fix(skills): replace polling/pre-spawning with Stop-Wait pattern across all team coordinators

All team coordinator roles now follow the Stop-Wait design principle: - Phase 2: Remove worker pre-spawning, workers are spawned per-stage in Phase 4 - Phase 4: Add Stop-Wait principle note (synchronous Task calls instead of polling) - monitor.md: Replace while+sleep polling loops with synchronous Task execution Teams updated: team-brainstorm, team-frontend, team-issue, team-iterdev, team-lifecycle, team-lifecycle-v2, team-quality-assurance, team-tech-debt, team-testing, team-uidesign, team-ultra-analyze
2026-02-28 09:23:08 +08:00 · 2026-02-24 00:05:29 +08:00
parent 695045787f
commit e92c6ce0b1
17 changed files with 616 additions and 445 deletions
--- a/.claude/skills/team-brainstorm/roles/coordinator.md
+++ b/.claude/skills/team-brainstorm/roles/coordinator.md
@@ -94,7 +94,7 @@ AskUserQuestion({
 })
 ```

-### Phase 2: Create Team + Spawn Workers
+### Phase 2: Create Team + Initialize Session

 ```javascript
 TeamCreate({ team_name: teamName })
@@ -135,7 +135,9 @@ const teamSession = {
 Write(`${sessionFolder}/team-session.json`, JSON.stringify(teamSession, null, 2))
 ```

-Spawn workers (see SKILL.md Coordinator Spawn Template).
+// ⚠️ Workers are NOT pre-spawned here.
+// Workers are spawned per-stage in Phase 4 via Stop-Wait Task(run_in_background: false).
+// See SKILL.md Coordinator Spawn Template for worker prompt templates.

 ### Phase 3: Create Task Chain

@@ -212,6 +214,13 @@ TaskUpdate({ taskId: evalId, owner: "evaluator", addBlockedBy: [synthId] })

 ### Phase 4: Coordination Loop + Generator-Critic Control

+> **设计原则（Stop-Wait）**: 模型执行没有时间概念，禁止任何形式的轮询等待。
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态
+> - ✅ 采用: 同步 `Task(run_in_background: false)` 调用，Worker 返回 = 阶段完成信号
+>
+> 按 Phase 3 创建的任务链顺序，逐阶段 spawn worker 同步执行。
+> Worker prompt 使用 SKILL.md Coordinator Spawn Template。
+
 | Received Message | Action |
 |-----------------|--------|
 | ideator: ideas_ready | Read ideas → team_msg log → TaskUpdate completed → unblock CHALLENGE |
--- a/.claude/skills/team-frontend/roles/coordinator/role.md
+++ b/.claude/skills/team-frontend/roles/coordinator/role.md
@@ -149,7 +149,7 @@ AskUserQuestion({
 })
 ```

-### Phase 2: Create Team + Session + Spawn Teammates
+### Phase 2: Create Team + Initialize Session

 ```javascript
 // Create session directory
@@ -183,9 +183,11 @@ Write(`${sessionFolder}/shared-memory.json`, JSON.stringify({
  industry_context: { industry: industryChoice, config: industry }
 }, null, 2))

-// Create team and spawn workers
+// Create team
 TeamCreate({ team_name: teamName })
-// → Spawn analyst, architect, developer, qa (see SKILL.md Coordinator Spawn Template)
+// ⚠️ Workers are NOT pre-spawned here.
+// Workers are spawned per-stage in Phase 4 via Stop-Wait Task(run_in_background: false).
+// See SKILL.md Coordinator Spawn Template for worker prompt templates.
 ```

 ### Phase 3: Create Task Chain
@@ -227,6 +229,13 @@ if (pipeline === 'system') {

 ### Phase 4: Coordination Loop

+> **设计原则（Stop-Wait）**: 模型执行没有时间概念，禁止任何形式的轮询等待。
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态
+> - ✅ 采用: 同步 `Task(run_in_background: false)` 调用，Worker 返回 = 阶段完成信号
+>
+> 按 Phase 3 创建的任务链顺序，逐阶段 spawn worker 同步执行。
+> Worker prompt 使用 SKILL.md Coordinator Spawn Template。
+
 Receive teammate messages, dispatch based on content.
 **Before each decision**: `team_msg list` to check recent messages.
 **After each decision**: `team_msg log` to record.
--- a/.claude/skills/team-issue/roles/coordinator.md
+++ b/.claude/skills/team-issue/roles/coordinator.md
@@ -136,38 +136,18 @@ function detectMode(issueIds, userMode) {
 }
 ```

-### Phase 2: Create Team + Spawn Workers
+### Phase 2: Create Team + Initialize Session

 ```javascript
 TeamCreate({ team_name: "issue" })

-// Spawn workers based on mode
-const workersToSpawn = mode === 'quick'
-  ? ['explorer', 'planner', 'integrator', 'implementer']  // No reviewer in quick mode
-  : ['explorer', 'planner', 'reviewer', 'integrator', 'implementer']
-
-for (const workerName of workersToSpawn) {
-  Task({
-    subagent_type: "general-purpose",
-    team_name: "issue",
-    name: workerName,
-    prompt: `你是 team "issue" 的 ${workerName.toUpperCase()}。
-当你收到任务时，调用 Skill(skill="team-issue", args="--role=${workerName}") 执行。
-当前需求: 处理 issue ${issueIds.join(', ')}，模式: ${mode}
-约束: CLI-first data access, 所有 issue 操作通过 ccw issue 命令
-
-## 角色准则（强制）
- 所有输出必须带 [${workerName}] 标识前缀
- 仅与 coordinator 通信
- 每次 SendMessage 前，先调用 mcp__ccw-tools__team_msg 记录
-
-工作流程:
-1. TaskList → 找到分配给你的任务
-2. Skill(skill="team-issue", args="--role=${workerName}") 执行
-3. team_msg log + SendMessage 结果给 coordinator
-4. TaskUpdate completed → 检查下一个任务`
-  })
-}
+// ⚠️ Workers are NOT pre-spawned here.
+// Workers are spawned per-stage in Phase 4 via Stop-Wait Task(run_in_background: false).
+// See SKILL.md Coordinator Spawn Template for worker prompt templates.
+//
+// Worker roles available (spawned on-demand per pipeline stage):
+//   quick mode:  explorer, planner, integrator, implementer
+//   full mode:   explorer, planner, reviewer, integrator, implementer
 ```

 ### Phase 3: Create Task Chain
@@ -323,6 +303,13 @@ const marshalId = TaskCreate({

 ### Phase 4: Coordination Loop

+> **设计原则（Stop-Wait）**: 模型执行没有时间概念，禁止任何形式的轮询等待。
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态
+> - ✅ 采用: 同步 `Task(run_in_background: false)` 调用，Worker 返回 = 阶段完成信号
+>
+> 按 Phase 3 创建的任务链顺序，逐阶段 spawn worker 同步执行。
+> Worker prompt 使用 SKILL.md Coordinator Spawn Template。
+
 Receive teammate messages, dispatch based on type.

 | Received Message | Action |
--- a/.claude/skills/team-iterdev/roles/coordinator.md
+++ b/.claude/skills/team-iterdev/roles/coordinator.md
@@ -112,7 +112,9 @@ const teamSession = {
 Write(`${sessionFolder}/team-session.json`, JSON.stringify(teamSession, null, 2))
 ```

-Spawn workers (see SKILL.md Coordinator Spawn Template).
+// ⚠️ Workers are NOT pre-spawned here.
+// Workers are spawned per-stage in Phase 4 via Stop-Wait Task(run_in_background: false).
+// See SKILL.md Coordinator Spawn Template for worker prompt templates.

 ### Phase 3: Create Task Chain + Update Ledger

@@ -161,6 +163,13 @@ TaskUpdate({ taskId: reviewId, owner: "reviewer", addBlockedBy: [devId] })

 ### Phase 4: Coordination Loop + GC Control + Ledger Updates

+> **设计原则（Stop-Wait）**: 模型执行没有时间概念，禁止任何形式的轮询等待。
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态
+> - ✅ 采用: 同步 `Task(run_in_background: false)` 调用，Worker 返回 = 阶段完成信号
+>
+> 按 Phase 3 创建的任务链顺序，逐阶段 spawn worker 同步执行。
+> Worker prompt 使用 SKILL.md Coordinator Spawn Template。
+
 | Received Message | Action |
 |-----------------|--------|
 | architect: design_ready | Read design → update ledger → unblock DEV |
--- a/.claude/skills/team-lifecycle-v2/roles/coordinator/commands/monitor.md
+++ b/.claude/skills/team-lifecycle-v2/roles/coordinator/commands/monitor.md
@@ -10,62 +10,66 @@

