kenaz 0.0.0-stage → 0.3.2

package/core/.claude/commands/strategy.md

@@ -0,0 +1,141 @@
+# /strategy - 策略智庫團隊
+
+調用 Strategy Team 進行設計思考、方案討論、決策建議。
+
+## 用法
+
+```bash
+/strategy "您的問題或需求"
+```
+
+## 範例
+
+```bash
+# 架構決策
+/strategy "我們應該使用單體架構還是微服務架構？"
+
+# 技術選型
+/strategy "資料庫應該選擇 PostgreSQL 還是 MongoDB？"
+
+# 功能評估
+/strategy "這個 AI 助手功能值得開發嗎？需要投入多少資源？"
+
+# 方案設計
+/strategy "如何設計一個可擴展到千萬用戶的系統？"
+
+# 風險分析
+/strategy "重構代碼庫有哪些潛在風險？如何緩解？"
+```
+
+## 團隊特色
+
+- 🧠 **7 個專業成員**：產品、技術、UX、商業、創新、風險、策略
+- 💬 **成員互相討論**：就像真實的策略會議
+- 🤝 **與您互動**：主動提問、澄清需求
+- 📊 **方案對比**：提供 2-3 個可行方案
+- ⚖️ **權衡分析**：清楚說明每個方案的優缺點
+
+## 輸出
+
+1. **問題分析**
+2. **團隊討論摘要**（成員之間的對話）
+3. **2-3 個方案選項**
+4. **建議報告**（包含推薦方案與理由）
+5. **風險分析與緩解策略**
+6. **下一步行動計劃**
+
+## 執行時間
+
+- 人類預估：8-16 小時（策略會議 + 分析 + 報告）
+- Agent 預估：20-30 分鐘（AI 討論與分析）
+- 效率提升：約 30 倍
+
+## 調用邏輯
+
+當用戶執行 `/strategy` 時：
+
+1. 讀取 Strategy Agent Skill 定義
+2. 啟動策略分析流程
+3. 多視角獨立分析
+4. 向用戶提問（使用 AskUserQuestion）
+5. 設計 2-3 個方案
+6. 提出建議報告
+7. 記錄決策過程
+
+## 團隊成員
+
+| 成員 | 視角 | 關注點 |
+|------|------|--------|
+| Strategy Lead | 整合 | 協調、綜合、達成共識 |
+| Product Strategist | 產品 | 用戶價值、功能優先級 |
+| Technical Architect | 技術 | 可行性、架構設計 |
+| UX Strategist | UX | 用戶體驗、交互設計 |
+| Business Analyst | 商業 | ROI、成本效益 |
+| Innovation Specialist | 創新 | 新技術、突破性方案 |
+| Risk Analyst | 風險 | 風險識別、緩解策略 |
+
+## 適用場景
+
+✅ **推薦使用**：
+- 面對複雜決策（架構、技術選型）
+- 需要多視角分析
+- 需要評估商業價值
+- 需要風險分析
+- 需要創新方案
+
+❌ **不推薦使用**：
+- 已有明確方案（直接用執行團隊）
+- 簡單 Bug 修復
+- 緊急問題（需要快速響應）
+- 日常維護工作
+
+## 與其他團隊的區別
+
+| 特性 | Strategy Team | Developer/Backend Team |
+|------|---------------|----------------------|
+| 職責 | 設計思考、方案建議 | 代碼實現、功能開發 |
+| 輸出 | 建議報告、方案文檔 | 代碼、Git commits |
+| 時間 | 20-30 分鐘 | 5-15 分鐘 |
+| 價值 | 決策支援、風險規避 | 快速執行、交付功能 |
+
+## 範例輸出
+
+```markdown
+# 建議報告：資料庫選型
+
+## Executive Summary
+建議使用 PostgreSQL + Redis 組合方案，平衡性能、可擴展性與成本。
+
+## Options
+
+### 選項 1：PostgreSQL + Redis（推薦）⭐
+優點：成熟穩定、團隊熟悉、成本可控
+缺點：垂直擴展有上限
+ROI：高
+
+### 選項 2：MongoDB + ElasticSearch
+優點：水平擴展能力強
+缺點：學習成本高、複雜查詢較弱
+ROI：中
+
+### 選項 3：Supabase（雲原生）
+優點：運維成本低、快速上線
+缺點：雲服務依賴、長期成本高
+ROI：中
+
+## Recommendation
+推薦選項 1，理由：
+1. 符合團隊技能
+2. 可擴展到千萬級
+3. 成本最優
+
+## Next Steps
+1. 建立 PostgreSQL PoC
+2. 設計 Schema
+3. 建立監控策略
+```
+
+---
+
+**建立日期**: 2026-02-09
+**適用版本**: Claude Code v1.0+

