercontent.com/AsiaOstrich/universal-dev-standards/main/skills",
-      "version": "6.4.0",
+      "version": "6.6.0",
       "note": "Skills are now included in the main repository under skills/"
     }
   },
@@ -1728,7 +1728,19 @@
       },
       "category": "skill",
       "skillName": null,
-      "description": "Three-tier model selection strategy (fast/standard/capable) with cost-effective escalation"
+      "description": "Two-axis model selection: model tier (reasoning ceiling × specification definiteness) × effort (reasoning depth), plus reverse-exclusion hard boundaries and four-state capability routing"
+    },
+    {
+      "id": "agent-dispatch",
+      "name": "Agent Dispatch & Parallel Coordination",
+      "nameZh": "代理派遣與並行協調",
+      "source": {
+        "human": "core/agent-dispatch.md",
+        "ai": "ai/standards/agent-dispatch.ai.yaml"
+      },
+      "category": "skill",
+      "skillName": null,
+      "description": "Parallel sub-agent dispatch: independent-domain criterion (no shared mutable state), four-state status protocol, conflict detection, prompt design principles"
     },
     {
       "id": "git-worktree",
@@ -2248,7 +2260,7 @@
       "id": "license-compliance",
       "name": "License Compliance Standards",
       "nameZh": "授權合規標準",
-      "version": "6.4.0",
+      "version": "6.6.0",
       "source": {
         "human": "core/license-compliance.md",
         "ai": "ai/standards/license-compliance.ai.yaml"
@@ -2260,7 +2272,7 @@
       "id": "verification-oracle",
       "name": "Verification Oracle Standards",
       "nameZh": "驗證 Oracle 標準",
-      "version": "6.4.0",
+      "version": "6.6.0",
       "source": {
         "human": "core/verification-oracle.md",
         "ai": "ai/standards/verification-oracle.ai.yaml"
@@ -2272,7 +2284,7 @@
       "id": "model-provenance",
       "name": "Model Provenance Policy Standards",
       "nameZh": "模型來源政策標準",
-      "version": "6.4.0",
+      "version": "6.6.0",
       "source": {
         "human": "core/model-provenance.md",
         "ai": "ai/standards/model-provenance.ai.yaml"
@@ -2284,7 +2296,7 @@
       "id": "resource-cost-boundary",
       "name": "Resource / Cost Boundary Declaration Standards",
       "nameZh": "資源／成本邊界宣告標準",
-      "version": "6.4.0",
+      "version": "6.6.0",
       "source": {
         "human": "core/resource-cost-boundary.md",
         "ai": "ai/standards/resource-cost-boundary.ai.yaml"