 ## Coordination Loop

+> **设计原则**: 模型执行没有时间概念，禁止任何形式的轮询等待。
+> 使用同步 `Task(run_in_background: false)` 调用作为等待机制。
+> Worker 返回 = 阶段完成信号（天然回调），无需 sleep 轮询。
+
 ```javascript
-Output("[coordinator] Entering coordination loop...")
+Output("[coordinator] Entering coordination loop (Stop-Wait mode)...")

-let loopActive = true
-let checkpointPending = false
+// Sequentially execute each task by spawning its worker
+const allTasks = TaskList()
+const pendingTasks = allTasks.filter(t => t.status !== 'completed' && t.assigned_to !== 'coordinator')

-while (loopActive) {
-  // Load current session state
-  const session = Read(sessionFile)
-  const teamState = TeamGet(session.team_id)
-  const allTasks = teamState.tasks
+for (const task of pendingTasks) {
+  // Check if all dependencies are met
+  const allDepsMet = (task.dependencies || []).every(depId => {
+    const dep = TaskGet(depId)
+    return dep.status === "completed"
+  })

-  // Check for incoming messages
-  const messages = TeamGetMessages(session.team_id)
-
-  for (const message of messages) {
-    Output(`[coordinator] Received message: ${message.type} from ${message.sender}`)
-
-    switch (message.type) {
-      case "task_complete":
-        handleTaskComplete(message)
-        break
-
-      case "task_blocked":
-        handleTaskBlocked(message)
-        break
-
-      case "discussion_needed":
-        handleDiscussionNeeded(message)
-        break
-
-      case "research_complete":
-        handleResearchComplete(message)
-        break
-
-      default:
-        Output(`[coordinator] Unknown message type: ${message.type}`)
-    }
+  if (!allDepsMet) {
+    Output(`[coordinator] Task ${task.task_id} blocked by dependencies, skipping`)
+    continue
  }

-  // Check if all tasks complete
-  const completedTasks = allTasks.filter(t => t.status === "completed")
-  const totalTasks = allTasks.length
+  Output(`[coordinator] Starting task: ${task.task_id} (assigned to: ${task.assigned_to})`)

-  if (completedTasks.length === totalTasks) {
-    Output("[coordinator] All tasks completed!")
-    loopActive = false
-    break
+  // Mark as active
+  TaskUpdate(task.task_id, {
+    status: "active",
+    started_at: new Date().toISOString()
+  })
+
+  // Spawn worker — blocks until worker returns (Stop-Wait)
+  Task({
+    subagent_type: "general-purpose",
+    prompt: `Execute task ${task.task_id}: ${task.description}
+Assigned role: ${task.assigned_to}
+When complete, mark TaskUpdate(${task.task_id}, { status: "completed" })`,
+    run_in_background: false
+  })
+
+  // Worker returned — check status
+  const completedTask = TaskGet(task.task_id)
+  Output(`[coordinator] Task ${task.task_id} status: ${completedTask.status}`)
+
+  if (completedTask.status === "completed") {
+    handleTaskComplete({ task_id: task.task_id, output: completedTask.output })
  }

  // Update session progress
-  session.tasks_completed = completedTasks.length
+  const session = Read(sessionFile)
+  const allTasksNow = TaskList()
+  session.tasks_completed = allTasksNow.filter(t => t.status === "completed").length
  Write(sessionFile, session)

-  // Sleep before next iteration
-  sleep(5000) // 5 seconds
+  // Check if all tasks complete
+  const remaining = allTasksNow.filter(t => t.status !== "completed" && t.assigned_to !== 'coordinator')
+  if (remaining.length === 0) {
+    Output("[coordinator] All tasks completed!")
+    break
+  }
 }

 Output("[coordinator] Coordination loop complete")
--- a/.claude/skills/team-lifecycle-v2/roles/coordinator/role.md
+++ b/.claude/skills/team-lifecycle-v2/roles/coordinator/role.md
@@ -377,9 +377,9 @@ goto Phase2

 ---

-### Phase 2: Create Team + Spawn Workers
+### Phase 2: Create Team + Initialize Session

-**Purpose**: Initialize team and spawn worker subagents
+**Purpose**: Initialize team and session state

 ```javascript
 Output("[coordinator] Phase 2: Team Creation")
@@ -430,43 +430,17 @@ const sessionData = {
 Write(sessionFile, sessionData)
 Output(`[coordinator] Session file created: ${sessionFile}`)

-// Spawn workers conditionally based on pipeline mode
-const isFE = ['fe-only', 'fullstack', 'full-lifecycle-fe'].includes(requirements.mode)
-const isBE = ['impl-only', 'fullstack', 'full-lifecycle', 'full-lifecycle-fe'].includes(requirements.mode)
-const isSpec = ['spec-only', 'full-lifecycle', 'full-lifecycle-fe'].includes(requirements.mode)
-
-if (isSpec) {
-  TeamSpawn({ team_id: teamId, role: "spec-writer", count: 1 })
-  Output("[coordinator] Spawned spec-writer")
-}
-
-if (isBE) {
-  TeamSpawn({ team_id: teamId, role: "implementer", count: 1 })
-  Output("[coordinator] Spawned implementer")
-}
-
-if (isFE) {
-  TeamSpawn({ team_id: teamId, role: "fe-developer", count: 1 })
-  Output("[coordinator] Spawned fe-developer")
-  TeamSpawn({ team_id: teamId, role: "fe-qa", count: 1 })
-  Output("[coordinator] Spawned fe-qa")
-
-  // Initialize shared memory for frontend pipeline
-  const sharedMemoryPath = `${sessionFolder}/shared-memory.json`
-  Write(sharedMemoryPath, JSON.stringify({
-    design_intelligence: {},
-    design_token_registry: {},
-    component_inventory: [],
-    style_decisions: [],
-    qa_history: [],
-    industry_context: {}
-  }, null, 2))
-  Output("[coordinator] Initialized shared-memory.json for frontend pipeline")
-}
-
-// Always spawn researcher for ambiguity resolution
-TeamSpawn({ team_id: teamId, role: "researcher", count: 1 })
-Output("[coordinator] Spawned researcher")
+// ⚠️ Workers are NOT pre-spawned here.
+// Workers are spawned per-stage in Phase 4 via Stop-Wait Task(run_in_background: false).
+// See SKILL.md Coordinator Spawn Template for worker prompt templates.
+//
+// Worker roles by mode (spawned on-demand):
+//   spec-only: spec-writer
+//   impl-only: implementer
+//   fe-only: fe-developer, fe-qa
+//   fullstack: implementer, fe-developer, fe-qa
+//   full-lifecycle / full-lifecycle-fe: spec-writer + relevant impl roles
+//   Always available: researcher (for ambiguity resolution)

 goto Phase3
 ```
@@ -495,6 +469,13 @@ goto Phase4

 **Purpose**: Monitor task progress and route messages

+> **设计原则（Stop-Wait）**: 模型执行没有时间概念，禁止任何形式的轮询等待。
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态
+> - ✅ 采用: 同步 `Task(run_in_background: false)` 调用，Worker 返回 = 阶段完成信号
+>
+> 按 Phase 3 创建的任务链顺序，逐阶段 spawn worker 同步执行。
+> Worker prompt 使用 SKILL.md Coordinator Spawn Template。
+
 ```javascript
 Output("[coordinator] Phase 4: Coordination Loop")

--- a/.claude/skills/team-lifecycle/roles/coordinator.md
+++ b/.claude/skills/team-lifecycle/roles/coordinator.md
@@ -188,15 +188,11 @@ if (isResume) {
      if (meta) neededRoles.add(meta.owner)
    })

-    // Spawn only needed workers using Phase 2 spawn template (see SKILL.md Coordinator Spawn Template)
-    // Each worker is spawned with prompt that:
-    // 1. Identifies their role
-    // 2. Instructs to call Skill(skill="team-lifecycle", args="--role=<name>")
-    // 3. Includes session context: taskDescription, sessionFolder, constraints
-    // 4. Instructs immediate TaskList polling on startup
+    // Spawn only needed workers using Phase 4 Stop-Wait pattern (see SKILL.md Coordinator Spawn Template)
+    // Workers are spawned per-stage via Task(run_in_background: false) in Phase 4 coordination loop.
+    // neededRoles is used to determine which workers will be spawned on-demand.
    neededRoles.forEach(role => {
-      // → Use SKILL.md Coordinator Spawn Template for each role
-      // → Worker prompt includes: "Session: ${sessionFolder}", "需求: ${taskDescription}"
+      // → Worker prompt template in SKILL.md (spawned per-stage in Phase 4, not pre-spawned here)
    })

    // ============================================================
@@ -419,7 +415,7 @@ if (mode === 'impl-only' || mode === 'full-lifecycle') {
 }
 ```

-### Phase 2: Create Team + Spawn Workers
+### Phase 2: Create Team + Initialize Session

 ```javascript
 TeamCreate({ team_name: teamName })
@@ -462,9 +458,11 @@ const teamSession = {
 Write(`${sessionFolder}/team-session.json`, JSON.stringify(teamSession, null, 2))
 ```

-**Conditional spawn based on mode** (see SKILL.md Coordinator Spawn Template for full prompts):
+**Workers are NOT pre-spawned here.** Workers are spawned per-stage in Phase 4 via Stop-Wait `Task(run_in_background: false)`. See SKILL.md Coordinator Spawn Template for worker prompt templates.

-| Mode | Spawned Workers |
+Worker roles by mode (spawned on-demand):
+
+| Mode | Worker Roles |
 |------|-----------------|
 | spec-only | analyst, writer, discussant, reviewer (4) |
 | impl-only | planner, executor, tester, reviewer (4) |
@@ -562,6 +560,13 @@ TaskUpdate({ taskId: planId, owner: "planner", addBlockedBy: [discuss6Id] })

 ### Phase 4: Coordination Loop

+> **设计原则（Stop-Wait）**: 模型执行没有时间概念，禁止任何形式的轮询等待。
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态
+> - ✅ 采用: 同步 `Task(run_in_background: false)` 调用，Worker 返回 = 阶段完成信号
+>
+> 按 Phase 3 创建的任务链顺序，逐阶段 spawn worker 同步执行。
+> Worker prompt 使用 SKILL.md Coordinator Spawn Template。
+
 Receive teammate messages and make dispatch decisions. **Before each decision: `team_msg list` to review recent messages. After each decision: `team_msg log` to record.**

 #### Spec Messages
--- a/.claude/skills/team-quality-assurance/roles/coordinator/commands/monitor.md
+++ b/.claude/skills/team-quality-assurance/roles/coordinator/commands/monitor.md
@@ -20,8 +20,15 @@

 ### 设计原则

-> **模型执行没有时间概念**。禁止空转 while 循环检查状态。
-> 使用固定 sleep 间隔 + 最大轮询次数，避免无意义的 API 调用浪费。
+> **模型执行没有时间概念，禁止任何形式的轮询等待。**
+>
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态（空转浪费 API 轮次）
+> - ❌ 禁止: `Bash(sleep N)` / `Bash(timeout /t N)` 作为等待手段
+> - ✅ 采用: 同步 `Task()` 调用（`run_in_background: false`），call 本身即等待
+> - ✅ 采用: Worker 返回 = 阶段完成信号（天然回调）
+>
+> **原理**: `Task(run_in_background: false)` 是阻塞调用，coordinator 自动挂起直到 worker 返回。
+> 无需 sleep，无需轮询，无需消息总线监控。Worker 的返回就是回调。

 ### Decision Logic

@@ -47,14 +54,16 @@ const routingTable = {
 }
 ```