package/core/.claude/commands/sunny.md

@@ -0,0 +1 @@
+讀取並遵循 `.claude/skills/sunny/SKILL.md` 中的指示。

package/core/.claude/commands//347/224/242/345/223/201.md

@@ -0,0 +1,15 @@
+# /產品 - 產品需求分析
+
+正在讀取產品經理 Agent...
+
+請按照以下步驟執行：
+
+1. **讀取 Agent Skill**：讀取 `.claude/skills/product_manager/SKILL.md` 內容並按照其框架執行。
+
+2. **執行產品經理職責**：
+   - 深度理解用戶需求
+   - 進行需求分析
+   - 將需求拆解為開發任務
+   - 將任務寫入 Pipeline 系統
+
+請開始執行產品需求分析。

package/core/.claude/commands//347/224/242/345/223/201/346/252/242/350/246/226.md

@@ -0,0 +1,129 @@
+# /產品檢視 - 自動化專案檢視模式
+
+## 用途
+此指令用於自動化排程的自我增強迭代。當所有開發任務完成後，自動呼叫此指令檢視專案狀態並決定下一步。
+
+## 執行步驟
+
+請按照以下步驟執行專案檢視分析：
+
+### 1. 讀取必要文件
+- 讀取 `.claude/skills/product_manager/SKILL.md` 採用產品經理角色
+- 讀取 `dev/PRODUCT_SPEC.md` 了解產品規格（若不存在則讀取 `dev/PRODUCT_SPEC_FOR_AI_template.md`）
+- 讀取 `dev/pipeline/PROGRESS.md` 了解目前進度
+- 列出 `dev/pipeline/completed/` 目錄了解已完成任務
+
+### 2. 專案差距分析
+根據讀取的資訊進行分析：
+1. **功能覆蓋率**：比對 PRODUCT_SPEC 中的功能需求與已完成任務
+2. **核心功能檢查**：確認 P0/P1 核心功能是否都已實現
+3. **測試覆蓋**：檢查是否有足夠的測試任務
+4. **品質檢查**：是否需要 QA 驗證或優化任務
+
+### 3. 決策輸出
+**重要限制**：
+- 必須緊貼 PRODUCT_SPEC，不可天馬行空新增功能
+- 只新增 PRODUCT_SPEC 中明確定義但尚未實現的功能
+- 專案完成標準：PRODUCT_SPEC 中所有核心功能 (P0/P1) 已實現
+
+根據分析結果，做出以下決策之一：
+
+#### 決策 A：需要繼續開發
+如果 PRODUCT_SPEC 中有尚未實現的核心功能：
+1. 在 `dev/pipeline/tasks/` 建立新任務
+2. 更新 `dev/pipeline/PROGRESS.md`
+3. 寫入狀態文件：
+
+```json
+// dev/pipeline/SELF_ENHANCEMENT_STATUS.json
+{
+  "status": "continue",
+  "decision": "新增開發任務",
+  "reason": "PRODUCT_SPEC 中尚有未完成的核心功能：[列出功能]",
+  "new_tasks": ["TASK_XXX", "TASK_YYY"],
+  "timestamp": "ISO 時間格式"
+}
+```
+
+#### 決策 B：需要測試/QA
+如果核心功能已實現但缺乏測試：
+1. 在 `dev/pipeline/tasks/` 建立 QA 測試任務
+2. 更新 `dev/pipeline/PROGRESS.md`
+3. 寫入狀態文件：
+
+```json
+{
+  "status": "continue",
+  "decision": "新增測試任務",
+  "reason": "核心功能已完成，需要進行品質驗證",
