cc-devflow 4.1.5 → 4.2.0

package/.claude/skills/workflow/flow-dev/CLAUDE.md

@@ -1,78 +1,16 @@
-# flow-dev/ - Development Execution Skill
-
-> L2 | 父级: `.claude/skills/workflow/CLAUDE.md`
-
-## Purpose
-
-执行开发任务，遵循 TDD Iron Law，默认 Autonomous 模式。
-
-## Members
-
-| File | Purpose | Lines |
-|------|---------|-------|
-| `SKILL.md` | 核心指令：TDD 执行流程、模式、错误处理 | ~100 |
-| `context.jsonl` | 通用上下文注入定义 (fallback) | ~10 |
-| `dev-implementer.jsonl` | dev-implementer Agent 专用上下文 [NEW: v4.4] | ~10 |
-| `scripts/entry-gate.sh` | 入口检查：TASKS.md 存在、状态验证 | ~100 |
-| `scripts/exit-gate.sh` | 出口检查：任务完成度、测试验证 | ~100 |
-| `scripts/task-orchestrator.sh` | 任务调度：解析 TASKS.md、标记完成 | ~100 |
-| `references/dev-implementer.md` | Agent 指令链接 | symlink |
-| `assets/IMPLEMENTATION_PLAN_TEMPLATE.md` | 实现计划模板 | ~80 |
-
-## Context Injection (v4.4)
-
-Hook `inject-agent-context.ts` 在 Task 调用前自动注入上下文：
-
-```
-Task(subagent_type="dev-implementer", prompt="...")
-    ↓
-Hook 检测到 dev-implementer
-    ↓
-读取 dev-implementer.jsonl
-    ↓
-解析 JSONL，加载文件内容
-    ↓
-注入到 prompt 中
-```
-
-JSONL 格式 (Trellis 风格):
-```jsonl
-{"file": "devflow/requirements/{REQ}/TASKS.md", "reason": "Task list with DoD"}
-{"file": "devflow/spec/frontend/index.md", "reason": "Frontend conventions", "optional": true}
-```
-
-## TDD Iron Law
-
-```
-NO PRODUCTION CODE WITHOUT A FAILING TEST FIRST
-```
-
-## Execution Flow
-
-```
-Entry Gate → Task Loop (TDD) → Exit Gate
-
-For each task:
-  1. Protocol 2: 读取任务 DoD
-  2. Write failing test (Phase 2)
-  3. Verify test fails
-  4. Write implementation (Phase 3)
-  5. Verify test passes
-  6. Mark task [x]
-  7. Protocol 3: 读取下一任务 + ERROR_LOG
-```
-
-## Mode Matrix
-
-| Mode | Flag | Behavior |
-|------|------|----------|
-| Autonomous | (default) | 自动重试，迭代直到完成 |
-| Manual | `--manual` | 遇错停止，等待用户 |
-
-## Agent Reference
-
-- `references/dev-implementer.md` → `.claude/agents/dev-implementer.md`
-
----
+# flow-dev/
+> L2 | 父级: /Users/dimon/001Area/80-CodeWorld/002-devflow/cc-devflow/.claude/skills/workflow/CLAUDE.md
+
+成员清单
+SKILL.md: flow:dev 的核心指令，定义 dispatch/resume 的参数与失败恢复策略。
+context.jsonl: flow-dev 上下文注入定义（task-manifest/context-package/runtime）。
+dev-implementer.jsonl: 旧 Agent 上下文（兼容保留，非主链关键路径）。
+scripts/entry-gate.sh: 旧入口脚本资产（兼容保留）。
+scripts/exit-gate.sh: 旧出口脚本资产（兼容保留）。
+scripts/task-orchestrator.sh: 旧任务调度脚本资产（兼容保留）。
+assets/IMPLEMENTATION_PLAN_TEMPLATE.md: 旧实现计划模板资产（兼容保留）。
+references/dev-implementer.md: 旧 Agent 指令引用（兼容保留）。
+
+法则: 成员完整·一行一文件·父级链接·技术词前置
 