-### 等待策略常量
+### Stage-Worker 映射表

 ```javascript
-const POLL_INTERVAL_SEC = 300  // 每次检查间隔 5 分钟（测试执行可能很慢）
-const MAX_POLLS_PER_STAGE = 6  // 单阶段最多等待 6 次（~30 分钟）
-const SLEEP_CMD = process.platform === 'win32'
-  ? `timeout /t ${POLL_INTERVAL_SEC} /nobreak >nul 2>&1`
-  : `sleep ${POLL_INTERVAL_SEC}`
+const STAGE_WORKER_MAP = {
+  'SCOUT':   { role: 'scout',      skillArgs: '--role=scout' },
+  'QASTRAT': { role: 'strategist', skillArgs: '--role=strategist' },
+  'QAGEN':   { role: 'generator',  skillArgs: '--role=generator' },
+  'QARUN':   { role: 'executor',   skillArgs: '--role=executor' },
+  'QAANA':   { role: 'analyst',    skillArgs: '--role=analyst' }
+}

 // ★ 统一 auto mode 检测：-y/--yes 从 $ARGUMENTS 或 ccw 传播
 const autoYes = /\b(-y|--yes)\b/.test(args)
@@ -83,94 +92,125 @@ const pipelineTasks = allTasks
  .sort((a, b) => Number(a.id) - Number(b.id))
 ```

-### Step 2: Stage-Driven Execution
+### Step 2: Sequential Stage Execution (Stop-Wait)

-> **核心改动**: 不再使用 while 轮询循环。按 pipeline 阶段顺序，逐阶段等待完成。
-> 每个阶段：sleep → 检查消息 → 确认任务状态 → 处理结果 → 下一阶段。
+> **核心**: 逐阶段 spawn worker，同步阻塞等待返回。
+> Worker 返回 = 阶段完成。无 sleep、无轮询、无消息总线监控。

 ```javascript
 // 按依赖顺序处理每个阶段
 for (const stageTask of pipelineTasks) {
-  // --- 等待当前阶段完成 ---
-  let stageComplete = false
-  let pollCount = 0
+  // 1. 提取阶段前缀 → 确定 worker 角色
+  const stagePrefix = stageTask.subject.match(/^([\w-]+)-\d/)?.[1]?.replace(/-L\d$/, '')
+  const workerConfig = STAGE_WORKER_MAP[stagePrefix]

-  while (!stageComplete && pollCount < MAX_POLLS_PER_STAGE) {
-    // ★ 固定等待：sleep 30s，让 worker 有执行时间
-    Bash(SLEEP_CMD)
-    pollCount++
-
-    // 1. 检查消息总线（主要信号源）
-    const messages = mcp__ccw-tools__team_msg({
-      operation: "list",
-      team: teamName,
-      last: 5
+  if (!workerConfig) {
+    mcp__ccw-tools__team_msg({
+      operation: "log", team: teamName, from: "coordinator",
+      to: "user", type: "error",
+      summary: `[coordinator] 未知阶段前缀: ${stagePrefix}，跳过`
    })
-
-    // 2. 路由消息
-    for (const msg of messages) {
-      const handler = routingTable[msg.type]
-      if (!handler) continue
-      processMessage(msg, handler)
-    }
-
-    // 3. 确认任务状态（兜底）
-    const currentTask = TaskGet({ taskId: stageTask.id })
-    stageComplete = currentTask.status === 'completed' || currentTask.status === 'deleted'
+    continue
  }

-  // --- 阶段超时处理 ---
-  if (!stageComplete) {
-    const elapsedMin = Math.round(pollCount * POLL_INTERVAL_SEC / 60)
+  // 2. 标记任务为执行中
+  TaskUpdate({ taskId: stageTask.id, status: 'in_progress' })

+  mcp__ccw-tools__team_msg({
+    operation: "log", team: teamName, from: "coordinator",
+    to: workerConfig.role, type: "task_unblocked",
+    summary: `[coordinator] 启动阶段: ${stageTask.subject} → ${workerConfig.role}`
+  })
+
+  // 3. 同步 spawn worker — 阻塞直到 worker 返回（Stop-Wait 核心）
+  const workerResult = Task({
+    subagent_type: "general-purpose",
+    prompt: `你是 team "${teamName}" 的 ${workerConfig.role.toUpperCase()}。
+
+## ⚠️ 首要指令（MUST）
+Skill(skill="team-quality-assurance", args="${workerConfig.skillArgs}")
+
+## 当前任务
+- 任务 ID: ${stageTask.id}
+- 任务: ${stageTask.subject}
+- 描述: ${stageTask.description || taskDescription}
+- Session: ${sessionFolder}
+
+## 角色准则（强制）
+- 所有输出必须带 [${workerConfig.role}] 标识前缀
+- 仅与 coordinator 通信
+
+## 工作流程
+1. Skill(skill="team-quality-assurance", args="${workerConfig.skillArgs}") 获取角色定义
+2. 执行任务 → 汇报结果
+3. TaskUpdate({ taskId: "${stageTask.id}", status: "completed" })`,
+    run_in_background: false
+  })
+
+  // 4. Worker 已返回 — 直接处理结果
+  const taskState = TaskGet({ taskId: stageTask.id })
+
+  if (taskState.status !== 'completed') {
+    // Worker 返回但未标记 completed → 异常处理
    if (autoYes) {
-      // 自动模式：记录日志，自动跳过
      mcp__ccw-tools__team_msg({
        operation: "log", team: teamName, from: "coordinator",
        to: "user", type: "error",
-        summary: `[coordinator] [auto] 阶段 ${stageTask.subject} 超时 (${elapsedMin}min)，自动跳过`
+        summary: `[coordinator] [auto] 阶段 ${stageTask.subject} 未完成，自动跳过`
      })
      TaskUpdate({ taskId: stageTask.id, status: 'deleted' })
      continue
    }

-    // 交互模式：由用户决定
    const decision = AskUserQuestion({
      questions: [{
-        question: `阶段 "${stageTask.subject}" 已等待 ${elapsedMin} 分钟仍未完成。如何处理？`,
-        header: "Stage Wait",
+        question: `阶段 "${stageTask.subject}" worker 返回但未完成。如何处理？`,
+        header: "Stage Fail",
        multiSelect: false,
        options: [
-          { label: "继续等待", description: `再等 ${MAX_POLLS_PER_STAGE} 轮（~${Math.round(MAX_POLLS_PER_STAGE * POLL_INTERVAL_SEC / 60)}min）` },
-          { label: "跳过此阶段", description: "标记为跳过，继续后续流水线" },
-          { label: "终止流水线", description: "停止整个 QA 流程，汇报当前结果" }
+          { label: "重试", description: "重新 spawn worker 执行此阶段" },
+          { label: "跳过", description: "标记为跳过，继续后续流水线" },
+          { label: "终止", description: "停止整个 QA 流程，汇报当前结果" }
        ]
      }]
    })

-    const answer = decision["Stage Wait"]
-
-    if (answer === "继续等待") {
-      // 重置计数器，继续等待当前阶段
-      pollCount = 0
-      // 重新进入当前阶段的等待循环（需要用 while 包裹，此处用 goto 语义）
-      continue // 注意：实际执行中需要将 for 改为可重入的逻辑
-    } else if (answer === "跳过此阶段") {
-      mcp__ccw-tools__team_msg({
-        operation: "log", team: teamName, from: "coordinator",
-        to: "user", type: "error",
-        summary: `[coordinator] 用户选择跳过阶段 ${stageTask.subject}`
-      })
+    const answer = decision["Stage Fail"]
+    if (answer === "跳过") {
      TaskUpdate({ taskId: stageTask.id, status: 'deleted' })
      continue
-    } else {
-      // 终止流水线
+    } else if (answer === "终止") {
      mcp__ccw-tools__team_msg({
        operation: "log", team: teamName, from: "coordinator",
        to: "user", type: "shutdown",
        summary: `[coordinator] 用户终止流水线，当前阶段: ${stageTask.subject}`
      })
-      break // 跳出 for 循环，进入 Step 3 汇报
+      break
+    }
+    // 重试: continue to next iteration will re-process if logic wraps
+  } else {
+    mcp__ccw-tools__team_msg({
+      operation: "log", team: teamName, from: "coordinator",
+      to: "user", type: "quality_gate",
+      summary: `[coordinator] 阶段完成: ${stageTask.subject}`
+    })
+  }
+
+  // 5. 阶段间检查（QARUN 阶段检查覆盖率，决定 GC 循环）
+  if (stagePrefix === 'QARUN') {
+    const latestMemory = JSON.parse(Read(`${sessionFolder}/shared-memory.json`))
+    const coverage = latestMemory.execution_results?.coverage || 0
+    const targetLayer = stageTask.metadata?.layer || 'L1'
+    const target = coverageTargets[targetLayer] || 80
+
+    if (coverage < target && gcIteration < MAX_GC_ITERATIONS) {
+      gcIteration++
+      mcp__ccw-tools__team_msg({
+        operation: "log", team: teamName, from: "coordinator",
+        to: "generator", type: "gc_loop_trigger",
+        summary: `[coordinator] GC循环 #${gcIteration}: 覆盖率 ${coverage}% < ${target}%，请修复`
+      })
+      // 创建 GC 修复任务追加到 pipeline
    }
  }
 }
@@ -284,10 +324,8 @@ const summary = {

 | Scenario | Resolution |
 |----------|------------|
-| Message bus unavailable | Fall back to TaskList polling only |
-| Stage timeout (交互模式) | AskUserQuestion：继续等待 / 跳过 / 终止流水线 |
-| Stage timeout (自动模式 `-y`/`--yes`，`autoYes`) | 自动跳过，记录日志，继续流水线 |
-| Teammate unresponsive (2x no response) | Respawn teammate with same task |
-| Deadlock detected (tasks blocked indefinitely) | Identify cycle, manually unblock |
+| Worker 返回但未 completed (交互模式) | AskUserQuestion: 重试 / 跳过 / 终止 |
+| Worker 返回但未 completed (自动模式) | 自动跳过，记录日志 |
+| Worker spawn 失败 | 重试一次，仍失败则上报用户 |
 | Quality gate FAIL | Report to user, suggest targeted re-run |
 | GC loop stuck >3 iterations | Accept current coverage, continue pipeline |
--- a/.claude/skills/team-quality-assurance/roles/coordinator/role.md
+++ b/.claude/skills/team-quality-assurance/roles/coordinator/role.md
@@ -99,7 +99,7 @@ if (!autoYes && (!taskDescription || taskDescription.length < 10)) {
 }
 ```

