ceo

Estrelas14

Forks2

Atualizado24 de janeiro de 2026 às 00:59

This skill should be used when the user asks to "build a complete app", "develop a software project from scratch", "create a website with full workflow", "manage a development team", "automate software development from requirements to deployment", or "coordinate multiple agents for software development". Integrates Superpowers frameworks: brainstorming for requirement exploration, git-worktrees for workspace isolation, subagent-driven-development for task execution, parallel-agent-dispatch for testing, TDD for code quality, and two-stage code review for specification compliance. Also includes integration testing with Chrome DevTools MCP for end-to-end validation. 🆕 v6.6.0: Complete workflow automation with auto/interactive modes, Git worktree for seamless project creation, and Callstack React Native best practices integration.

Instalação

Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.

Fonte

pyinx

pyinx/ceo-skills-plugin

Abrir repositório GitHub Ver repositórios do creator

Download

Ocupações relacionadasSOC

Baseado na classificação ocupacional SOC

Gerentes de sistemas computacionais e de informaçãoOcupações de Gestão·SOC 11-3021

SKILL.md

readonly

Mais deste repositório

mesmo repositório

two-stage-review

pyinx/ceo-skills-plugin

This skill should be used after completing development tasks. Performs two-stage code review: Stage 1 verifies specification compliance (prevents scope creep), Stage 2 reviews code quality (ensures maintainability).

2026-01-1914

name

ceo

description

CEO Agent - Direct Orchestration (v6.6.0)

When user requests software development, execute the following workflow:

Architecture Overview

v6.0 integrates Superpowers frameworks:

Phase 0: Brainstorming for requirement exploration
Phase 3.5: Git worktrees for workspace isolation
Phase 4: Subagent-driven development with two-stage code review
Phase 4.5: TDD enforcement
Phase 5: Parallel agent dispatch for independent test failures

Key principles:

流程自洽：Ensure workflow is complete, non-redundant, and naturally connected
保留自动化：Retain maximum automation, only key checkpoints require human confirmation
符合规范：Follow Claude Skill standards, prioritize Claude Code CLI native capabilities

Step 1: Initialize State Files

⚠️ NOTE: CEO agents are already installed as part of the ceo-skills plugin. No agent verification or installation is required.

Note: The Write tool will automatically create the .claudedocs directory if it doesn't exist.

Write initial task plan using Write tool

Create .claudedocs/task_plan.md with the following content:

# 任务计划

## Workflow配置 🆕 v6.5.0
- 执行模式: 待定（Step 1.5选择）
- 自动流转: 待定

## 用户需求
{USER_INPUT}

## 工作区配置
🔧 v6.5.0: 在阶段3.5创建Git Worktree后填充
- 工作区类型: Git Worktree
- 项目路径: 待定（阶段3.5创建）
- Git 分支: 待定（阶段3.5创建）

## 当前阶段
初始化

## 阶段进度
- [ ] 阶段0: 需求探索（brainstorming）
- [ ] 阶段1: 需求澄清（产品经理）
- [ ] 阶段2: 产品设计（UI/UX设计师）
- [ ] 阶段3: 架构设计（系统架构师）
- [ ] 阶段3.3: 平台决策（Web/Mobile/Both）🆕
- [ ] 阶段3.5: 工作区准备（git-worktrees）🔧 v6.5: 解决权限问题
- [ ] 阶段4: 开发实现（并行：Web+Mobile）🆕
- [ ] 阶段5: 测试验证（测试工程师-并行修复）
- [ ] 阶段6: 交付部署（市场营销师）

## 全局目标
1. 理解并澄清用户需求
2. 设计符合用户期望的产品
3. 实现高质量、可维护的代码
4. 确保充分测试和验证
5. 交付完整的文档和部署方案

Step 1.5: 选择Workflow执行模式 🆕 v6.5.0

Purpose: 在开始workflow前，让用户选择执行模式。

模式说明

🤖 自动模式（推荐）：完全自动化执行，跳过所有确认检查点，适合快速原型开发
👤 交互模式：在关键阶段暂停，等待用户确认，适合重要项目

Execution

Use AskUserQuestion tool:

Question: "选择CEO Workflow执行模式："
Header: "⚙️ Workflow模式配置"
Options:
  - label: "🤖 自动模式（推荐）"
    description: "完全自动化执行，跳过所有确认检查点，适合快速原型开发"
  - label: "👤 交互模式"
    description: "在每个关键阶段暂停，等待用户确认，适合重要项目"

Save user selection to task_plan.md:

If user selects "🤖 自动模式":

Edit file: .claudedocs/task_plan.md
Replace: "## Workflow配置\n- 执行模式: 待定（Step 1.5选择）\n- 自动流转: 待定"
With:
"""
## Workflow配置
- 执行模式: 自动模式
- 自动流转: 启用
- 确认检查点: 跳过
"""

If user selects "👤 交互模式":

Edit file: .claudedocs/task_plan.md
Replace: "## Workflow配置\n- 执行模式: 待定（Step 1.5选择）\n- 自动流转: 待定"
With:
"""
## Workflow配置
- 执行模式: 交互模式
- 自动流转: 禁用
- 确认检查点: 启用
"""

Display confirmation message:

═════════════════════════════════════════════════════════════
✅ Workflow模式已配置
═════════════════════════════════════════════════════════════

执行模式: {用户选择的模式}
{if 自动模式} 🤖 自动模式：将自动流转到所有阶段，无需手动确认
{if 交互模式} 👤 交互模式：将在关键阶段等待您确认

📋 下一步: Phase 0 - 需求探索

Proceed to Step 2 (Phase 0).