 [PROTOCOL]: 变更时更新此头部，然后检查 CLAUDE.md

package/.claude/skills/workflow/flow-dev/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: flow-dev
-description: 'Execute development tasks with TDD and Autonomous mode. Usage: /flow-dev "REQ-123" [--manual] [--max-iterations N]'
+description: 'Execute task-manifest with dependency-aware parallel dispatch and checkpoint recovery. Use when implementing planned tasks for a requirement.'
 ---
 
 # Flow-Dev Skill
@@ -9,88 +9,86 @@ description: 'Execute development tasks with TDD and Autonomous mode. Usage: /fl
 
 ## Purpose
 
-执行开发任务，遵循 TDD Iron Law，默认 Autonomous 模式。
+执行 `task-manifest.json` 中的任务，默认并行调度并写入 checkpoint/events 供恢复。
 
-## TDD Iron Law
+## Long-Running Harness Protocol（Initializer/Worker/Done Gate）
 
-```
-NO PRODUCTION CODE WITHOUT A FAILING TEST FIRST
-```
+`/flow:dev` 是最长链路阶段，必须严格执行“先读状态、再增量、后留证据”。
 
-### TDD Checkpoint
+### Session Start（禁止盲跑）
 
-```yaml
-⚠️ TEST VERIFICATION CHECKPOINT:
-  1. 运行所有 Phase 2 测试
-  2. 验证所有新测试 FAIL
-  3. 只有全部 FAIL 后才能进入 Phase 3
-  4. 如果已经写了实现代码 → DELETE 重来
-```
+每个窗口开始时先同步以下信息：
+- `devflow/requirements/${REQ_ID}/session-checklist.json`（若存在）
+- `devflow/requirements/${REQ_ID}/session-progress.md`（若存在）
+- `devflow/requirements/${REQ_ID}/session-handoff.md`（若存在）
+- `devflow/requirements/${REQ_ID}/task-manifest.json`
+- `.harness/runtime/${REQ_ID}/**/checkpoint.json` 与 `events.jsonl`（若存在）
 
-## Input Format
+然后执行一次基础健康检查：
+1. 不存在损坏的 `running` 残留状态
+2. 上轮失败任务与失败原因可定位
 
-```
-/flow-dev "REQ_ID" [--manual] [--max-iterations N] [--task T###]
-```
+### Worker Session（最小可执行前沿）
 
-- **默认**: Autonomous 模式 (自动重试)
-- **--manual**: Manual 模式 (遇错停止)
-- **--max-iterations**: 最大迭代次数 (默认 10)
-- **--task**: 从指定任务开始
+单窗口仅推进一个最小前沿：
+- 仅处理当前 `dependsOn` 已满足的一组任务
+- 每完成一组任务即写入 checkpoint/events
+- 遇到失败立即收敛原因并记录，不并行扩散新的失败面
 
-## 运行模式对比
+会话收尾必须：
+1. 更新 `task-manifest.json` 状态
+2. 更新 `session-progress.md` 与 `session-handoff.md`
+3. 推荐提交一个可回滚的最小 git commit（若仓库策略允许）
 
-| | Manual | Autonomous (默认) |
-|--|--------|------------------|
-| 遇到错误 | 停止 | 自动重试 |
-| 注意力刷新 | Protocol 2 | Protocol 2+3+4 |
-| 适用场景 | 复杂需求 | 清晰需求 |
+### Done Gate
 
-## Execution Flow
+仅当以下条件满足才标记 `flow:dev` 完成：
+- 全部任务为 `passed` 或 `skipped`
+- 无 `running`/`failed` 残留
+- `session-checklist.json` 中 `flow:dev.passes == true`
 
-### Stage 1: Entry Gate
+## Input Format
 
-1. TASKS.md 存在
-2. EPIC.md 存在
-3. Status: `epic_complete`
+```bash
+/flow:dev "REQ_ID" [--parallel N] [--resume] [--max-retries N]
+```
 