+  "new_tasks": ["TASK_XXX_QA"],
+  "timestamp": "ISO 時間格式"
+}
+```
+
+#### 決策 C：專案已完成
+如果 PRODUCT_SPEC 中所有核心功能已實現且有基本測試：
+1. 寫入狀態文件：
+
+```json
+{
+  "status": "complete",
+  "decision": "專案已完成",
+  "reason": "PRODUCT_SPEC 中所有核心功能 (P0/P1) 已實現，專案達到可交付狀態",
+  "summary": {
+    "total_completed": 數字,
+    "core_features": ["已完成的核心功能列表"]
+  },
+  "timestamp": "ISO 時間格式"
+}
+```
+
+#### 決策 D：專案已完成但有優化建議
+如果 PRODUCT_SPEC 核心功能已完成，但 AI 判斷有優化空間或可延伸的功能（不在 PRODUCT_SPEC 範圍內）：
+1. **不自動建立任務**（因為超出 PRODUCT_SPEC 範圍）
+2. 寫入狀態文件，記錄建議供用戶參考：
+
+```json
+{
+  "status": "complete_with_suggestions",
+  "decision": "專案已完成，有優化建議",
+  "reason": "PRODUCT_SPEC 核心功能已實現，以下為額外優化建議",
+  "summary": {
+    "total_completed": 數字,
+    "core_features": ["已完成的核心功能列表"]
+  },
+  "suggestions": [
+    {
+      "type": "optimization/feature/enhancement",
+      "title": "建議名稱",
+      "description": "詳細說明",
+      "priority": "P2/P3",
+      "effort": "low/medium/high"
+    }
+  ],
+  "next_action": "建議用戶評估後，可手動執行 /產品 新增需求",
+  "timestamp": "ISO 時間格式"
+}
+```
+
+> **注意**：此決策仍視為「完成」，排程將結束。建議內容僅供用戶參考，不會自動執行。
+
+### 4. 輸出格式
+執行完成後，輸出簡短摘要：
+
+```
+========================================
+專案檢視分析完成
+========================================
+決策：[continue/complete/complete_with_suggestions]
+原因：[簡短說明]
+新增任務：[如有]
+優化建議：[如有，列出建議數量]
+========================================
+```
+
+## 注意事項
+- 此指令專為自動化排程設計
+- 不需要與用戶互動，直接進行分析和決策
+- 嚴格遵循 PRODUCT_SPEC，不可自行發想新功能
+- 狀態文件 `SELF_ENHANCEMENT_STATUS.json` 是排程腳本的關鍵依據

package/core/.claude/commands//350/250/255/350/250/210.md

@@ -0,0 +1,14 @@
+# /設計 - UI/UX 設計
+
+正在讀取設計師 Agent...
+
+請按照以下步驟執行：
+
+1. **讀取 Agent Skill**：讀取 `.claude/skills/designer/SKILL.md` 內容並按照其框架執行。
+
+2. **執行設計師職責**：
+   - 制定設計策略
+   - 創建完整的設計規範
+   - 輸出 DRD.md 文件
+
+請開始執行 UI/UX 設計。