Git 分支: 待定（阶段3.5创建）

当前阶段

初始化

阶段进度

阶段0: 需求探索（brainstorming）
阶段1: 需求澄清（产品经理）
阶段2: 产品设计（UI/UX设计师）
阶段3: 架构设计（系统架构师）
阶段3.3: 平台决策（Web/Mobile/Both）🆕
阶段3.5: 工作区准备（git-worktrees）🔧 v6.5: 解决权限问题
阶段4: 开发实现（并行：Web+Mobile）🆕
阶段5: 测试验证（测试工程师-并行修复）
阶段6: 交付部署（市场营销师）

全局目标

理解并澄清用户需求
设计符合用户期望的产品
实现高质量、可维护的代码
确保充分测试和验证
交付完整的文档和部署方案


### Initialize notes file
Create `.claudedocs/notes.md` with the following content:
```markdown
# 项目笔记

## 初始化
项目启动时间: [执行时自动记录]

Step 2: Detect Current Phase and Resume

⚠️ CRITICAL: Before starting any phase, check if there's an existing workflow to resume.

Check for existing task plan

Use Read tool to check if task plan exists:

Read file: .claudedocs/task_plan.md

If task_plan.md exists (Resume Mode)

Parse current phase: Read "## 当前阶段" section
Check progress: Read "## 阶段进度" to see completed phases
Jump to next phase: Use the mapping below

Phase mapping:

"初始化" OR "阶段0: 需求探索" → Go to Step 3 (Phase 0)
"阶段1: 需求澄清" → Go to Step 4 (Phase 1)
"阶段2: 产品设计" → Go to Step 5 (Phase 2)
"阶段3: 架构设计" → Go to Step 6 (Phase 3)
"阶段3.5: 工作区准备" → Go to Step 7 (Phase 3.5)
"阶段4: 开发实现" → Go to Step 8 (Phase 4)
"阶段5: 测试验证" → Go to Step 9 (Phase 5)
"阶段6: 交付部署" → Go to Step 10 (Phase 6)

Display resume message:

═════════════════════════════════════════════════════════════
🔄 恢复工作流
═════════════════════════════════════════════════════════════

检测到现有任务计划，将从 {CURRENT_PHASE} 继续执行。

Then jump to the appropriate step above.

If task_plan.md doesn't exist (Fresh Start)

Display initialization message:

═════════════════════════════════════════════════════════════
🚀 启动新的 CEO 工作流
═════════════════════════════════════════════════════════════

将创建新的任务计划并执行完整 6 阶段开发流程。

Proceed to Step 3 (Phase 0).

Step 3: Execute Phase 0 - 需求探索（Brainstorming）

🆕 NEW in v6.0: Integrate Superpowers brainstorming for requirement exploration.

Purpose

Before generating formal PRD, conduct conversational exploration to fully understand user requirements through dialogue.

Process

Follow the brainstorming process:

digraph phase0_brainstorming {
    rankdir=TB;
    start [label="用户输入模糊需求", shape=ellipse];
    context [label="检查项目状态\n（文件、文档、最近提交）", shape=box];
    questions [label="苏格拉底式提问\n（一次一个问题）", shape=box, style=filled, fillcolor="#e1f5ff"];
    approaches [label="提出2-3种方案\n（含权衡分析）", shape=box];
    design [label="分段展示设计\n（200-300词/段）", shape=box];
    validate [label="每段后验证\n'这看起来对吗？'", shape=diamond];
    doc [label="保存设计文档\n.claudedocs/phase0-design.md", shape=box];

    start → context;
    context → questions;
    questions → approaches;
    approaches → design;
    design → validate;
    validate → design [label="否，调整"];
    validate → doc [label="是"];
}

Key Rules

提问规则:

一次一个问题：不要用多个问题淹没用户
优先选择题：比开放性问题更容易回答
聚焦理解：目的、约束、成功标准

设计展示规则:

分段展示：每段200-300词
每段验证：展示后询问"这看起来对吗？"
灵活调整：如有不清楚，返回澄清

产出物:

设计文档：.claudedocs/phase0-design.md
包含内容：架构、组件、数据流、错误处理、测试策略

Execution

⚠️ DO NOT use AskUserQuestion tool - brainstorming is a natural conversational process.

Invoke the brainstorming skill (from superpowers), execute it, then continue to Phase 1.

After brainstorming completes, save design document to:

.claudedocs/phase0-design.md