-### Stage 2: Task Execution (Ralph Loop)
+## Execution Steps
 
-```
-For each task in TASKS.md:
-  1. Protocol 2: 读取任务 DoD
-  2. 执行任务
-  3. 验证完成
-  4. 标记 [x]
-  5. Protocol 3: 读取下一任务 + ERROR_LOG
+1. 解析参数：
+   - 默认 `parallel = 3`
+   - 默认不 resume
+2. 分支执行：
+   - 普通执行：
+
+```bash
+npm run harness:dispatch -- --change-id "${REQ_ID}" --parallel ${PARALLEL} [--max-retries ${MAX_RETRIES}]
 ```
 
-### Stage 3: Error Handling
+   - 恢复执行：
 
+```bash
+npm run harness:resume -- --change-id "${REQ_ID}" --parallel ${PARALLEL} [--max-retries ${MAX_RETRIES}]
 ```
-On Error:
-  1. Protocol 4: 读取 ERROR_LOG.md
-  2. 记录错误到 ERROR_LOG.md
-  3. Autonomous: 自动重试
-  4. Manual: 停止等待
-```
 
-### Stage 4: Exit Gate
+3. 检查执行结果：
+   - `task-manifest.json` 状态更新
+   - `.harness/runtime/${REQ_ID}/<TASK_ID>/events.jsonl`
+   - `.harness/runtime/${REQ_ID}/<TASK_ID>/checkpoint.json`
+
+## Exit Criteria
 
-1. 所有任务完成
-2. 测试通过
-3. Status: `development_complete`
+- 所有任务状态为 `passed` 或 `skipped`
+- 没有 `running` 残留状态
 
-## Attention Refresh Protocols
+## Failure Handling
 
-| Protocol | Trigger | Reads |
-|----------|---------|-------|
-| 2 | Task start | TASKS.md T### + DoD |
-| 3 | Iteration start | TASKS.md + ERROR_LOG |
-| 4 | After error | ERROR_LOG.md |
+- 若存在 `failed` 任务：
+  1. 修复失败原因
+  2. 重新运行 `--resume`
 