package/core/.claude/commands//350/251/225/344/274/260.md

@@ -0,0 +1,16 @@
+# /評估 - 評估命令
+
+正在啟動評估 Agent...
+
+請按照以下步驟執行：
+
+1. **讀取評估 Agent Skill**：讀取 `.claude/skills/assessment_agent/SKILL.md` 內容並按照其框架執行。
+
+2. **執行評估職責**：
+   - 深度分析目標（功能/頁面/組件/API）
+   - 識別優點、缺點、技術債務
+   - 提出結構化的改良建議
+   - 規劃優先級和執行策略
+   - 生成評估報告 YAML
+
+請開始執行評估分析。

package/core/.claude/commands//351/200/262/345/272/246.md

@@ -0,0 +1,15 @@
+# /進度 - 查看 Pipeline 開發進度
+
+正在讀取開發進度...
+
+請執行以下操作：
+
+1. 讀取 `dev/pipeline/PROGRESS.md` 文件內容
+2. 以清晰的格式顯示：
+   - 當前執行中的任務
+   - 待處理任務佇列
+   - 已完成任務列表
+   - 整體進度統計
+3. 如有阻塞項目，特別標註
+
+請顯示當前 Pipeline 開發進度。

package/core/.claude/hooks/__tests__/substrate.test.js

@@ -0,0 +1,122 @@
+#!/usr/bin/env node
+/**
+ * Smoke tests for substrate-path detection (TASK_1434; TASK_2335 dropped the pulse-before-* hooks).
+ *
+ * Run: node .claude/hooks/__tests__/substrate.test.js
+ *
+ * Hooks are exercised through the one remaining hook that consumes substrate paths.
+ * Hooks are spawned as child processes with mock JSON stdin — no server needed
+ * (Dashboard offline → hooks fall through to deterministic non-server logic).
+ */
+'use strict';
+
+const assert = require('assert');
+const { execSync } = require('child_process');
+const fs = require('fs');
+const path = require('path');
+
+const HOOKS_DIR = path.join(__dirname, '..');
+let passed = 0;
+let failed = 0;
+
+function ok(label, val) {
+  if (val) {
+    console.log(`  ✓ ${label}`);
+    passed++;
+  } else {
+    console.error(`  ✗ ${label}`);
+    failed++;
+  }
+}
+
+// ---------------------------------------------------------------------------
+// Shared helper: spawn a hook with JSON stdin, return parsed stdout JSON
+// ---------------------------------------------------------------------------
+function runHook(hookFile, jsonInput, envOverrides = {}) {
+  const hookPath = path.join(HOOKS_DIR, hookFile);
+  const env = { ...process.env, ...envOverrides };
+  // Clear advisory mode by default so we get 'ask' on substrate
+  if (!('ANSUZ_HOOK_MODE' in envOverrides)) {
+    delete env.ANSUZ_HOOK_MODE;
+  }
+  const result = execSync(`node "${hookPath}"`, {
+    input: JSON.stringify(jsonInput),
+    timeout: 6000,
+    env,
+  });
+  return JSON.parse(result.toString());
+}
+
+// ---------------------------------------------------------------------------
+// 1. _substrate-paths.js — SUBSTRATE_PATH_RE and isSubstratePath
+// ---------------------------------------------------------------------------
+console.log('\n--- _substrate-paths.js ---');
+const { isSubstratePath } = require('../_substrate-paths');
+
+// substrate triggers
+ok('.claude/agents/test.md → substrate',      isSubstratePath('.claude/agents/test.md'));
+ok('.claude/hooks/turn-context.js → substrate', isSubstratePath('.claude/hooks/turn-context.js'));
+ok('.specialist/Ansuz/ → substrate',           isSubstratePath('.specialist/Ansuz/'));
+ok('.specialist/Ansuz/concepts/c.yaml → substrate', isSubstratePath('.specialist/Ansuz/concepts/c.yaml'));
+ok('.specialist/Ansuz/phases/p0.yaml → substrate',  isSubstratePath('.specialist/Ansuz/phases/p0.yaml'));
+ok('.specialist/Ansuz/guards/L-005.yaml → substrate', isSubstratePath('.specialist/Ansuz/guards/L-005.yaml'));
+ok('.specialist/Ansuz/elixir.yaml → substrate', isSubstratePath('.specialist/Ansuz/elixir.yaml'));