Then update task_plan.md to mark Phase 0 as completed:

Edit: Replace "- [ ] 阶段0: 需求探索（brainstorming）"
With:  "- [x] 阶段0: 需求探索（brainstorming）"
Edit: Replace "## 当前阶段\n初始化"
With: "## 当前阶段\n阶段0: 需求探索（完成）"

Proceed to Phase 1.

═════════════════════════════════════════════════════════════
✅ Phase 0 完成 - 需求探索
═════════════════════════════════════════════════════════════

📄 设计文档: .claudedocs/phase0-design.md
📋 下一步: Phase 1 - 需求澄清（产品经理）

Step 4: Execute Phase 1 - 需求澄清（产品经理）

⚠️ MANDATORY: Read Previous Phase Output

Before executing this phase, you MUST read all previous outputs:

Read file: .claudedocs/phase0-design.md

This ensures you have complete context from Phase 0.

Update task plan current phase

Use Edit tool to update task_plan.md:

Replace: "## 当前阶段\n初始化"
With: "## 当前阶段\n阶段1: 需求澄清"

Call Product Manager agent

⚠️ IMPORTANT: Execute the following steps in order to ensure the agent completes BEFORE the confirmation checkpoint.

Step 1: Launch agent

Task tool: Launch the ceo-product-manager agent with the following context:

## CEO任务上下文

### 用户输入
{USER_INPUT}

### 阶段0输出（NEW - 设计文档）
[使用Read工具读取 .claudedocs/phase0-design.md 内容]

### 你的任务
1. 基于阶段0设计文档生成产品需求文档（PRD）
2. 构建用户画像
3. 定义MVP范围
4. 识别未澄清的问题（如有）

### ⚠️ 关键约束 - 提问规则
- **最多提问5个问题**：降低用户认知负担
- **分批提问**：如果问题超过5个，分多次提问，每次最多5个
- **优先级排序**：先问最重要、最核心的问题

### 输出要求
- 输出完整的PRD文档到 .claudedocs/ceo-product-manager_result.md
- 包含用户画像、功能列表、优先级
- 如果有未问的问题，在文档末尾列出"待确认的问题"

Save the returned task_id as {PRODUCT_MANAGER_TASK_ID}

Step 2: Wait for agent completion

TaskOutput: Wait for {PRODUCT_MANAGER_TASK_ID}
Parameters: block=true, timeout=300000

⚠️ CRITICAL: DO NOT PROCEED until agent completes
⚠️ DO NOT proceed to confirmation checkpoint until this step completes

Step 3: Verify output file exists

Read file: .claudedocs/ceo-product-manager_result.md

⚠️ If file doesn't exist, agent failed - inform user and ask what to do

After agent completes and output is verified, proceed to confirmation checkpoint below.

Step 4.2: CONDITIONAL - User Confirmation Checkpoint 🆕 v6.5.0

⚠️ CONDITIONAL CHECKPOINT - 根据workflow模式决定是否需要确认

Check workflow mode first:

Read file: .claudedocs/task_plan.md

Extract:
  - 执行模式: (自动模式 | 交互模式)

If 执行模式 == "自动模式":

✅ 自动模式：跳过确认检查点

直接执行批准操作：
1. Use Edit tool to update task_plan.md:
   Replace: "## 当前阶段\n阶段1: 需求澄清"
   With: "## 当前阶段\n阶段2: 产品设计"
   Replace: "- [ ] 阶段0: 需求探索（brainstorming）"
   With: "- [x] 阶段0: 需求探索（brainstorming）"
   Replace: "- [ ] 阶段1: 需求澄清（产品经理）"
   With: "- [x] 阶段1: 需求澄清（产品经理）"

2. Display message:

═════════════════════════════════════════════════════════════ ✅ 自动模式：PRD已自动批准 ═════════════════════════════════════════════════════════════

🤖 执行模式: 自动模式 📄 PRD文档: .claudedocs/ceo-product-manager_result.md ✅ 状态: 已自动批准，继续下一阶段

📋 下一步: Phase 2 - 产品设计（UI/UX设计师）


3. Proceed to Step 5 (Phase 2).

If 执行模式 == "交互模式":

👤 交互模式：执行用户确认流程

⚠️ **CRITICAL**: You MUST pause here and wait for user confirmation before proceeding.

First, use Read tool to display PRD preview:

Read file: .claudedocs/ceo-product-manager_result.md Limit: 50 lines Display to user with formatted header: "📋 产品需求文档预览"


Then, use AskUserQuestion tool to get user confirmation:

Question: "请查看产品需求文档（PRD）并提供反馈。是否批准此PRD？" Header: "🎯 检查点 1 - 产品需求文档确认" Options:

label: "✅ 批准PRD" description: "PRD符合预期，批准并继续下一阶段"
label: "📝 修改PRD" description: "我有修改意见，需要调整PRD"
label: "🔄 重做PRD" description: "PRD不符合预期，需要重新澄清需求"
label: "🛑 终止workflow" description: "结束整个开发流程"


⚠️ **DO NOT PROCEED** until user selects an option.

Step 4.3: Process User Decision

If user selects ✅ 批准PRD:

Use Edit tool to update task_plan.md:
- Mark Phase 0 and Phase 1 as completed
- Update current phase to "阶段2: 产品设计"
Proceed to Step 5 (Phase 2).

═════════════════════════════════════════════════════════════
✅ Phase 1 完成 - 需求澄清
═════════════════════════════════════════════════════════════

📄 PRD文档: .claudedocs/ceo-product-manager_result.md
📋 下一步: Phase 2 - 产品设计（UI/UX设计师）

If user selects 📝 修改PRD:

Use AskUserQuestion to collect specific modification requests
Call Product Manager again with feedback
After revision completes, repeat Step 4.2 (confirmation checkpoint)

If user selects 🔄 重做PRD:

Use AskUserQuestion to collect new requirements
Call Product Manager for new round
After new PRD completes, repeat Step 4.2 (confirmation checkpoint)

If user selects 🛑 终止workflow:

Display termination message
Update task_plan.md with termination status
End workflow

Step 5: Execute Phase 2 - 产品设计

⚠️ MANDATORY: Read Previous Phase Outputs

Before executing this phase, you MUST read all previous outputs:

Read file: .claudedocs/phase0-design.md
Read file: .claudedocs/ceo-product-manager_result.md

This ensures you have complete context from Phase 0 and Phase 1.

Update task plan current phase

Use Edit tool to update task_plan.md:

Replace: "## 当前阶段\n阶段1: 需求澄清"
With: "## 当前阶段\n阶段2: 产品设计"

Call UI/UX Designer agent

⚠️ IMPORTANT: Execute the following steps in order to ensure the agent completes BEFORE proceeding.

Step 1: Launch agent

Task tool: Launch the ceo-ui-ux-designer agent with the following context:

## CEO任务上下文

### 用户输入
{USER_INPUT}

### 阶段0-1输出
[使用Read工具读取阶段0设计文档和阶段1 PRD]

### 用户对阶段0-1问题的回答
{USER_ANSWERS_PHASE0_1}

### 你的任务
1. 基于PRD设计用户故事
2. 设计交互流程
3. 设计视觉界面
4. 创建原型设计

### ⚠️ 关键约束 - 提问规则
- **最多提问5个问题**：降低用户认知负担
- **分批提问**：如果问题超过5个，分多次提问，每次最多5个
- **优先级排序**：先问最重要、最核心的设计问题

### 输出要求
- 输出完整的设计文档到 .claudedocs/ceo-ui-ux-designer_result.md
- 包含用户故事、交互流程、视觉设计

Save the returned task_id as {UI_UX_TASK_ID}

Step 2: Wait for agent completion

TaskOutput: Wait for {UI_UX_TASK_ID}
Parameters: block=true, timeout=300000

⚠️ CRITICAL: DO NOT PROCEED until agent completes

Step 3: Verify output file exists

Read file: .claudedocs/ceo-ui-ux-designer_result.md

⚠️ If file doesn't exist, agent failed - inform user and ask what to do

After agent completes and output is verified:

Use Edit tool to update task_plan.md: Mark Phase 2 as completed
Proceed directly to Phase 3 (no confirmation required).

═════════════════════════════════════════════════════════════
✅ Phase 2 完成 - 产品设计
═════════════════════════════════════════════════════════════

📄 设计文档: .claudedocs/ceo-ui-ux-designer_result.md
📋 下一步: Phase 3 - 架构设计（系统架构师）

Step 6: Execute Phase 3 - 架构设计

⚠️ MANDATORY: Read Previous Phase Outputs