-### Phase 2: Create Team + Spawn Teammates
+### Phase 2: Create Team + Initialize Session

 ```javascript
 const teamName = "quality-assurance"
@@ -121,8 +121,10 @@ Write(`${sessionFolder}/shared-memory.json`, JSON.stringify({

 TeamCreate({ team_name: teamName })

-// Spawn teammates (see SKILL.md Coordinator Spawn Template)
-// Scout, Strategist, Generator, Executor, Analyst
+// ⚠️ Workers are NOT pre-spawned here.
+// Workers are spawned per-stage in Phase 4 via Stop-Wait Task(run_in_background: false).
+// See SKILL.md Coordinator Spawn Template for worker prompt templates.
+// Worker roles: Scout, Strategist, Generator, Executor, Analyst
 ```

 ### Phase 3: Create Task Chain
@@ -151,6 +153,13 @@ SCOUT-001 → QASTRAT-001 → [QAGEN-001(L1) + QAGEN-002(L2)](parallel) → [QAR

 ### Phase 4: Coordination Loop

+> **设计原则（Stop-Wait）**: 模型执行没有时间概念，禁止任何形式的轮询等待。
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态
+> - ✅ 采用: 同步 `Task(run_in_background: false)` 调用，Worker 返回 = 阶段完成信号
+>
+> 按 Phase 3 创建的任务链顺序，逐阶段 spawn worker 同步执行。
+> Worker prompt 使用 SKILL.md Coordinator Spawn Template。
+
 ```javascript
 // Read commands/monitor.md for full implementation
 Read("commands/monitor.md")
--- a/.claude/skills/team-tech-debt/SKILL.md
+++ b/.claude/skills/team-tech-debt/SKILL.md
@@ -243,14 +243,21 @@ TDFIX → TDVAL → (if regression or quality drop) → TDFIX-fix → TDVAL-2

 ## Coordinator Spawn Template

+> **注意**: 以下模板作为 worker prompt 参考。在 Stop-Wait 策略下，coordinator 不再在 Phase 2 预先 spawn 所有 worker。
+> 而是在 Phase 4 (monitor) 中，按 pipeline 阶段逐个 spawn worker（同步阻塞 `Task(run_in_background: false)`），
+> worker 返回即阶段完成。详见 `roles/coordinator/commands/monitor.md`。
+
 ```javascript
 TeamCreate({ team_name: teamName })

-// Scanner
+// Worker 按需 spawn（monitor.md Phase 4 调用）
+// 以下为各角色的 prompt 模板参考：
+```
+
+### Scanner Prompt Template
+```javascript
 Task({
  subagent_type: "general-purpose",
-  team_name: teamName,
-  name: "scanner",
  prompt: `你是 team "${teamName}" 的 SCANNER。

 ## ⚠️ 首要指令（MUST）
@@ -274,14 +281,15 @@ Skill(skill="team-tech-debt", args="--role=scanner")
 1. 调用 Skill(skill="team-tech-debt", args="--role=scanner") 获取角色定义和执行逻辑
 2. 按 role.md 中的 5-Phase 流程执行（TaskList → 找到 TDSCAN-* 任务 → 执行 → 汇报）
 3. team_msg log + SendMessage 结果给 coordinator（带 [scanner] 标识）
-4. TaskUpdate completed → 检查下一个任务 → 回到步骤 1`
+4. TaskUpdate completed → 检查下一个任务 → 回到步骤 1`,
+  run_in_background: false  // 同步阻塞
 })
+```

-// Assessor
+### Assessor Prompt Template
+```javascript
 Task({
  subagent_type: "general-purpose",
-  team_name: teamName,
-  name: "assessor",
  prompt: `你是 team "${teamName}" 的 ASSESSOR。

 ## ⚠️ 首要指令（MUST）
@@ -302,14 +310,15 @@ Skill(skill="team-tech-debt", args="--role=assessor")
 1. Skill(skill="team-tech-debt", args="--role=assessor") 获取角色定义
 2. TaskList → 找到 TDEVAL-* 任务 → 执行 → 汇报
 3. team_msg log + SendMessage 结果给 coordinator
-4. TaskUpdate completed → 检查下一个任务`
+4. TaskUpdate completed`,
+  run_in_background: false
 })
+```

-// Planner
+### Planner Prompt Template
+```javascript
 Task({
  subagent_type: "general-purpose",
-  team_name: teamName,
-  name: "planner",
  prompt: `你是 team "${teamName}" 的 PLANNER。

 ## ⚠️ 首要指令（MUST）
@@ -330,14 +339,15 @@ Skill(skill="team-tech-debt", args="--role=planner")
 1. Skill(skill="team-tech-debt", args="--role=planner") 获取角色定义
 2. TaskList → 找到 TDPLAN-* 任务 → 执行 → 汇报
 3. team_msg log + SendMessage 结果给 coordinator
-4. TaskUpdate completed → 检查下一个任务`
+4. TaskUpdate completed`,
+  run_in_background: false
 })
+```

-// Executor
+### Executor Prompt Template
+```javascript
 Task({
  subagent_type: "general-purpose",
-  team_name: teamName,
-  name: "executor",
  prompt: `你是 team "${teamName}" 的 EXECUTOR。

 ## ⚠️ 首要指令（MUST）
@@ -358,14 +368,15 @@ Skill(skill="team-tech-debt", args="--role=executor")
 1. Skill(skill="team-tech-debt", args="--role=executor") 获取角色定义
 2. TaskList → 找到 TDFIX-* 任务 → 执行 → 汇报
 3. team_msg log + SendMessage 结果给 coordinator
-4. TaskUpdate completed → 检查下一个任务`
+4. TaskUpdate completed`,
+  run_in_background: false
 })
+```

-// Validator
+### Validator Prompt Template
+```javascript
 Task({
  subagent_type: "general-purpose",
-  team_name: teamName,
-  name: "validator",
  prompt: `你是 team "${teamName}" 的 VALIDATOR。

 ## ⚠️ 首要指令（MUST）
@@ -386,7 +397,8 @@ Skill(skill="team-tech-debt", args="--role=validator")
 1. Skill(skill="team-tech-debt", args="--role=validator") 获取角色定义
 2. TaskList → 找到 TDVAL-* 任务 → 执行 → 汇报
 3. team_msg log + SendMessage 结果给 coordinator
-4. TaskUpdate completed → 检查下一个任务`
+4. TaskUpdate completed`,
+  run_in_background: false
 })
 ```

--- a/.claude/skills/team-tech-debt/roles/coordinator/commands/monitor.md
+++ b/.claude/skills/team-tech-debt/roles/coordinator/commands/monitor.md
@@ -1,12 +1,12 @@
 # Command: monitor

-> 阶段驱动的协调循环。按 pipeline 阶段顺序等待 worker 完成，路由消息，处理 Fix-Verify 循环，检测完成。
+> 停止等待（Stop-Wait）协调。按 pipeline 阶段顺序，逐阶段 spawn worker 同步执行，worker 返回即阶段完成，无需轮询。

 ## When to Use

 - Phase 4 of Coordinator
- 任务链已创建并分发
- 需要持续监控直到所有任务完成
+- 任务链已创建（dispatch 完成）
+- 需要逐阶段驱动 worker 执行直到所有任务完成

 **Trigger conditions**:
 - dispatch 完成后立即启动
@@ -16,199 +16,225 @@

 ### Delegation Mode

-**Mode**: Stage-driven（按阶段顺序等待，非轮询）
+**Mode**: Stop-Wait（同步阻塞 Task call，非轮询）

 ### 设计原则

-> **模型执行没有时间概念**。禁止空转 while 循环检查状态。
-> 使用固定 sleep 间隔 + 最大轮询次数，避免无意义的 API 调用浪费。
+> **模型执行没有时间概念，禁止任何形式的轮询等待。**
+>
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态（空转浪费 API 轮次）
+> - ❌ 禁止: `Bash(sleep N)` / `Bash(timeout /t N)` 作为等待手段
+> - ✅ 采用: 同步 `Task()` 调用（`run_in_background: false`），call 本身即等待
+> - ✅ 采用: Worker 返回 = 阶段完成信号（天然回调）
+>
+> **原理**: `Task(run_in_background: false)` 是阻塞调用，coordinator 自动挂起直到 worker 返回。
+> 无需 sleep，无需轮询，无需消息总线监控。Worker 的返回就是回调。

-### Decision Logic
+### Stage-Worker 映射表