+ok('.rule/general/synapse.yaml → substrate',   isSubstratePath('.rule/general/synapse.yaml'));
+ok('.failure/FAIL-001.yaml → substrate',       isSubstratePath('.failure/FAIL-001.yaml'));
+
+// non-substrate
+ok('src/foo.ts → NOT substrate',              !isSubstratePath('src/foo.ts'));
+ok('dashboard/build/ → NOT substrate',        !isSubstratePath('dashboard/build/'));
+ok('.claude/settings.json → NOT substrate',   !isSubstratePath('.claude/settings.json'));
+ok('null → NOT substrate',                    !isSubstratePath(null));
+ok('empty → NOT substrate',                   !isSubstratePath(''));
+
+// ---------------------------------------------------------------------------
+// 4. synapse-first-investigation.js
+// ---------------------------------------------------------------------------
+console.log('\n--- synapse-first-investigation.js ---');
+
+// Helper: parse additionalContext from investigation hook output
+// Isolated project dir: the hook's 5-minute same-command dedupe reads the
+// project's telemetry file, so running this suite twice within 5 minutes
+// against the real repo silenced the ls nudges and failed the asserts.
+const INVESTIGATION_PROJ = fs.mkdtempSync(path.join(require('os').tmpdir(), 'substrate-nudge-'));
+fs.mkdirSync(path.join(INVESTIGATION_PROJ, '.kenaz', 'specialist'), { recursive: true });
+fs.writeFileSync(path.join(INVESTIGATION_PROJ, '.kenaz', 'project.yaml'), 'name: substrate-fixture\n');
+function runInvestigationHook(cmd) {
+  const input = { toolInput: { command: cmd } };
+  return runHook('synapse-first-investigation.js', input, { CLAUDE_PROJECT_DIR: INVESTIGATION_PROJ });
+}
+
+// 4a. ls -la .specialist/Ansuz/ → NO nudge (the substrate/pulse nudge is gone)
+const lsSubstrateResult = runInvestigationHook('ls -la .specialist/Ansuz/');
+ok('ls .specialist/Ansuz/ → no nudge',
+  !lsSubstrateResult.hookSpecificOutput.additionalContext);
+
+// 4b. ls -la dashboard/build/ → NO nudge
+const lsNonSubstrateResult = runInvestigationHook('ls -la dashboard/build/');
+ok('ls dashboard/build/ → NO nudge',
+  !lsNonSubstrateResult.hookSpecificOutput.additionalContext);
+
+// 4c. cat .specialist/Ansuz/elixir.yaml → NO nudge (2026-09-23: reading a file
+//     you already chose is not investigation; pulse cannot replace its content)
+const catSubstrateResult = runInvestigationHook('cat .specialist/Ansuz/elixir.yaml');
+ok('cat substrate file → no nudge',
+  !catSubstrateResult.hookSpecificOutput.additionalContext);
+
+// 4d. A symbol grep over source → LSP nudge (kenaz_lsp_*), no removed tool names
+const grepSymbolResult = runInvestigationHook('grep -rn "resolveBootPolicy" .formation/');
+const grepAdditional = grepSymbolResult.hookSpecificOutput.additionalContext || '';
+ok('symbol grep → LSP nudge mentions kenaz_lsp_references',
+  grepAdditional.includes('kenaz_lsp_references') && !/synapse_(pulse|search|context|impact)/.test(grepAdditional));
+
+// ---------------------------------------------------------------------------
+// Summary
+// ---------------------------------------------------------------------------
+console.log(`\n${'─'.repeat(48)}`);
+console.log(`Passed: ${passed}  Failed: ${failed}`);
+if (failed > 0) {
+  process.exit(1);
+} else {
+  console.log('✓ All substrate hook tests passed');
+}