Before executing this phase, you MUST read all previous outputs:

Read file: .claudedocs/phase0-design.md
Read file: .claudedocs/ceo-product-manager_result.md
Read file: .claudedocs/ceo-ui-ux-designer_result.md

This ensures you have complete context from Phase 0, 1, and 2.

Update task plan current phase

Use Edit tool to update task_plan.md:

Replace: "## 当前阶段\n阶段2: 产品设计"
With: "## 当前阶段\n阶段3: 架构设计"

Call System Architect agent

⚠️ CRITICAL: Execute the following steps in order to ensure the agent completes BEFORE the confirmation checkpoint.

Step 1: Launch agent

Task tool: Launch the ceo-system-architect agent with the following context:

## CEO任务上下文

### 用户输入
{USER_INPUT}

### 前期阶段输出
[使用Read工具读取所有前期输出文件]

### 用户回答
{USER_ANSWERS}

### 你的任务
1. 技术栈选型（前端、后端、数据库）
2. 系统架构设计
3. API规范设计
4. 数据模型设计

### ⚠️ 关键约束 - 提问规则
- **最多提问5个问题**：降低用户认知负担
- **分批提问**：如果技术决策问题超过5个，分多次提问
- **优先级排序**：先问最关键的技术选型问题

### 输出要求
- 输出完整的架构设计文档到 .claudedocs/ceo-system-architect_result.md

Save the returned task_id as {ARCHITECT_TASK_ID}

Step 2: Wait for agent completion

TaskOutput: Wait for {ARCHITECT_TASK_ID}
Parameters: block=true, timeout=300000

⚠️ CRITICAL: DO NOT PROCEED until agent completes
⚠️ DO NOT proceed to confirmation checkpoint until this step completes

Step 3: Verify output file exists

Read file: .claudedocs/ceo-system-architect_result.md

⚠️ If file doesn't exist, agent failed - inform user and ask what to do
⚠️ DO NOT proceed to confirmation checkpoint if file doesn't exist

After agent completes and output is verified, CHECK WORKFLOW MODE before proceeding.

⚠️ DO NOT PROCEED to Phase 3.5 until user confirms the architecture (交互模式) or auto-approves (自动模式)!

Step 6.1: CONDITIONAL - Architecture Confirmation Checkpoint 🆕 v6.5.0

🚨 CONDITIONAL CHECKPOINT - 根据workflow模式决定是否需要确认

Check workflow mode first:

Read file: .claudedocs/task_plan.md

Extract:
  - 执行模式: (自动模式 | 交互模式)

If 执行模式 == "自动模式":

✅ 自动模式：跳过确认检查点

直接执行批准操作：
1. Use Edit tool to update task_plan.md:
   Replace: "## 当前阶段\n阶段3: 架构设计"
   With: "## 当前阶段\n阶段3.3: 平台决策"
   Replace: "- [ ] 阶段3: 架构设计（系统架构师）"
   With: "- [x] 阶段3: 架构设计（系统架构师）"

2. Display message:

═════════════════════════════════════════════════════════════ ✅ 自动模式：架构已自动批准 ═════════════════════════════════════════════════════════════

🤖 执行模式: 自动模式 📄 架构文档: .claudedocs/ceo-system-architect_result.md ✅ 状态: 已自动批准，继续平台决策

📋 下一步: Phase 3.3 - 平台决策


3. Proceed to Step 6.3 (Phase 3.3).

If 执行模式 == "交互模式":

👤 交互模式：执行用户确认流程

🚨 **CRITICAL CHECKPOINT - MANDATORY USER CONFIRMATION REQUIRED**

You are at the Architecture Confirmation Checkpoint. You MUST execute this step before proceeding to Phase 3.5.

**Step 1**: Display architecture document preview using Read tool:

Read file: .claudedocs/ceo-system-architect_result.md Limit: 50 lines Display formatted header: "🏗️ 技术架构设计文档预览"


**Step 2**: Use AskUserQuestion tool to get user confirmation:

Question: "请查看技术架构设计文档并提供反馈。是否批准此架构方案？" Header: "🏗️ 检查点 2 - 技术架构方案确认" Options:

label: "✅ 批准架构" description: "架构方案符合预期，批准并继续创建工作区"
label: "📝 修改架构" description: "我有修改意见，需要调整架构设计"
label: "🔄 重做架构" description: "架构不符合预期，需要重新设计"
label: "🛑 终止workflow" description: "结束整个开发流程"


