speccrew 0.7.17 → 0.7.19

package/.speccrew/agents/speccrew-product-manager.md

@@ -51,6 +51,54 @@ When involving related domains:
 - `{workspace_path}/knowledge/domain/glossary/` → Business terminology glossary
 - `{workspace_path}/knowledge/domain/qa/` → Common problem solutions
 
+# 🛑 CRITICAL: dispatch-to-worker 执行协议
+
+### 定义
+当 orchestration workflow 中出现 `action="dispatch-to-worker"` 时：
+
+**你（PM Agent）必须：**
+1. 使用 **Agent tool** 创建一个新的子 Agent
+2. 子 Agent 角色指定为 **speccrew-task-worker**
+3. 在 Task 描述中传递 Skill 名称和所有 context 参数
+4. **等待 Worker 完成**后再继续下一个 block
+
+**你（PM Agent）绝对不能：**
+- ❌ 用 Skill tool 直接调用 Phase Skill
+- ❌ 自己运行脚本（包括 update-progress.js）
+- ❌ 自己读写业务文件（如 .clarification-summary.md）
+- ❌ 把 "dispatch" 理解为 "自己执行"
+
+### 正确 vs 错误示例
+
+**❌ 错误 — PM 自己执行：**
+```
+PM 读取需求文件 → PM 生成澄清总结 → PM 运行 update-progress.js
+```
+
+**✅ 正确 — PM dispatch 给 Worker：**
+```
+PM 用 Agent tool 创建 speccrew-task-worker 子 Agent
+  → 传递: skill=speccrew-pm-requirement-clarify, context={...}
+  → Worker 加载 Skill 并执行所有步骤
+  → Worker 返回结果给 PM
+PM 继续下一个 orchestration block
+```
+
+### 适用范围：ALL Phases (0-6)
+
+| Phase | Skill 名称 | dispatch? |
+|-------|-----------|-----------|
+| Phase 0 | speccrew-pm-phase0-init | ✅ dispatch-to-worker |
+| Phase 1 | speccrew-pm-phase1-knowledge-check | ✅ dispatch-to-worker |
+| Phase 2 | speccrew-pm-phase2-complexity-assess | ✅ dispatch-to-worker |
+| Phase 3 | speccrew-pm-requirement-clarify | ✅ dispatch-to-worker |
+| Phase 4a | speccrew-pm-requirement-model | ✅ dispatch-to-worker |
+| Phase 4b | speccrew-pm-requirement-analysis | ✅ dispatch-to-worker |
+| Phase 5 | speccrew-pm-phase5-subprd-dispatch | ✅ dispatch-to-worker |
+| Phase 6 | speccrew-pm-phase6-verify-confirm | ✅ dispatch-to-worker |
+
+---
+
 # Workflow
 
 ## AgentFlow Definition
@@ -241,8 +289,9 @@ After Phase 1 completes, store:
 
 ## Phase 4: Invoke PRD Skill
 
-> ⚠️ **PM AGENT ORCHESTRATION PRINCIPLE (Phase 4-6)**
+> ⚠️ **PM AGENT ORCHESTRATION PRINCIPLE (ALL PHASES 0-6)**
 > You are the ORCHESTRATOR, NOT the WRITER:
+> - Phase 0-3: DO NOT run scripts or create files yourself → Dispatch Worker
 > - Phase 4a (Model): DO NOT do ISA-95 analysis yourself → Dispatch Worker
 > - Phase 4a.5 (Confirm): MUST stop for user confirmation after module design
 > - Phase 4b (Generate): DO NOT generate Master PRD yourself → Dispatch Worker
@@ -597,12 +646,20 @@ Phase 0 → Phase 1 → Phase 2 → Phase 3 → Phase 4 → Phase 5 (if complex)
 - **NEVER assume requirement completeness** — clarification skill handles all verification
 - **IF clarification skill fails: ABORT** — do NOT generate clarification yourself
 
-## MANDATORY WORKER DISPATCH RULE
+## MANDATORY WORKER DISPATCH RULE (ALL PHASES)
+
+🛑 **UNIVERSAL**: PM Agent 在所有 Phase (0-6) 中都必须通过 Agent tool 创建 speccrew-task-worker 子 Agent 来执行 Skill。
+
+**执行方式：**
+- 每个 Phase Skill 都通过 Agent tool 创建子 Agent（speccrew-task-worker）
+- Worker Agent 接收 Skill 名称和 context 参数后自行加载执行
+- PM Agent 等待 Worker 完成后继续 orchestration 流程
 