 ## Next Step
 
-```
-/flow-quality "${REQ_ID}"
+```bash
+/flow:verify "${REQ_ID}"
 ```

package/.claude/skills/workflow/flow-dev/context.jsonl

@@ -1,8 +1,4 @@
-{"file": "devflow/requirements/{REQ}/TASKS.md", "reason": "Task list"}
-{"file": "devflow/requirements/{REQ}/EPIC.md", "reason": "Epic overview"}
-{"file": "devflow/requirements/{REQ}/BRAINSTORM.md", "reason": "Original intent"}
-{"file": "devflow/requirements/{REQ}/ERROR_LOG.md", "reason": "Previous errors", "optional": true}
-{"file": "devflow/requirements/{REQ}/TECH_DESIGN.md", "reason": "Technical design", "optional": true}
-{"file": "devflow/spec/frontend/index.md", "reason": "Frontend conventions", "optional": true}
-{"file": "devflow/spec/backend/index.md", "reason": "Backend conventions", "optional": true}
-{"file": ".claude/rules/project-constitution.md", "reason": "Quality rules"}
+{"file": "devflow/requirements/{REQ}/task-manifest.json", "reason": "Executable task graph"}
+{"file": "devflow/requirements/{REQ}/context-package.md", "reason": "Execution constraints", "optional": true}
+{"file": "devflow/requirements/{REQ}/harness-state.json", "reason": "Lifecycle status", "optional": true}
+{"file": ".claude/rules/project-constitution.md", "reason": "Quality rules and constraints"}

package/.claude/skills/workflow/flow-init/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: flow-init
-description: 'Initialize requirement structure with brainstorming and research. Usage: /flow-init "REQ-123|Title|URLs". Creates BRAINSTORM.md (north star) and research artifacts.'
+description: 'Initialize a requirement with harness state and context package. Use when starting a new REQ/BUG and preparing deterministic execution context for the next stages.'
 ---
 
 # Flow-Init Skill
@@ -9,174 +9,76 @@ description: 'Initialize requirement structure with brainstorming and research.
 
 ## Purpose
 
-初始化需求结构，包含头脑风暴和研究阶段。生成 BRAINSTORM.md 作为需求的「北极星」。
+初始化需求目录的 harness 运行状态，并生成可执行上下文包。
 
-## Input Format
-
-```
-/flow-init "REQ_ID|TITLE|PLAN_URLS?"
-/flow-init --interactive
-```
+## Long-Running Harness Protocol（Initializer/Worker/Done Gate）
 
-- **REQ_ID**: `^(REQ|BUG)-[0-9]+$`
-- **TITLE**: 需求简短标题
-- **PLAN_URLS**: 计划文档URL (可选，逗号分隔)
+为与 `/core:*` 保持一致，`/flow:init` 也采用会话分层协议。
 
-## Execution Flow
+### Initializer Session（为后续窗口建立可恢复上下文）
 
-### Stage 1: Entry Gate
+在执行 `harness:init + harness:pack` 后，确保以下工件存在：
+- `devflow/requirements/${REQ_ID}/session-checklist.json`
+  - 至少包含 `flow:init/flow:spec/flow:dev/flow:verify/flow:release` 五个阶段，默认 `passes=false`
+- `devflow/requirements/${REQ_ID}/session-progress.md`
+  - 记录本窗口完成内容、风险和阻塞
+- `devflow/requirements/${REQ_ID}/session-handoff.md`
+  - 明确下一窗口唯一优先动作（通常是 `/flow:spec "${REQ_ID}"`）
 
-1. **参数解析**
-   - 验证 REQ_ID 格式
-   - 提取 TITLE 和 PLAN_URLS
+### Worker Session（只做 init 的原子闭环）
 
-2. **前置检查**
-   ```bash
-   bash scripts/check-prerequisites.sh --json --paths-only
-   ```
-   - devflow/ 目录存在
+每次仅执行一个最小闭环：
+1. 读取 `session-progress.md` 与 `session-handoff.md`
+2. 运行 `harness:init`
+3. 运行 `harness:pack`
+4. 验证产物并更新 session 工件
 
-3. **唯一性检查**
-   - `devflow/requirements/${REQ_ID}/` 不存在
+### Done Gate（禁止口头完成）
 
-### Stage 1.5: Context Loading
+仅当以下条件同时满足才允许标记 `flow:init` 为完成：
+- `harness-state.json.status == "initialized"`
+- `context-package.md` 存在且包含 Next Commands
+- `session-checklist.json` 中 `flow:init.passes == true`
 
-检查并加载 (如存在):
-- `devflow/ROADMAP.md`
-- `devflow/ARCHITECTURE.md`
-
-### Stage 2: Directory Init
+## Input Format
 