package/core/.claude/hooks/__tests__/task-1866-cli-guard.test.js

@@ -0,0 +1,193 @@
+'use strict';
+/**
+ * TASK_1866 — worktree-create.js entry guards.
+ *
+ * Defect: any unrecognized CLI arg (e.g. --help) fell through to main() and
+ * silently created a worktree + branch, exit 0. Hook mode (require.main ===
+ * undefined, i.e. node -e require(...)) fabricated a session id when stdin
+ * had no JSON and did the same.
+ *
+ * Tests (ALL git activity confined to a throwaway repo under os.tmpdir()):
+ *   1. CONTROL (highest priority): hook mode WITH valid stdin JSON still
+ *      creates a worktree and emits its path — the fix must not turn the hook
+ *      into a never-creates hook (that broke every dispatch twice before).
+ *   2. hook mode with EMPTY stdin → non-zero exit, zero new worktrees, even
+ *      with CLAUDE_SESSION_ID set (gate is stdin-based, not env-saved).
+ *   3. --help / unknown subcommand → usage on stderr, exit 1, counts frozen.
+ *   4. no-arg CLI with no stdin → non-zero (stdin gate), counts frozen.
+ *   5. no-arg CLI with valid stdin JSON → creates (manual hook invocation).
+ *   6. mark-landed + gc --dry-run do not regress.
+ *
+ * Run: node .claude/hooks/__tests__/task-1866-cli-guard.test.js
+ * Self-cleaning: every created worktree/branch is removed; final assertion is
+ * counts back to baseline; tmp dir removed.
+ *
+ * NEVER point this at the main repo or a live worktree — the throwaway repo is
+ * the only permitted target (see TASK_1866 card).
+ *
+ * Source style note: this file deliberately contains ZERO backslash characters
+ * (line splitting via String.fromCharCode(10) + trim) so it survives shells
+ * that eat an escaping level.
+ */
+
+const fs = require('fs');
+const os = require('os');
+const path = require('path');
+const { execFileSync, spawnSync } = require('child_process');
+
+const HOOK = path.resolve(__dirname, '..', 'worktree-create.js');
+const NODE = process.execPath;
+const NL = String.fromCharCode(10);
+
+function git(repo, args) {
+  return execFileSync('git', ['-C', repo, ...args], { stdio: ['ignore', 'pipe', 'pipe'] })
+    .toString().trim();
+}
+
+function counts(repo) {
+  const wt = git(repo, ['worktree', 'list', '--porcelain'])
+    .split(NL).map((l) => l.trim()).filter((l) => l.startsWith('worktree ')).length;
+  const br = git(repo, ['branch', '--list', '--format=%(refname:short)'])
+    .split(NL).map((l) => l.trim()).filter(Boolean).length;
+  return { wt, br };
+}
+
+let failures = 0;
+function check(name, cond, detail) {
+  if (cond) { console.log('  PASS  ' + name); }
+  else { failures += 1; console.error('  FAIL  ' + name + (detail ? ' — ' + detail : '')); }
+}
+
+const base = fs.mkdtempSync(path.join(os.tmpdir(), 't1866-'));
+const repo = path.join(base, 'repo');
+fs.mkdirSync(repo);
+execFileSync('git', ['init', '-q', repo]);
+git(repo, ['config', 'user.email', 't1866@test.local']);
+git(repo, ['config', 'user.name', 'T1866']);
+fs.writeFileSync(path.join(repo, 'README.md'), 'throwaway');
+git(repo, ['add', '-A']);
+git(repo, ['commit', '-q', '-m', 'init']);
+
+const createdWorktrees = [];
+let created = 0;
+let cleaned = 0;
+
+function cleanupWorktree(wtPath) {