 ```javascript
-// 消息路由表
-const routingTable = {
-  // Scanner 完成
-  'scan_complete':        { action: 'Mark TDSCAN complete, unblock TDEVAL' },
-  'debt_items_found':     { action: 'Mark TDSCAN complete with items, unblock TDEVAL' },
-  // Assessor 完成
-  'assessment_complete':  { action: 'Mark TDEVAL complete, unblock TDPLAN' },
-  // Planner 完成
-  'plan_ready':           { action: 'Mark TDPLAN complete, unblock TDFIX' },
-  'plan_revision':        { action: 'Plan revised, re-evaluate dependencies' },
-  // Executor 完成
-  'fix_complete':         { action: 'Mark TDFIX complete, unblock TDVAL' },
-  'fix_progress':         { action: 'Log progress, continue waiting' },
-  // Validator 完成
-  'validation_complete':  { action: 'Mark TDVAL complete, evaluate quality gate', special: 'quality_gate' },
-  'regression_found':     { action: 'Evaluate regression, decide Fix-Verify loop', special: 'fix_verify_decision' },
-  // 错误
-  'error':                { action: 'Assess severity, retry or escalate', special: 'error_handler' }
+const STAGE_WORKER_MAP = {
+  'TDSCAN': { role: 'scanner',   skillArgs: '--role=scanner' },
+  'TDEVAL': { role: 'assessor',  skillArgs: '--role=assessor' },
+  'TDPLAN': { role: 'planner',   skillArgs: '--role=planner' },
+  'TDFIX':  { role: 'executor',  skillArgs: '--role=executor' },
+  'TDVAL':  { role: 'validator', skillArgs: '--role=validator' }
 }
 ```

-### 等待策略常量
-
-```javascript
-const POLL_INTERVAL_SEC = 300  // 每次检查间隔 5 分钟
-const MAX_POLLS_PER_STAGE = 6  // 单阶段最多等待 6 次（~30 分钟）
-const SLEEP_CMD = process.platform === 'win32'
-  ? `timeout /t ${POLL_INTERVAL_SEC} /nobreak >nul 2>&1`
-  : `sleep ${POLL_INTERVAL_SEC}`
-
-// 统一 auto mode 检测
-const autoYes = /\b(-y|--yes)\b/.test(args)
-```
-
 ## Execution Steps

 ### Step 1: Context Preparation

 ```javascript
-// 从 shared memory 获取上下文
 const sharedMemory = JSON.parse(Read(`${sessionFolder}/shared-memory.json`))

 let fixVerifyIteration = 0
 const MAX_FIX_VERIFY_ITERATIONS = 3

-// 获取 pipeline 阶段列表
+// 获取 pipeline 阶段列表（按创建顺序 = 依赖顺序）
 const allTasks = TaskList()
 const pipelineTasks = allTasks
  .filter(t => t.owner && t.owner !== 'coordinator')
  .sort((a, b) => Number(a.id) - Number(b.id))
+
+// 统一 auto mode 检测
+const autoYes = /\b(-y|--yes)\b/.test(args)
 ```

-### Step 2: Stage-Driven Execution
+### Step 2: Sequential Stage Execution (Stop-Wait)

-> **核心**: 按 pipeline 阶段顺序，逐阶段等待完成。
-> 每个阶段：sleep → 检查消息 → 确认任务状态 → 处理结果 → 下一阶段。
+> **核心**: 逐阶段 spawn worker，同步阻塞等待返回。
+> Worker 返回 = 阶段完成。无 sleep、无轮询、无消息总线监控。

 ```javascript
 for (const stageTask of pipelineTasks) {
-  let stageComplete = false
-  let pollCount = 0
+  // 1. 提取阶段前缀 → 确定 worker 角色
+  const stagePrefix = stageTask.subject.match(/^(TD\w+)-/)?.[1]
+  const workerConfig = STAGE_WORKER_MAP[stagePrefix]

-  while (!stageComplete && pollCount < MAX_POLLS_PER_STAGE) {
-    Bash(SLEEP_CMD)
-    pollCount++
-
-    // 1. 检查消息总线
-    const messages = mcp__ccw-tools__team_msg({
-      operation: "list",
-      team: teamName,
-      last: 5
+  if (!workerConfig) {
+    mcp__ccw-tools__team_msg({
+      operation: "log", team: teamName, from: "coordinator",
+      to: "user", type: "error",
+      summary: `[coordinator] 未知阶段前缀: ${stagePrefix}，跳过`
    })
-
-    // 2. 路由消息
-    for (const msg of messages) {
-      const handler = routingTable[msg.type]
-      if (!handler) continue
-      processMessage(msg, handler)
-    }
-
-    // 3. 确认任务状态
-    const currentTask = TaskGet({ taskId: stageTask.id })
-    stageComplete = currentTask.status === 'completed' || currentTask.status === 'deleted'
+    continue
  }

-  // 阶段超时处理
-  if (!stageComplete) {
-    const elapsedMin = Math.round(pollCount * POLL_INTERVAL_SEC / 60)
+  // 2. 标记任务为执行中
+  TaskUpdate({ taskId: stageTask.id, status: 'in_progress' })

-    if (autoYes) {
-      mcp__ccw-tools__team_msg({
-        operation: "log", team: teamName, from: "coordinator",
-        to: "user", type: "error",
-        summary: `[coordinator] [auto] 阶段 ${stageTask.subject} 超时 (${elapsedMin}min)，自动跳过`
-      })
-      TaskUpdate({ taskId: stageTask.id, status: 'deleted' })
-      continue
-    }
+  mcp__ccw-tools__team_msg({
+    operation: "log", team: teamName, from: "coordinator",
+    to: workerConfig.role, type: "task_unblocked",
+    summary: `[coordinator] 启动阶段: ${stageTask.subject} → ${workerConfig.role}`
+  })

-    const decision = AskUserQuestion({
-      questions: [{
-        question: `阶段 "${stageTask.subject}" 已等待 ${elapsedMin} 分钟仍未完成。如何处理？`,
-        header: "Stage Wait",
-        multiSelect: false,
-        options: [
-          { label: "继续等待", description: `再等 ${MAX_POLLS_PER_STAGE} 轮` },
-          { label: "跳过此阶段", description: "标记为跳过，继续后续流水线" },
-          { label: "终止流水线", description: "停止整个流程，汇报当前结果" }
-        ]
-      }]
+  // 3. 同步 spawn worker — 阻塞直到 worker 返回（Stop-Wait 核心）
+  //    Task() 本身就是等待机制，无需 sleep/poll
+  const workerResult = Task({
+    subagent_type: "general-purpose",
+    prompt: buildWorkerPrompt(stageTask, workerConfig, sessionFolder, taskDescription),
+    run_in_background: false  // ← 同步阻塞 = 天然回调
+  })
+
+  // 4. Worker 已返回 — 直接处理结果（无需检查状态）
+  const taskState = TaskGet({ taskId: stageTask.id })
+
+  if (taskState.status !== 'completed') {
+    // Worker 返回但未标记 completed → 异常处理
+    handleStageFailure(stageTask, taskState, workerConfig, autoYes)
+  } else {
+    mcp__ccw-tools__team_msg({
+      operation: "log", team: teamName, from: "coordinator",
+      to: "user", type: "quality_gate",
+      summary: `[coordinator] 阶段完成: ${stageTask.subject}`
    })
+  }

-    const answer = decision["Stage Wait"]
-    if (answer === "跳过此阶段") {
-      TaskUpdate({ taskId: stageTask.id, status: 'deleted' })
-      continue
-    } else if (answer === "终止流水线") {
-      mcp__ccw-tools__team_msg({
-        operation: "log", team: teamName, from: "coordinator",
-        to: "user", type: "shutdown",
-        summary: `[coordinator] 用户终止流水线，当前阶段: ${stageTask.subject}`
-      })
-      break
+  // 5. 阶段间质量检查（仅 TDVAL 阶段）
+  if (stagePrefix === 'TDVAL') {
+    const needsFixVerify = evaluateValidationResult(sessionFolder)
+    if (needsFixVerify && fixVerifyIteration < MAX_FIX_VERIFY_ITERATIONS) {
+      fixVerifyIteration++
+      const fixVerifyTasks = createFixVerifyTasks(fixVerifyIteration, sessionFolder)
+      // 将 Fix-Verify 任务追加到 pipeline 末尾继续执行
+      pipelineTasks.push(...fixVerifyTasks)
    }
  }
 }
 ```

-### Step 2.1: Message Processing (processMessage)
+### Step 2.1: Worker Prompt Builder