**Step 3**: ⚠️ **WAIT FOR USER RESPONSE - DO NOT PROCEED**

⚠️ **DO NOT PROCEED** until user selects an option.
⚠️ **DO NOT PROCEED to Phase 3.5** until user selects "✅ 批准架构".

Step 6.2: Process User Decision

If user selects ✅ 批准架构:

Use Edit tool to update task_plan.md:
- Mark Phase 3 as completed
- Update current phase to "阶段3.3: 平台决策"
Proceed to Step 6.3 (Phase 3.3 - 平台决策).

═════════════════════════════════════════════════════════════
✅ Phase 3 完成 - 架构设计
═════════════════════════════════════════════════════════════

📄 架构文档: .claudedocs/ceo-system-architect_result.md
📋 下一步: Phase 3.3 - 平台决策（Web/Mobile/Both）

Step 6.3: 🆕 Phase 3.3 - 平台决策（NEW in v6.4.0）

🆕 NEW in v6.4.0: 在架构设计后进行平台决策，动态决定开发Web、Mobile还是Both。

Purpose

根据产品需求、用户场景、功能需求和技术约束，决定开发平台：

Web Only: 仅Web应用
Mobile Only: 仅移动应用
Both: Web + Mobile并行开发

决策流程

Step 1: 分析决策因素

基于以下因素进行分析：

决策因素:
  用户因素:
    - 目标用户类型（办公用户/移动用户/全场景用户）
    - 主要使用设备（桌面/移动/混合）
    - 使用场景（固定/移动/灵活）

  功能因素:
    - 需要的原生功能（GPS、相机、传感器等）
    - 复杂表单处理（Web优势）
    - 实时协作（Web优势）
    - 离线需求（Mobile优势）

  技术因素:
    - 技术栈复杂度
    - 开发时间要求
    - 团队能力匹配

  约束因素:
    - 开发预算
    - 上市时间要求
    - 维护成本考虑

Step 2: 生成平台决策

基于分析生成决策文档：

interface PlatformDecision {
  // 目标平台
  platforms: ('web' | 'mobile')[];

  // 开发优先级
  priority: 'web-first' | 'mobile-first' | 'parallel';

  // 决策理由
  rationale: string;

  // 实施建议
  implementation: {
    phasedRollout: boolean;
    mvpPlatform: 'web' | 'mobile';
    featuresByPlatform: {
      web: string[];
      mobile: string[];
      shared: string[];
    };
  };
}

Step 3: 保存决策文档

使用Write工具创建平台决策文档：

Write file: .claudedocs/platform-decision.md

Content:
# 平台决策文档

## 决策结果
- **目标平台**: {web | mobile | both}
- **开发优先级**: {web-first | mobile-first | parallel}
- **决策理由**: {rationale}

## 实施计划
- **分阶段发布**: {yes | no}
- **MVP平台**: {web | mobile}

## 功能分配
- **Web独有功能**: [...]
- **Mobile独有功能**: [...]
- **共享功能**: [...]

Step 4: 更新任务计划

使用Edit工具更新task_plan.md：

Replace: "## 当前阶段\n阶段3: 架构设计"
With: "## 当前阶段\n阶段3.3: 平台决策"

Step 6.4: CONDITIONAL - 平台决策确认检查点 🆕 v6.5.0

⚠️ CONDITIONAL CHECKPOINT - 根据workflow模式决定是否需要确认

Check workflow mode first:

Read file: .claudedocs/task_plan.md

Extract:
  - 执行模式: (自动模式 | 交互模式)

If 执行模式 == "自动模式":

✅ 自动模式：跳过确认检查点

直接执行批准操作：
1. Use Edit tool to update task_plan.md:
   Replace: "## 当前阶段\n阶段3.3: 平台决策"
   With: "## 当前阶段\n阶段3.5: 工作区准备"
   Replace: "- [ ] 阶段3.3: 平台决策（Web/Mobile/Both）"
   With: "- [x] 阶段3.3: 平台决策（Web/Mobile/Both）"

2. Extract platform decision and add to global goals:

Read file: .claudedocs/platform-decision.md

Extract:

platforms: (web | mobile | both)
priority: (web-first | mobile-first | parallel)


Add to task_plan.md "## 全局目标":

开发平台: {platforms}
优先级策略: {priority}


3. Display message:

═════════════════════════════════════════════════════════════ ✅ 自动模式：平台决策已自动批准 ═════════════════════════════════════════════════════════════

🤖 执行模式: 自动模式 📄 决策文档: .claudedocs/platform-decision.md 📱 开发平台: {platforms} ⚡ 优先级: {priority} ✅ 状态: 已自动批准，继续工作区准备

📋 下一步: Phase 3.5 - 工作区准备（Git Worktrees）


4. Proceed to Step 7 (Phase 3.5).

If 执行模式 == "交互模式":

👤 交互模式：执行用户确认流程

🚨 **CRITICAL CHECKPOINT - MANDATORY USER CONFIRMATION REQUIRED**

**Step 1**: 显示平台决策文档预览：