+  const branch = 'ansuz-' + path.basename(wtPath);
+  try { git(repo, ['worktree', 'remove', '--force', wtPath]); } catch (e) {
+    fs.rmSync(wtPath, { recursive: true, force: true, maxRetries: 3 });
+    try { git(repo, ['worktree', 'prune']); } catch { /* best-effort */ }
+  }
+  try { git(repo, ['branch', '-D', branch]); } catch { /* may not exist */ }
+  cleaned += 1;
+}
+
+const hookJson = () => JSON.stringify({
+  session_id: 'deadbeefcafe4711',
+  transcript_path: path.join(base, 'fake-transcript.jsonl'),
+  cwd: repo,
+  hook_event_name: 'WorktreeCreate',
+  worktree_path: path.join(repo, 'ignored-proposal'),
+  isolation_reason: 'test',
+});
+
+const childEnv = { ...process.env, CLAUDE_PROJECT_DIR: repo, CLAUDE_SESSION_ID: 'env-fabrication-bait' };
+
+const baseline = counts(repo);
+console.log('baseline: worktrees=' + baseline.wt + ' branches=' + baseline.br);
+
+try {
+  // ── 1. CONTROL: hook mode (node -e require) + valid stdin JSON → creates ──
+  console.log('[1] control: hook mode with valid stdin JSON');
+  let r = spawnSync(NODE, ['-e', 'require(' + JSON.stringify(HOOK) + ')'],
+    { input: hookJson(), cwd: repo, env: childEnv, encoding: 'utf8' });
+  const emitted = (r.stdout || '').trim();
+  check('exit 0', r.status === 0, 'status=' + r.status + ' stderr=' + r.stderr);
+  check('emits a path', emitted.length > 0, 'stdout empty');
+  check('path exists on disk', emitted !== '' && fs.existsSync(emitted), emitted);
+  let after = counts(repo);
+  check('worktree count +1', after.wt === baseline.wt + 1, baseline.wt + ' -> ' + after.wt);
+  check('branch count +1', after.br === baseline.br + 1, baseline.br + ' -> ' + after.br);
+  if (emitted !== '' && fs.existsSync(emitted)) { createdWorktrees.push(emitted); created += 1; }
+
+  const afterControl = counts(repo);
+
+  // ── 2. hook mode, EMPTY stdin (env session id present) → refuse ──
+  console.log('[2] hook mode with empty stdin');
+  r = spawnSync(NODE, ['-e', 'require(' + JSON.stringify(HOOK) + ')'],
+    { input: '', cwd: repo, env: childEnv, encoding: 'utf8' });
+  check('non-zero exit', r.status !== 0, 'status=' + r.status);
+  check('names stdin in message', (r.stderr || '').includes('stdin'), r.stderr);
+  check('emits no path', (r.stdout || '').trim() === '', r.stdout);
+  after = counts(repo);
+  check('counts frozen', after.wt === afterControl.wt && after.br === afterControl.br,
+    JSON.stringify(after) + ' vs ' + JSON.stringify(afterControl));
+
+  // ── 2b. hook mode, garbage stdin → refuse ──
+  console.log('[2b] hook mode with non-JSON stdin');
+  r = spawnSync(NODE, ['-e', 'require(' + JSON.stringify(HOOK) + ')'],
+    { input: 'not json at all', cwd: repo, env: childEnv, encoding: 'utf8' });
+  check('non-zero exit', r.status !== 0, 'status=' + r.status);
+  after = counts(repo);
+  check('counts frozen', after.wt === afterControl.wt && after.br === afterControl.br);
+
+  // ── 3. --help and unknown subcommand → usage, exit 1, counts frozen ──
+  for (const arg of ['--help', 'frobnicate']) {
+    console.log('[3] CLI unknown arg: ' + arg);
+    r = spawnSync(NODE, [HOOK, arg], { input: '', cwd: repo, env: childEnv, encoding: 'utf8' });