 ```javascript
-function processMessage(msg, handler) {
-  switch (handler.special) {
-    case 'quality_gate': {
-      const latestMemory = JSON.parse(Read(`${sessionFolder}/shared-memory.json`))
-      const debtBefore = latestMemory.debt_score_before || 0
-      const debtAfter = latestMemory.debt_score_after || 0
-      const improved = debtAfter < debtBefore
+function buildWorkerPrompt(stageTask, workerConfig, sessionFolder, taskDescription) {
+  return `你是 team "${teamName}" 的 ${workerConfig.role.toUpperCase()}。

-      let status = 'PASS'
-      if (!improved && latestMemory.validation_results?.regressions > 0) status = 'FAIL'
-      else if (!improved) status = 'CONDITIONAL'
+## ⚠️ 首要指令（MUST）
+你的所有工作必须通过调用 Skill 获取角色定义后执行，禁止自行发挥：
+Skill(skill="team-tech-debt", args="${workerConfig.skillArgs}")
+此调用会加载你的角色定义（role.md）、可用命令（commands/*.md）和完整执行逻辑。

-      mcp__ccw-tools__team_msg({
-        operation: "log", team: teamName, from: "coordinator",
-        to: "user", type: "quality_gate",
-        summary: `[coordinator] 质量门控: ${status} (债务分 ${debtBefore} → ${debtAfter})`
-      })
-      break
-    }
+## 当前任务
+- 任务 ID: ${stageTask.id}
+- 任务: ${stageTask.subject}
+- 描述: ${stageTask.description || taskDescription}
+- Session: ${sessionFolder}

-    case 'fix_verify_decision': {
-      const regressions = msg.data?.regressions || 0
-      if (regressions > 0 && fixVerifyIteration < MAX_FIX_VERIFY_ITERATIONS) {
-        fixVerifyIteration++
-        mcp__ccw-tools__team_msg({
-          operation: "log", team: teamName, from: "coordinator",
-          to: "executor", type: "task_unblocked",
-          summary: `[coordinator] Fix-Verify #${fixVerifyIteration}: 发现 ${regressions} 个回归，请修复`,
-          data: { iteration: fixVerifyIteration, regressions }
-        })
-        // 创建 Fix-Verify 修复任务（参见 dispatch.md createFixVerifyTasks）
-      } else {
-        mcp__ccw-tools__team_msg({
-          operation: "log", team: teamName, from: "coordinator",
-          to: "user", type: "quality_gate",
-          summary: `[coordinator] Fix-Verify 循环已达上限(${MAX_FIX_VERIFY_ITERATIONS})，接受当前结果`
-        })
-      }
-      break
-    }
+## 角色准则（强制）
+- 你只能处理 ${stageTask.subject.match(/^(TD\w+)-/)?.[1] || 'TD'}-* 前缀的任务
+- 所有输出必须带 [${workerConfig.role}] 标识前缀
+- 仅与 coordinator 通信，不得直接联系其他 worker

-    case 'error_handler': {
-      const severity = msg.data?.severity || 'medium'
-      if (severity === 'critical') {
-        SendMessage({
-          content: `## [coordinator] Critical Error from ${msg.from}\n\n${msg.summary}`,
-          summary: `[coordinator] Critical error: ${msg.summary}`
-        })
-      }
-      break
-    }
+## 消息总线（必须）
+每次 SendMessage 前，先调用 mcp__ccw-tools__team_msg 记录。
+
+## 工作流程（严格按顺序）
+1. 调用 Skill(skill="team-tech-debt", args="${workerConfig.skillArgs}") 获取角色定义和执行逻辑
+2. 按 role.md 中的 5-Phase 流程执行
+3. team_msg log + SendMessage 结果给 coordinator
+4. TaskUpdate({ taskId: "${stageTask.id}", status: "completed" })`
+}
+```
+
+### Step 2.2: Stage Failure Handler
+
+```javascript
+function handleStageFailure(stageTask, taskState, workerConfig, autoYes) {
+  if (autoYes) {
+    mcp__ccw-tools__team_msg({
+      operation: "log", team: teamName, from: "coordinator",
+      to: "user", type: "error",
+      summary: `[coordinator] [auto] 阶段 ${stageTask.subject} 未完成 (status=${taskState.status})，自动跳过`
+    })
+    TaskUpdate({ taskId: stageTask.id, status: 'deleted' })
+    return 'skip'
  }
+
+  const decision = AskUserQuestion({
+    questions: [{
+      question: `阶段 "${stageTask.subject}" worker 返回但未完成 (status=${taskState.status})。如何处理？`,
+      header: "Stage Fail",
+      multiSelect: false,
+      options: [
+        { label: "重试", description: "重新 spawn worker 执行此阶段" },
+        { label: "跳过", description: "标记为跳过，继续后续流水线" },
+        { label: "终止", description: "停止整个流程，汇报当前结果" }
+      ]
+    }]
+  })
+
+  const answer = decision["Stage Fail"]
+  if (answer === "重试") {
+    // 重新 spawn worker（递归单次）
+    TaskUpdate({ taskId: stageTask.id, status: 'in_progress' })
+    const retryResult = Task({
+      subagent_type: "general-purpose",
+      prompt: buildWorkerPrompt(stageTask, workerConfig, sessionFolder, taskDescription),
+      run_in_background: false
+    })
+    const retryState = TaskGet({ taskId: stageTask.id })
+    if (retryState.status !== 'completed') {
+      TaskUpdate({ taskId: stageTask.id, status: 'deleted' })
+    }
+    return 'retried'
+  } else if (answer === "跳过") {
+    TaskUpdate({ taskId: stageTask.id, status: 'deleted' })
+    return 'skip'
+  } else {
+    mcp__ccw-tools__team_msg({
+      operation: "log", team: teamName, from: "coordinator",
+      to: "user", type: "shutdown",
+      summary: `[coordinator] 用户终止流水线，当前阶段: ${stageTask.subject}`
+    })
+    return 'abort'
+  }
+}
+```
+
+### Step 2.3: Validation Evaluation
+
+```javascript
+function evaluateValidationResult(sessionFolder) {
+  const latestMemory = JSON.parse(Read(`${sessionFolder}/shared-memory.json`))
+  const debtBefore = latestMemory.debt_score_before || 0
+  const debtAfter = latestMemory.debt_score_after || 0
+  const regressions = latestMemory.validation_results?.regressions || 0
+  const improved = debtAfter < debtBefore
+
+  let status = 'PASS'
+  if (!improved && regressions > 0) status = 'FAIL'
+  else if (!improved) status = 'CONDITIONAL'
+
+  mcp__ccw-tools__team_msg({
+    operation: "log", team: teamName, from: "coordinator",
+    to: "user", type: "quality_gate",
+    summary: `[coordinator] 质量门控: ${status} (债务分 ${debtBefore} → ${debtAfter}, 回归 ${regressions})`
+  })
+
+  return regressions > 0
 }
 ```

@@ -237,19 +263,15 @@ const summary = {
 ### Tasks: [completed]/[total]
 ### Fix-Verify Iterations: [count]
 ### Debt Score: [before] → [after]
-
-### Message Log (last 10)
- [timestamp] [from] → [to]: [type] - [summary]
 ```

 ## Error Handling

 | Scenario | Resolution |
 |----------|------------|
-| Message bus unavailable | Fall back to TaskList polling only |
-| Stage timeout (交互模式) | AskUserQuestion: 继续等待 / 跳过 / 终止 |
-| Stage timeout (自动模式 `-y`/`--yes`) | 自动跳过，记录日志 |
-| Teammate unresponsive (2x no response) | Respawn teammate with same task |
-| Deadlock detected | Identify cycle, manually unblock |
+| Worker 返回但未 completed (交互模式) | AskUserQuestion: 重试 / 跳过 / 终止 |
+| Worker 返回但未 completed (自动模式) | 自动跳过，记录日志 |
+| Worker spawn 失败 | 重试一次，仍失败则上报用户 |
 | Quality gate FAIL | Report to user, suggest targeted re-run |
 | Fix-Verify loop stuck >3 iterations | Accept current state, continue pipeline |
+| Shared memory 读取失败 | 降级为 TaskList 状态判断 |
--- a/.claude/skills/team-tech-debt/roles/coordinator/role.md
+++ b/.claude/skills/team-tech-debt/roles/coordinator/role.md
@@ -121,7 +121,7 @@ if (!autoYes && (!taskDescription || taskDescription.length < 10)) {
 }
 ```

-### Phase 2: Create Team + Spawn Teammates
+### Phase 2: Create Team + Initialize Session

 ```javascript
 const teamName = "tech-debt"
@@ -143,8 +143,9 @@ Write(`${sessionFolder}/shared-memory.json`, JSON.stringify({

 TeamCreate({ team_name: teamName })

-// Spawn teammates (see SKILL.md Coordinator Spawn Template)
-// Scanner, Assessor, Planner, Executor, Validator
+// ⚠️ 不在此阶段 spawn worker
+// Worker 在 Phase 4 (monitor) 中按阶段按需 spawn（Stop-Wait 策略）
+// 这避免了 worker 先启动但无任务可做的鸡生蛋问题
 ```

 ### Phase 3: Create Task Chain
@@ -171,28 +172,33 @@ TDSCAN-001(扫描) → TDEVAL-001(评估) → TDPLAN-001(规划) → TDFIX-001(
 TDPLAN-001(规划) → TDFIX-001(修复) → TDVAL-001(验证)
 ```