 ```bash
-bash scripts/create-requirement.sh "${REQ_ID}" --title "${TITLE}"
+/flow:init "REQ_ID|TITLE|PLAN_URLS?"
 ```
 
-生成:
-- `devflow/requirements/${REQ_ID}/README.md`
-- `devflow/requirements/${REQ_ID}/EXECUTION_LOG.md`
-- `devflow/requirements/${REQ_ID}/orchestration_status.json`
-- `devflow/requirements/${REQ_ID}/research/`
-
-### Stage 2.3: Brainstorming (MANDATORY)
-
-**Iron Law**: `NO FLOW EXECUTION WITHOUT BRAINSTORM ALIGNMENT`
-
-触发 `flow-brainstorming` skill，执行:
-
-1. **Understanding** - 一次问一个问题
-2. **Exploring** - 提出 2-3 种方案
-3. **Presenting** - 分段呈现设计
-4. **Documentation** - 输出 BRAINSTORM.md
-
-**输出**: `devflow/requirements/${REQ_ID}/BRAINSTORM.md`
-
-**验证**:
-- [ ] BRAINSTORM.md 存在
-- [ ] 包含「原始需求」章节
-- [ ] 包含「成功标准」章节
-- [ ] 包含「方案探索」章节
-- [ ] 包含「最终决策」章节
-
-### Stage 2.5: Research (Subagent)
-
-调用 `flow-researcher` subagent:
-
-```json
-{
-  "reqId": "${REQ_ID}",
-  "reqDir": "devflow/requirements/${REQ_ID}",
-  "title": "${TITLE}",
-  "planUrls": ["..."],
-  "contextFiles": {
-    "brainstorm": "devflow/requirements/${REQ_ID}/BRAINSTORM.md"
-  }
-}
-```
+- `REQ_ID`: `REQ-123` 或 `BUG-123`
+- `TITLE`: 需求标题
+- `PLAN_URLS`: 可选，逗号分隔
 
-**输出**:
-- `research/internal/codebase-overview.md`
-- `research/mcp/YYYYMMDD/**`
-- `research/research-summary.md`
-- `research/tasks.json`
-- `research/research.md`
+## Execution Steps
 
-### Stage 3: README Generation
-
-生成 `devflow/requirements/${REQ_ID}/README.md`
-
-### Stage 4: Exit Gate (5-Level)
+1. 解析输入，提取 `REQ_ID`、`TITLE`、`PLAN_URLS`。
+2. 组装 goal 文本：
+   - `Deliver <REQ_ID>: <TITLE>`
+   - 若有 URL，追加 `Sources: <URLS>`。
+3. 运行初始化：
 
 ```bash
-# Level 2: Structure validation
-bash scripts/validate-research.sh "${REQ_DIR}" --strict
+npm run harness:init -- --change-id "${REQ_ID}" --goal "${GOAL}"
 ```
 
-**验证**:
-1. File Existence
-2. Research.md Structure
-3. Content Quality (无 TODO/PLACEHOLDER)
-4. Tasks.json Validation
-5. Git & Status & Constitution
+4. 运行上下文打包：
 
-## Output Artifacts
-
-```
-devflow/requirements/${REQ_ID}/
-├── README.md
-├── BRAINSTORM.md           # 北极星
-├── research/
-│   ├── internal/
-│   │   └── codebase-overview.md
-│   ├── mcp/YYYYMMDD/
-│   ├── research.md
-│   ├── research-summary.md
-│   └── tasks.json
-├── EXECUTION_LOG.md
-└── orchestration_status.json
+```bash
+npm run harness:pack -- --change-id "${REQ_ID}" --goal "${GOAL}"
 ```
 
-## Scripts Reference
-
-| Script | Purpose |
-|--------|---------|
-| `scripts/check-prerequisites.sh` | 前置条件检查 |
-| `scripts/create-requirement.sh` | 创建目录结构 |
-| `scripts/generate-research-tasks.sh` | 生成研究任务 |
-| `scripts/populate-research-tasks.sh` | 填充任务内容 |
-| `scripts/consolidate-research.sh` | 整合研究结果 |
-| `scripts/validate-research.sh` | 验证研究质量 |
-
-## Templates Reference
-
-| Template | Purpose |
-|----------|---------|
-| `assets/INIT_FLOW_TEMPLATE.md` | 详细执行流程 |
-| `assets/RESEARCH_TEMPLATE.md` | 研究文档格式 |
-| `assets/BRAINSTORM_TEMPLATE.md` | 头脑风暴模板 |
+5. 验证输出文件：
+   - `devflow/requirements/${REQ_ID}/harness-state.json`
+   - `devflow/requirements/${REQ_ID}/context-package.md`
 