+    check('exit 1', r.status === 1, 'status=' + r.status);
+    check('prints usage', (r.stderr || '').includes('usage:'), r.stderr);
+    check('emits no path', (r.stdout || '').trim() === '', r.stdout);
+    after = counts(repo);
+    check('counts frozen', after.wt === afterControl.wt && after.br === afterControl.br,
+      JSON.stringify(after) + ' vs ' + JSON.stringify(afterControl));
+  }
+
+  // ── 4. no-arg CLI, no stdin → stdin gate refuses ──
+  console.log('[4] no-arg CLI with no stdin');
+  r = spawnSync(NODE, [HOOK], { input: '', cwd: repo, env: childEnv, encoding: 'utf8' });
+  check('non-zero exit', r.status !== 0, 'status=' + r.status);
+  after = counts(repo);
+  check('counts frozen', after.wt === afterControl.wt && after.br === afterControl.br);
+
+  // ── 5. no-arg CLI, valid stdin JSON → manual hook invocation still works ──
+  console.log('[5] no-arg CLI with valid stdin JSON');
+  r = spawnSync(NODE, [HOOK], { input: hookJson(), cwd: repo, env: childEnv, encoding: 'utf8' });
+  const emitted5 = (r.stdout || '').trim();
+  check('exit 0', r.status === 0, 'status=' + r.status + ' stderr=' + r.stderr);
+  check('emits a path', emitted5 !== '' && fs.existsSync(emitted5), emitted5);
+  after = counts(repo);
+  check('worktree count +1', after.wt === afterControl.wt + 1, afterControl.wt + ' -> ' + after.wt);
+  if (emitted5 !== '' && fs.existsSync(emitted5)) { createdWorktrees.push(emitted5); created += 1; }
+
+  // ── 6. mark-landed + gc --dry-run regressions ──
+  console.log('[6] mark-landed');
+  r = spawnSync(NODE, [HOOK, 'mark-landed', 'wt-11111111-1', '--repo', repo, '--task', 'TASK_T1866'],
+    { input: '', cwd: repo, env: childEnv, encoding: 'utf8' });
+  check('exit 0', r.status === 0, 'status=' + r.status + ' stderr=' + r.stderr);
+  const ledger = path.join(repo, '.formation', 'worktrees', 'landed-ledger.jsonl');
+  check('ledger entry written', fs.existsSync(ledger)
+    && fs.readFileSync(ledger, 'utf8').includes('ansuz-wt-11111111-1'));
+
+  console.log('[6b] gc --dry-run');
+  const preGc = counts(repo);
+  r = spawnSync(NODE, [HOOK, 'gc', '--dry-run', '--repo', repo],
+    { input: '', cwd: repo, env: childEnv, encoding: 'utf8' });
+  check('exit 0', r.status === 0, 'status=' + r.status + ' stderr=' + r.stderr);
+  check('dry-run announced', (r.stderr || '').includes('DRY-RUN'), r.stderr);
+  after = counts(repo);
+  check('dry-run removed nothing', after.wt === preGc.wt && after.br === preGc.br);
+} finally {
+  // ── cleanup: every created worktree + branch, then verify baseline ──
+  for (const p of createdWorktrees) cleanupWorktree(p);
+  const finalCounts = counts(repo);
+  check('cleanup returns to baseline',
+    finalCounts.wt === baseline.wt && finalCounts.br === baseline.br,
+    JSON.stringify(finalCounts) + ' vs baseline ' + JSON.stringify(baseline));
+  console.log('Worktrees created during testing: ' + created);
+  console.log('Worktrees cleaned up:             ' + cleaned);
+  fs.rmSync(base, { recursive: true, force: true, maxRetries: 5, retryDelay: 200 });
+}
+
+if (created !== cleaned) { console.error('LEAK: created !== cleaned'); process.exit(1); }
+process.exit(failures === 0 ? 0 : 1);