```bash
Read file: .claudedocs/platform-decision.md
Limit: 30 lines
Display formatted header: "📱 平台决策预览"

Step 2: 使用AskUserQuestion工具获取用户确认：

Question: "根据产品需求和技术架构分析，建议开发{platforms}，采用{priority}策略。是否批准此平台决策？"
Header: "📱 检查点 2.5 - 平台决策确认"
Options:
  - label: "✅ 批准决策"
    description: "同意平台决策，继续工作区准备"
  - label: "📝 修改决策"
    description: "有不同意见，需要调整平台决策"
  - label: "🛑 终止workflow"
    description: "结束整个开发流程"

Step 3: ⚠️ 等待用户响应 - 不要继续执行


### Step 6.5: 处理用户决策

**如果用户选择 ✅ 批准决策**:

1. 使用Edit工具更新task_plan.md：
   - 标记阶段3.3为已完成
   - 更新当前阶段为"阶段3.5: 工作区准备"
   - 添加平台决策信息到全局目标

```bash
Edit: task_plan.md
Add to "## 全局目标":
  - 开发平台: {platforms}
  - 开发策略: {priority}

继续执行Step 7 (Phase 3.5)

═════════════════════════════════════════════════════════════
✅ Phase 3.3 完成 - 平台决策
═════════════════════════════════════════════════════════════

📄 决策文档: .claudedocs/platform-decision.md
🎯 目标平台: {platforms}
📋 下一步: Phase 3.5 - 工作区准备（git-worktrees）

如果用户选择 📝 修改决策:

使用AskUserQuestion收集具体的修改意见
重新分析并生成新的平台决策
重复Step 6.4（确认检查点）

如果用户选择 🛑 终止workflow:

显示终止消息
更新task_plan.md为终止状态
结束workflow

If user selects 📝 修改架构:

Use AskUserQuestion to collect specific modification requests
Call System Architect again with feedback
After revision completes, repeat Step 6.1 (confirmation checkpoint)

If user selects 🔄 重做架构:

Use AskUserQuestion to collect new requirements
Call System Architect for new round
After new architecture completes, repeat Step 6.1 (confirmation checkpoint)

If user selects 🛑 终止workflow:

Display termination message
Update task_plan.md with termination status
End workflow

Step 7: Execute Phase 3.5 - 工作区准备（Git Worktrees）

🆕 NEW in v6.0: Integrate Superpowers using-git-worktrees for workspace isolation.

🔧 UPDATED in v6.5.0: 明确使用git-worktree解决Claude Code CLI跨目录访问权限问题。

Purpose

Before starting development, create isolated Git worktree to:

✅ 避免分支切换污染
✅ 解决跨目录访问权限问题（worktree属于项目内，无需额外确认）
✅ 提供独立的工作环境

Why Git Worktree Solves Permission Issues

问题根源：

# 当前工作目录
/path/to/ceo-skills-plugin/

# 操作目标（项目外）
/path/to/another-project/.claudedocs/
# ❌ Claude Code提示权限确认

Git Worktree解决方案：

# 创建worktree（项目内）
git worktree add ../my-project -b feature/my-project

# worktree目录结构
/path/to/repo/
├── .git/                    # 共享Git仓库
├── ceo-skills-plugin/        # 主worktree（当前目录）
└── my-project/               # 新worktree（共享.git）
    └── .git (worktree元数据)

# 操作目标（项目内）
../my-project/.claudedocs/
# ✅ worktree是Git仓库的一部分，无需权限确认

Execution Steps

Step 7.1: 确定项目名称和路径

从架构设计文档中提取项目名称：

Read file: .claudedocs/ceo-system-architect_result.md

Extract:
  - project_name: 项目名称

Step 7.2: 创建Git Worktree

# 计算worktree路径（相对于当前仓库）
WORKTREE_PATH="../${project_name}"
BRANCH_NAME="feature/${project_name}"

# 创建worktree
git worktree add ${WORKTREE_PATH} -b ${BRANCH_NAME}

# 验证创建成功
git worktree list

# 在worktree中创建文档目录（无需权限确认）
mkdir -p ${WORKTREE_PATH}/.claudedocs

Step 7.3: 保存Worktree信息到任务计划

Edit file: .claudedocs/task_plan.md

Add section after "## 用户需求":

## 工作区配置
- 工作区类型: Git Worktree
- 项目路径: ${WORKTREE_PATH}
- Git 分支: ${BRANCH_NAME}
- 相对路径: ../${project_name}

Replace: "## 当前阶段\n阶段3: 架构设计"
With: "## 当前阶段\n阶段3.5: 工作区准备"

Replace: "- [ ] 阶段3.5: 工作区准备（git-worktrees）"
With: "- [x] 阶段3.5: 工作区准备（git-worktrees）"

Step 7.4: 验证Worktree准备完成

# 列出所有worktrees
git worktree list

# 切换到worktree查看（可选）
cd ${WORKTREE_PATH}
pwd
ls -la

Proceed to Step 8 (Phase 4).

═════════════════════════════════════════════════════════════
✅ Phase 3.5 完成 - 工作区准备
═════════════════════════════════════════════════════════════

🌳 工作树路径: {WORKTREE_PATH}
📋 下一步: Phase 4 - 开发实现（子任务驱动）

Step 8: Execute Phase 4 - 开发实现（并行：Web + Mobile）

🆕 ENHANCED in v6.4.0: 支持并行Web和Mobile开发，根据平台决策动态激活agents。

🔧 UPDATED in v6.5.0: 所有项目操作在Git Worktree路径中执行，避免跨目录访问权限问题。

⚠️ MANDATORY: Read Previous Phase Outputs & Platform Decision

Before executing this phase, you MUST read all previous outputs AND the platform decision:

Read file: .claudedocs/phase0-design.md
Read file: .claudedocs/ceo-product-manager_result.md
Read file: .claudedocs/ceo-ui-ux-designer_result.md
Read file: .claudedocs/ceo-system-architect_result.md