-## Agent Reference
+## Exit Criteria
 
-| Agent | Purpose |
-|-------|---------|
-| `references/flow-researcher.md` | 研究 subagent 指令 |
-
-## Error Handling
-
-| Error | Solution |
-|-------|----------|
-| Invalid REQ_ID | 检查格式 `^(REQ\|BUG)-[0-9]+$` |
-| REQ_ID exists | 使用不同 ID 或 --force |
-| Git not clean | commit/stash changes |
-| Research validation failed | 手动补充 research.md |
+- `harness-state.json.status == "initialized"`
+- `context-package.md` 存在并包含 Next Commands 段落
 
 ## Next Step
 
+```bash
+/flow:spec "${REQ_ID}"
 ```
-/flow-prd "${REQ_ID}"
-```
-
----
-
-**Related Skills**:
-- `flow-brainstorming` - 头脑风暴
-- `flow-prd` - PRD 生成

package/.claude/skills/workflow/flow-init/assets/RESEARCH_TEMPLATE.md

@@ -145,7 +145,7 @@ Generated: {GENERATED_AT}
 2. **Direct Editing**:
    - Copy this template to `research/research.md`
    - Fill in decisions based on your research
-   - Run validation script before proceeding to /flow-prd
+   - Run validation script before proceeding to /flow:spec
 
 ---
 

package/.claude/skills/workflow/flow-init/context.jsonl

@@ -1,5 +1,5 @@
-{"file": ".claude/rules/project-constitution.md", "reason": "Quality rules and constraints"}
 {"file": "devflow/ROADMAP.md", "reason": "Project roadmap context", "optional": true}
 {"file": "devflow/ARCHITECTURE.md", "reason": "Architecture context", "optional": true}
-{"file": "devflow/spec/frontend/index.md", "reason": "Frontend conventions", "optional": true}
-{"file": "devflow/spec/backend/index.md", "reason": "Backend conventions", "optional": true}
+{"file": "devflow/requirements/{REQ}/harness-state.json", "reason": "Existing harness state", "optional": true}
+{"file": "devflow/requirements/{REQ}/context-package.md", "reason": "Existing context package", "optional": true}
+{"file": ".claude/rules/project-constitution.md", "reason": "Quality rules and constraints"}

package/.claude/skills/workflow/flow-init/scripts/check-prerequisites.sh

@@ -134,7 +134,7 @@ fi
 # Validate required directories and files
 if [[ ! -d "$REQ_DIR" ]]; then
     echo "ERROR: Requirement directory not found: $REQ_DIR" >&2
-    echo "Run /flow-new first to create the requirement structure." >&2
+    echo "Run /flow:init first to create the requirement structure." >&2
     exit 1
 fi
 

package/.claude/skills/workflow/flow-init/scripts/validate-research.sh

@@ -311,7 +311,7 @@ main() {
   echo "════════════════════════════════════════════════════════════════════════"
   if [[ $total_errors -eq 0 ]]; then
     echo "✅ ALL VALIDATIONS PASSED"
-    echo "research.md is ready for /flow-prd"
+    echo "research.md is ready for /flow:spec"
     echo "════════════════════════════════════════════════════════════════════════"
     return 0
   else

package/.claude/skills/workflow/flow-release/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: flow-release
-description: 'Complete requirement and update progress. Usage: /flow-release "REQ-123" or /flow-release'
+description: 'Release a verified requirement and run runtime cleanup. Use only after flow-verify has passed.'
 ---
 
 # Flow-Release Skill
@@ -9,74 +9,41 @@ description: 'Complete requirement and update progress. Usage: /flow-release "RE
 
 ## Purpose
 
-完成需求发布：更新进度文档、提交变更。Git 分支/PR/合并由用户自行管理。
+在验证通过后生成发布说明并标记需求为 released，同时执行 runtime 清理。
 
 ## Input Format
 
+```bash
+/flow:release "REQ_ID" [--janitor-hours N]
 ```
-/flow-release "REQ_ID"
-/flow-release             # Auto-detect
-```
-
-## Entry Gate
-
-1. **PRD.md, EPIC.md, TASKS.md** 存在
-2. **TEST_REPORT.md, SECURITY_REPORT.md** Gate 均为 PASS
-3. **Status**: `quality_complete`（兼容 `qa_complete`）或 `release_failed`
-
-## Execution Flow
-
-### Stage 1: Context Preparation
-
-收集元数据:
-- REQ_ID, TITLE
-- coverage, security 状态
-- TASKS.md 完成情况
-
-### Stage 2: Release Manager Agent
-
-调用 `release-manager` agent:
-- 生成 RELEASE_PLAN.md (发布摘要)
 