-### Phase 4: Coordination Loop
+### Phase 4: Sequential Stage Execution (Stop-Wait)

 ```javascript
 // Read commands/monitor.md for full implementation
 Read("commands/monitor.md")
 ```

-| Received Message | Action |
-|-----------------|--------|
-| `scan_complete` | 标记 TDSCAN complete → 解锁 TDEVAL |
-| `assessment_complete` | 标记 TDEVAL complete → 解锁 TDPLAN |
-| `plan_ready` | 标记 TDPLAN complete → 解锁 TDFIX |
-| `fix_complete` | 标记 TDFIX complete → 解锁 TDVAL |
-| `validation_complete` | 标记 TDVAL complete → 评估质量门控 |
-| `regression_found` | 评估回归 → 触发 Fix-Verify 循环（max 3） |
-| Worker: `error` | 评估严重性 → 重试或上报用户 |
+> **策略**: 逐阶段 spawn worker，同步阻塞等待返回。Worker 返回即阶段完成，无需轮询。
+>
+> - ❌ 禁止: while 循环 + sleep + 检查状态
+> - ✅ 采用: `Task(run_in_background: false)` 同步调用 = 天然回调

-**Fix-Verify 循环逻辑**:
+**阶段流转**:
+
+| 当前阶段 | Worker | 完成后 |
+|----------|--------|--------|
+| TDSCAN-001 | scanner | → 启动 TDEVAL |
+| TDEVAL-001 | assessor | → 启动 TDPLAN |
+| TDPLAN-001 | planner | → 启动 TDFIX |
+| TDFIX-001 | executor | → 启动 TDVAL |
+| TDVAL-001 | validator | → 评估质量门控 |
+
+**Fix-Verify 循环**（TDVAL 阶段发现回归时）:
 ```javascript
 if (regressionFound && fixVerifyIteration < 3) {
  fixVerifyIteration++
-  // 创建 TDFIX-fix 任务 → TDVAL 重新验证
+  // 创建 TDFIX-fix + TDVAL-verify 任务，追加到 pipeline 继续执行
 } else if (fixVerifyIteration >= 3) {
  // 接受当前状态，继续汇报
  mcp__ccw-tools__team_msg({
--- a/.claude/skills/team-testing/SKILL.md
+++ b/.claude/skills/team-testing/SKILL.md
@@ -103,7 +103,6 @@ mcp__ccw-tools__team_msg({ summary: `[${role}] ...` })
 const TEAM_CONFIG = {
  name: "testing",
  sessionDir: ".workflow/.team/TST-{slug}-{date}/",
-  msgDir: ".workflow/.team-msg/testing/",
  sharedMemory: "shared-memory.json",
  testLayers: {
    L1: { name: "Unit Tests", coverage_target: 80 },
--- a/.claude/skills/team-testing/roles/coordinator.md
+++ b/.claude/skills/team-testing/roles/coordinator.md
@@ -86,7 +86,7 @@ AskUserQuestion({
 })
 ```

-### Phase 2: Create Team + Spawn Workers
+### Phase 2: Create Team + Initialize Session

 ```javascript
 TeamCreate({ team_name: teamName })
@@ -130,7 +130,9 @@ const teamSession = {
 Write(`${sessionFolder}/team-session.json`, JSON.stringify(teamSession, null, 2))
 ```

-Spawn workers (see SKILL.md Coordinator Spawn Template).
+// ⚠️ Workers are NOT pre-spawned here.
+// Workers are spawned per-stage in Phase 4 via Stop-Wait Task(run_in_background: false).
+// See SKILL.md Coordinator Spawn Template for worker prompt templates.

 ### Phase 3: Create Task Chain

@@ -178,6 +180,13 @@ TaskUpdate({ taskId: anaId, owner: "analyst", addBlockedBy: [run2Id] })

 ### Phase 4: Coordination Loop + Generator-Critic Control

+> **设计原则（Stop-Wait）**: 模型执行没有时间概念，禁止任何形式的轮询等待。
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态
+> - ✅ 采用: 同步 `Task(run_in_background: false)` 调用，Worker 返回 = 阶段完成信号
+>
+> 按 Phase 3 创建的任务链顺序，逐阶段 spawn worker 同步执行。
+> Worker prompt 使用 SKILL.md Coordinator Spawn Template。
+
 | Received Message | Action |
 |-----------------|--------|
 | strategist: strategy_ready | Read strategy → team_msg log → TaskUpdate completed |
--- a/.claude/skills/team-uidesign/roles/coordinator.md
+++ b/.claude/skills/team-uidesign/roles/coordinator.md
@@ -125,7 +125,7 @@ const industryConfig = {
 }[industryChoice] || { strictness: 'standard', mustHave: [] }
 ```

-### Phase 2: Create Team + Spawn Workers
+### Phase 2: Create Team + Initialize Session

 ```javascript
 TeamCreate({ team_name: teamName })
@@ -174,7 +174,9 @@ const teamSession = {
 }
 Write(`${sessionFolder}/team-session.json`, JSON.stringify(teamSession, null, 2))

-// Spawn 4 workers (see SKILL.md Coordinator Spawn Template)
+// ⚠️ Workers are NOT pre-spawned here.
+// Workers are spawned per-stage in Phase 4 via Stop-Wait Task(run_in_background: false).
+// See SKILL.md Coordinator Spawn Template for worker prompt templates.
 ```

 ### Phase 3: Create Task Chain
@@ -246,6 +248,13 @@ TaskUpdate({ taskId: audit3Id, owner: "reviewer", addBlockedBy: [build2Id] })

 ### Phase 4: Coordination Loop

+> **设计原则（Stop-Wait）**: 模型执行没有时间概念，禁止任何形式的轮询等待。
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态
+> - ✅ 采用: 同步 `Task(run_in_background: false)` 调用，Worker 返回 = 阶段完成信号
+>
+> 按 Phase 3 创建的任务链顺序，逐阶段 spawn worker 同步执行。
+> Worker prompt 使用 SKILL.md Coordinator Spawn Template。
+
 #### Message Handling

 | Received Message | Action |
--- a/.claude/skills/team-ultra-analyze/roles/coordinator/commands/monitor.md
+++ b/.claude/skills/team-ultra-analyze/roles/coordinator/commands/monitor.md
@@ -20,8 +20,15 @@

 ### 设计原则

-> **模型执行没有时间概念**。禁止空转 while 循环检查状态。
-> 使用固定 sleep 间隔 + 最大轮询次数，避免无意义的 API 调用浪费。
+> **模型执行没有时间概念，禁止任何形式的轮询等待。**
+>
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态（空转浪费 API 轮次）
+> - ❌ 禁止: `Bash(sleep N)` / `Bash(timeout /t N)` 作为等待手段
+> - ✅ 采用: 同步 `Task()` 调用（`run_in_background: false`），call 本身即等待
+> - ✅ 采用: Worker 返回 = 阶段完成信号（天然回调）
+>
+> **原理**: `Task(run_in_background: false)` 是阻塞调用，coordinator 自动挂起直到 worker 返回。
+> 无需 sleep，无需轮询，无需消息总线监控。Worker 的返回就是回调。

 ### Decision Logic

@@ -41,14 +48,15 @@ const routingTable = {
 }
 ```

-### 等待策略常量
+### Stage-Worker 映射表

 ```javascript
-const POLL_INTERVAL_SEC = 300  // 每次检查间隔 5 分钟
-const MAX_POLLS_PER_STAGE = 6  // 单阶段最多等待 6 次（~30 分钟）
-const SLEEP_CMD = process.platform === 'win32'
-  ? `timeout /t ${POLL_INTERVAL_SEC} /nobreak >nul 2>&1`
-  : `sleep ${POLL_INTERVAL_SEC}`
+const STAGE_WORKER_MAP = {
+  'EXPLORE': { role: 'explorer',    skillArgs: '--role=explorer' },
+  'ANALYZE': { role: 'analyst',     skillArgs: '--role=analyst' },
+  'DISCUSS': { role: 'discussant',  skillArgs: '--role=discussant' },
+  'SYNTH':   { role: 'synthesizer', skillArgs: '--role=synthesizer' }
+}

 // ★ 统一 auto mode 检测
 const autoYes = /\b(-y|--yes)\b/.test(args)
@@ -72,9 +80,10 @@ const pipelineTasks = allTasks
  .sort((a, b) => Number(a.id) - Number(b.id))
 ```

-### Step 2: Stage-Driven Execution (Exploration + Analysis)
+### Step 2: Sequential Stage Execution (Stop-Wait) — Exploration + Analysis