-- **For complex requirements (3+ modules): Phase 5 is MANDATORY**
-- **MUST dispatch `speccrew-task-worker` with `speccrew-pm-sub-prd-generate` for each Sub-PRD**
-- **DO NOT generate Sub-PRDs yourself** — you are the orchestrator, not the writer
-- **MUST use `update-progress.js` for all progress file operations**
+**禁止行为：**
+- ❌ PM 直接执行任何 Phase Skill
+- ❌ PM 直接运行脚本（update-progress.js 等）
+- ❌ PM 直接创建/修改业务文档（.clarification-summary.md、.module-design.md 等）
+- ❌ PM 用 Skill tool 调用 Phase Skill（必须用 Agent tool）
 
 ## MANDATORY TEMPLATE PATH
 

package/.speccrew/skills/speccrew-pm-requirement-clarify/SKILL.md

@@ -85,3 +85,12 @@ Applies ISA-95 Stage 1 (Domain Description) for clarification:
 - Use chat for complex requirements
 - Manually write JSON files
 - Auto-proceed to Phase 4 (PRD generation) without PM Agent's user confirmation gate
+- Auto-pass sufficiency checks without actual user answers
+- Write checkpoints in this Skill — checkpoints are managed by orchestration layer
+
+### MANDATORY: User Answer Verification Rule
+
+- Worker MUST NOT auto-pass sufficiency checks without actual user answers
+- Each clarification round MUST wait for user to fill in answers before proceeding
+- Checkpoint writing is FORBIDDEN in this Skill — checkpoints are managed by the orchestration layer
+- The sufficiency check result is ONLY valid when based on real user-provided answers

package/.speccrew/skills/speccrew-pm-requirement-clarify/workflow.agentflow.xml

@@ -118,8 +118,15 @@
             <field name="output" var="questions_file"/>
           </block>
 
-          <block type="event" id="E1" action="confirm" title="Clarification Round ${round_number}" type="wait" desc="Wait for user to answer questions">
-            <field name="message">Round ${round_number} questions written. Please fill answers and notify me.</field>
+          <!-- MANDATORY: Wait for user to answer clarification questions -->
+          <block type="event" id="E-ROUND-CONFIRM" action="user-confirm" desc="Wait for user to answer clarification questions">
+            <field name="prompt">📋 第 ${round_number} 轮澄清问题已生成: ${questions_file}
+
+请：
+1. 查看澄清问题文件
+2. 在文件中填写您的答案
+3. 填写完成后确认继续</field>
+            <field name="skippable" value="false"/>
           </block>
 
           <block type="task" id="B8" action="read-file" desc="Read answered questions">
@@ -127,10 +134,18 @@
             <field name="output" var="answered_questions"/>
           </block>
 
+          <!-- MANDATORY: User answers are required for sufficiency check -->
+          <block type="rule" id="R-ANSWER-REQUIRED" level="mandatory" desc="User answers are required for sufficiency check">
+            <field name="text">MANDATORY: Sufficiency check can ONLY pass when the user has actually provided answers to the clarification questions.</field>
+            <field name="text">If the clarification questions file still contains only the original template without user answers, ALL checks MUST fail.</field>
+            <field name="text">Worker MUST NOT self-evaluate and auto-pass the sufficiency check without real user input.</field>
+          </block>
+
           <!-- Sufficiency Check -->
           <block type="task" id="B9" action="analyze" desc="Perform sufficiency check">
             <field name="input" value="${answered_questions}"/>
             <field name="checks">
+              0. User has actually provided answers - The answered_questions content contains real user responses, not just the original template
               1. Scope boundaries are clear - Explicit in/out scope defined
               2. Key business rules are documented - At least 1 rule identified
               3. Priority/phasing decisions are made - Phase 1 scope is clear
@@ -177,12 +192,6 @@
       <field name="output" var="summary_path"/>
     </block>
 
-    <!-- Step 7: Initialize .checkpoints.json -->
-    <block type="task" id="B11" action="run-script" desc="Initialize checkpoints via script">
-      <field name="command">node "${workspace_path}/scripts/update-progress.js" write-checkpoint --file "${iteration_path}/01.product-requirement/.checkpoints.json" --checkpoint requirement_clarification --passed true</field>
-      <field name="output" var="checkpoint_result"/>
-    </block>
-
     <!-- Checkpoint: Clarification complete -->
     <block type="checkpoint" id="CP1" name="clarification-complete" desc="Verify clarification completed">
       <field name="file" value="${summary_path}"/>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "speccrew",
-  "version": "0.7.17",
+  "version": "0.7.19",
   "description": "Spec-Driven Development toolkit for AI-powered IDEs",
   "author": "charlesmu99",
   "repository": {