-### Stage 3: Progress Update
+- `janitor-hours` 默认 `72`
 
-1. **更新 BACKLOG.md**
-   - 找到对应 REQ 条目，更新状态为 `completed` 或 `released`
+## Execution Steps
 
-2. **更新 ROADMAP.md**
-   - 找到对应 REQ 条目，更新进度
+1. 检查 `report-card.json`：
+   - `overall` 必须为 `pass`
+2. 运行发布：
 
-3. **更新 orchestration_status.json**
-   - Status → `release_complete`
-
-### Stage 4: Commit Gate (MANDATORY)
-
-- 检查 `git status --porcelain`
-- 若存在未提交变更，按 `.claude/commands/util/git-commit.md` 规范提交
-- 提交格式: `chore(release): complete ${REQ_ID} - ${TITLE}`
-
-### Stage 5: Exit Gate
-
-1. RELEASE_PLAN.md 存在
-2. Status: `release_complete`
-3. 变更已提交
+```bash
+npm run harness:release -- --change-id "${REQ_ID}"
+```
 
-## Output
+3. 运行熵清理：
 
+```bash
+npm run harness:janitor -- --hours ${HOURS}
 ```
-devflow/requirements/${REQ_ID}/
-├── RELEASE_PLAN.md
-└── orchestration_status.json (release_complete)
 
-Updated:
-├── devflow/BACKLOG.md (REQ status updated)
-└── devflow/ROADMAP.md (REQ progress updated)
+4. 验证输出：
+   - `devflow/requirements/${REQ_ID}/RELEASE_NOTE.md`
+   - `devflow/requirements/${REQ_ID}/harness-state.json` 中 `status == "released"`
 
-Git:
-└── Changes committed (user handles branch/PR/merge externally)
-```
+## Exit Criteria
+
+- 发布文件存在且状态为 released
+- janitor 执行成功
 
 ## Next Step
 
-1. 用户自行处理分支合并 / PR 创建
-2. 可选: `/flow-verify` 复检
+- 进入 PR / merge 流程（仓库策略处理）

package/.claude/skills/workflow/flow-release/context.jsonl

@@ -1,7 +1,5 @@
-{"file": "devflow/requirements/{REQ}/PRD.md", "reason": "Product requirements"}
-{"file": "devflow/requirements/{REQ}/EPIC.md", "reason": "Epic overview"}
-{"file": "devflow/requirements/{REQ}/TASKS.md", "reason": "Task completion"}
-{"file": "devflow/requirements/{REQ}/TEST_REPORT.md", "reason": "Test results", "optional": true}
-{"file": "devflow/requirements/{REQ}/SECURITY_REPORT.md", "reason": "Security results", "optional": true}
-{"file": "devflow/requirements/{REQ}/quickstart.md", "reason": "Verification commands", "optional": true}
-{"file": ".claude/rules/project-constitution.md", "reason": "Quality rules"}
+{"file": "devflow/requirements/{REQ}/report-card.json", "reason": "Gate results before release"}
+{"file": "devflow/requirements/{REQ}/task-manifest.json", "reason": "Task completion summary"}
+{"file": "devflow/requirements/{REQ}/harness-state.json", "reason": "Lifecycle status", "optional": true}
+{"file": "devflow/requirements/{REQ}/RELEASE_NOTE.md", "reason": "Previous release note", "optional": true}
+{"file": ".claude/rules/project-constitution.md", "reason": "Quality rules and constraints"}