-> 按 pipeline 阶段顺序，逐阶段等待完成。
+> **核心**: 逐阶段 spawn worker，同步阻塞等待返回。
+> Worker 返回 = 阶段完成。无 sleep、无轮询、无消息总线监控。

 ```javascript
 // 处理 EXPLORE 和 ANALYZE 阶段
@@ -83,33 +92,57 @@ const preDiscussionTasks = pipelineTasks.filter(t =>
 )

 for (const stageTask of preDiscussionTasks) {
-  let stageComplete = false
-  let pollCount = 0
+  // 1. 提取阶段前缀 → 确定 worker 角色
+  const stagePrefix = stageTask.subject.match(/^(\w+)-/)?.[1]
+  const workerConfig = STAGE_WORKER_MAP[stagePrefix]

-  while (!stageComplete && pollCount < MAX_POLLS_PER_STAGE) {
-    Bash(SLEEP_CMD)
-    pollCount++
+  if (!workerConfig) continue

-    // 1. 检查消息总线
-    const messages = mcp__ccw-tools__team_msg({
-      operation: "list", team: teamName, last: 5
+  // 2. 标记任务为执行中
+  TaskUpdate({ taskId: stageTask.id, status: 'in_progress' })
+
+  mcp__ccw-tools__team_msg({
+    operation: "log", team: teamName, from: "coordinator",
+    to: workerConfig.role, type: "task_unblocked",
+    summary: `[coordinator] 启动阶段: ${stageTask.subject} → ${workerConfig.role}`
+  })
+
+  // 3. 同步 spawn worker — 阻塞直到 worker 返回（Stop-Wait 核心）
+  const workerResult = Task({
+    subagent_type: "general-purpose",
+    prompt: `你是 team "${teamName}" 的 ${workerConfig.role.toUpperCase()}。
+
+## ⚠️ 首要指令（MUST）
+Skill(skill="team-ultra-analyze", args="${workerConfig.skillArgs}")
+
+## 当前任务
+- 任务 ID: ${stageTask.id}
+- 任务: ${stageTask.subject}
+- 描述: ${stageTask.description || taskDescription}
+- Session: ${sessionFolder}
+
+## 角色准则（强制）
+- 所有输出必须带 [${workerConfig.role}] 标识前缀
+- 仅与 coordinator 通信
+
+## 工作流程
+1. Skill(skill="team-ultra-analyze", args="${workerConfig.skillArgs}") 获取角色定义
+2. 执行任务 → 汇报结果
+3. TaskUpdate({ taskId: "${stageTask.id}", status: "completed" })`,
+    run_in_background: false
+  })
+
+  // 4. Worker 已返回 — 检查结果
+  const taskState = TaskGet({ taskId: stageTask.id })
+
+  if (taskState.status !== 'completed') {
+    handleStageTimeout(stageTask, 0, autoYes)
+  } else {
+    mcp__ccw-tools__team_msg({
+      operation: "log", team: teamName, from: "coordinator",
+      to: "user", type: "quality_gate",
+      summary: `[coordinator] 阶段完成: ${stageTask.subject}`
    })
-
-    // 2. 路由消息
-    for (const msg of messages) {
-      const handler = routingTable[msg.type]
-      if (!handler) continue
-      processMessage(msg, handler)
-    }
-
-    // 3. 确认任务状态（兜底）
-    const currentTask = TaskGet({ taskId: stageTask.id })
-    stageComplete = currentTask.status === 'completed' || currentTask.status === 'deleted'
-  }
-
-  // 阶段超时处理
-  if (!stageComplete) {
-    handleStageTimeout(stageTask, pollCount, autoYes)
  }
 }
 ```
@@ -164,9 +197,21 @@ if (MAX_DISCUSSION_ROUNDS === 0) {
  // Then enter discussion loop

  while (discussionRound < MAX_DISCUSSION_ROUNDS) {
-    // 等待当前 DISCUSS 任务完成
+    // 等待当前 DISCUSS 任务完成（Stop-Wait: spawn discussant worker）
    const currentDiscussId = `DISCUSS-${String(discussionRound + 1).padStart(3, '0')}`
-    // ... wait for completion (same pattern as Step 2)
+    const discussTask = pipelineTasks.find(t => t.subject.startsWith(currentDiscussId))
+    if (discussTask) {
+      TaskUpdate({ taskId: discussTask.id, status: 'in_progress' })
+      const discussResult = Task({
+        subagent_type: "general-purpose",
+        prompt: `你是 team "${teamName}" 的 DISCUSSANT。
+Skill(skill="team-ultra-analyze", args="--role=discussant")
+当前任务: ${discussTask.subject}
+Session: ${sessionFolder}
+TaskUpdate({ taskId: "${discussTask.id}", status: "completed" })`,
+        run_in_background: false
+      })
+    }

    // 收集用户反馈
    const feedbackResult = AskUserQuestion({
@@ -291,14 +336,12 @@ ${data.updated_understanding || '(Updated by discussant)'}
  Write(`${sessionFolder}/discussion.md`, currentContent + roundContent)
 }

-function handleStageTimeout(stageTask, pollCount, autoYes) {
-  const elapsedMin = Math.round(pollCount * POLL_INTERVAL_SEC / 60)
-
+function handleStageTimeout(stageTask, _unused, autoYes) {
  if (autoYes) {
    mcp__ccw-tools__team_msg({
      operation: "log", team: teamName, from: "coordinator",
      to: "user", type: "error",
-      summary: `[coordinator] [auto] 阶段 ${stageTask.subject} 超时 (${elapsedMin}min)，自动跳过`
+      summary: `[coordinator] [auto] 阶段 ${stageTask.subject} worker 返回但未完成，自动跳过`
    })
    TaskUpdate({ taskId: stageTask.id, status: 'deleted' })
    return
@@ -306,18 +349,18 @@ function handleStageTimeout(stageTask, pollCount, autoYes) {

  const decision = AskUserQuestion({
    questions: [{
-      question: `阶段 "${stageTask.subject}" 已等待 ${elapsedMin} 分钟仍未完成。如何处理？`,
-      header: "Stage Wait",
+      question: `阶段 "${stageTask.subject}" worker 返回但未完成。如何处理？`,
+      header: "Stage Fail",
      multiSelect: false,
      options: [
-        { label: "继续等待", description: `再等 ${MAX_POLLS_PER_STAGE} 轮` },
+        { label: "重试", description: "重新 spawn worker 执行此阶段" },
        { label: "跳过此阶段", description: "标记为跳过，继续后续流水线" },
        { label: "终止流水线", description: "停止整个分析流程" }
      ]
    }]
  })

-  const answer = decision["Stage Wait"]
+  const answer = decision["Stage Fail"]
  if (answer === "跳过此阶段") {
    TaskUpdate({ taskId: stageTask.id, status: 'deleted' })
  } else if (answer === "终止流水线") {
@@ -333,8 +376,20 @@ function handleStageTimeout(stageTask, pollCount, autoYes) {
 ### Step 4: Wait for Synthesis + Result Processing

 ```javascript
-// 等待 SYNTH-001 完成
-// ... same wait pattern
+// 等待 SYNTH-001 完成（Stop-Wait: spawn synthesizer worker）
+const synthTask = pipelineTasks.find(t => t.subject.startsWith('SYNTH-'))
+if (synthTask) {
+  TaskUpdate({ taskId: synthTask.id, status: 'in_progress' })
+  const synthResult = Task({
+    subagent_type: "general-purpose",
+    prompt: `你是 team "${teamName}" 的 SYNTHESIZER。
+Skill(skill="team-ultra-analyze", args="--role=synthesizer")
+当前任务: ${synthTask.subject}
+Session: ${sessionFolder}
+TaskUpdate({ taskId: "${synthTask.id}", status: "completed" })`,
+    run_in_background: false
+  })
+}

 // 汇总所有结果
 const finalMemory = JSON.parse(Read(`${sessionFolder}/shared-memory.json`))
@@ -368,9 +423,8 @@ const summary = {

 | Scenario | Resolution |
 |----------|------------|
-| Message bus unavailable | Fall back to TaskList polling only |
-| Stage timeout (交互模式) | AskUserQuestion：继续等待 / 跳过 / 终止 |
-| Stage timeout (自动模式) | 自动跳过，记录日志 |
-| Teammate unresponsive (2x) | Respawn teammate with same task |
+| Worker 返回但未 completed (交互模式) | AskUserQuestion: 重试 / 跳过 / 终止 |
+| Worker 返回但未 completed (自动模式) | 自动跳过，记录日志 |
+| Worker spawn 失败 | 重试一次，仍失败则上报用户 |
 | Discussion loop stuck >5 rounds | Force synthesis, offer continuation |
 | Synthesis fails | Report partial results from analyses |
--- a/.claude/skills/team-ultra-analyze/roles/coordinator/role.md
+++ b/.claude/skills/team-ultra-analyze/roles/coordinator/role.md
@@ -163,7 +163,7 @@ if (!autoYes) {
 }
 ```

-### Phase 2: Create Team + Spawn Teammates
+### Phase 2: Create Team + Initialize Session

 ```javascript
 const teamName = "ultra-analyze"
@@ -210,7 +210,9 @@ Write(`${sessionFolder}/discussion.md`, `# Analysis Discussion

 TeamCreate({ team_name: teamName })

-// Spawn teammates (see SKILL.md Coordinator Spawn Template)
+// ⚠️ Workers are NOT pre-spawned here.
+// Workers are spawned per-stage in Phase 4 via Stop-Wait Task(run_in_background: false).
+// See SKILL.md Coordinator Spawn Template for worker prompt templates.
 // Quick mode: 1 explorer + 1 analyst (single agents)
 // Standard/Deep mode: N explorers + N analysts (parallel agents with distinct names)
 // explorer-1, explorer-2... / analyst-1, analyst-2... for true parallel execution
@@ -243,6 +245,13 @@ EXPLORE-001 → ANALYZE-001 → SYNTH-001

 ### Phase 4: Discussion Loop + Coordination

+> **设计原则（Stop-Wait）**: 模型执行没有时间概念，禁止任何形式的轮询等待。
+> - ❌ 禁止: `while` 循环 + `sleep` + 检查状态
+> - ✅ 采用: 同步 `Task(run_in_background: false)` 调用，Worker 返回 = 阶段完成信号
+>
+> 按 Phase 3 创建的任务链顺序，逐阶段 spawn worker 同步执行。
+> Worker prompt 使用 SKILL.md Coordinator Spawn Template。
+
 ```javascript
 // Read commands/monitor.md for full implementation
 Read("commands/monitor.md")