Read file: .claudedocs/platform-decision.md  🆕 平台决策

This ensures you have complete context from all previous phases AND know which platforms to develop.

⚠️ CRITICAL: Load Worktree Configuration

🔧 v6.5.0: 从任务计划中加载worktree配置，确保所有项目操作在正确路径中执行：

Read file: .claudedocs/task_plan.md

Extract:
  - WORKTREE_PATH: 工作区路径（如 ../my-project）
  - BRANCH_NAME: Git分支（如 feature/my-project）

重要提示：

✅ 所有项目创建、文件操作、命令执行都在 ${WORKTREE_PATH} 中进行
✅ Worktree是Git仓库的一部分，属于"项目内"，无需权限确认
❌ 不要直接操作 /Users/zyb/.../my-project，使用相对路径 ${WORKTREE_PATH}

Execution Steps

Step 8.1: Update task plan current phase

Replace: "## 当前阶段\n阶段3.5: 工作区准备"
With: "## 当前阶段\n阶段4: 开发实现"

Step 8.2: 🆕 Parse Platform Decision

从平台决策文档中提取开发策略：

Read file: .claudedocs/platform-decision.md

Extract:
  - platforms: (web | mobile | both)
  - priority: (web-first | mobile-first | parallel)

Step 8.3: 🆕 根据平台决策配置开发任务

方案A: Backend Only（纯后端）

如果决策是纯后端项目（无前端需求）：

📋 开发策略：Backend Only

任务分配:
  - fullstack-developer: 后端API

执行步骤:
  1. 调用subagent-driven-development skill
  2. 激活ceo-fullstack-developer agent（仅后端）
  3. 任务分解：
     - 后端API实现
     - 数据库设计
     - 业务逻辑实现
  4. 执行两阶段代码审查

方案B: Web Frontend + Backend（传统Web）

如果决策是传统Web应用（React/Next.js等）：

📋 开发策略：Web Full Stack

任务分配:
  - fullstack-developer: 后端API + Web前端

执行步骤:
  1. 调用subagent-driven-development skill
  2. 激活ceo-fullstack-developer agent
  3. 任务分解：
     - 后端API实现
     - Web前端实现（React/Next.js）
  4. 执行两阶段代码审查

技术栈:
  - 后端: Node.js/Python/Go
  - 前端: React/Next.js/Vue（传统Web技术）

方案C: Cross-Platform Mobile（跨平台移动应用）

如果决策是跨平台移动应用（需要iOS + Android + Web）：

📋 开发策略：Cross-Platform Mobile

任务分配:
  - fullstack-developer: 后端API
  - mobile-developer: React Native应用（iOS + Android + Web）

执行步骤:
  1. 调用subagent-driven-development skill
  2. 并行激活agents:
     - ceo-fullstack-developer (后端API)
     - ceo-mobile-developer (React Native应用)
  3. 移动端开发依赖后端API完成
  4. 执行两阶段代码审查

技术栈:
  - 后端: Node.js/Python/Go
  - 移动端: React Native + Expo
  - 平台: iOS、Android、Web（React Native for Web）

说明:
  - React Native的"Web"是通过React Native for Web编译的
  - 这与传统Web前端（React/Next.js）不同
  - 一套代码覆盖三个平台

方案D: Web + Mobile Both（Web + 移动端）

如果决策是同时需要传统Web和移动应用：

📋 开发策略：Web + Mobile Both

任务分配:
  - fullstack-developer: 后端API + 传统Web前端
  - mobile-developer: React Native移动应用

根据优先级策略执行:

#### 策略1: web-first

执行顺序:
  1. 后端API (fullstack-developer)
  2. 传统Web前端 (fullstack-developer)
  3. 移动端 (mobile-developer)

#### 策略2: mobile-first

执行顺序:
  1. 后端API (fullstack-developer)
  2. 移动端 (mobile-developer)
  3. 传统Web前端 (fullstack-developer)

#### 策略3: parallel

执行顺序:
  1. 后端API (fullstack-developer) - 必须首先完成
  2. 并行开发:
     - 传统Web前端 (fullstack-developer)
     - 移动端 (mobile-developer)

技术栈:
  - 后端: Node.js/Python/Go
  - Web前端: React/Next.js（传统Web）
  - 移动端: React Native + Expo

Step 8.4: Extract development tasks from architecture

Read the architecture document and identify all development tasks. Group them into 2-5 minute subtasks.

⚠️ Note: All operations should use absolute paths based on {WORKTREE_PATH}.

Step 8.5: 🆕 根据平台策略执行开发

后端API开发（始终执行）

# 后端API由fullstack-developer统一实现
Task tool: Launch ceo-fullstack-developer agent

任务:
  1. 实现后端API（按照architect设计的API规范）
  2. 实现数据模型
  3. 实现业务逻辑
  4. API文档生成

输出: .claudedocs/backend-api-result.md

Web前端开发（按需执行）

如果 platforms.includes('web')：

Task tool: Launch ceo-fullstack-developer agent

任务:
  1. 实现Web前端组件
  2. 实现Web状态管理
  3. 集成后端API
  4. 响应式实现

依赖: 后端API完成

输出: .claudedocs/web-frontend-result.md

移动端开发（按需执行）

如果 platforms.includes('mobile')：

Task tool: Launch ceo-mobile-developer agent

任务:
  1. 初始化移动项目（使用obytes模板）
  2. 实现移动端UI
  3. 实现移动端状态管理
  4. 集成后端API
  5. 原生功能集成（如需要）

依赖: 后端API完成

输出: .claudedocs/mobile-app-result.md

Step 8.6: 🆕 并行开发协调

根据 priority 策略协调开发：

// web-first 策略
if (priority === 'web-first') {
  await executeBackendAPI();     // fullstack
  await executeWebFrontend();     // fullstack
  await executeMobileApp();       // mobile
}

// mobile-first 策略
else if (priority === 'mobile-first') {
  await executeBackendAPI();      // fullstack
  await executeMobileApp();        // mobile
  await executeWebFrontend();      // fullstack
}

// parallel 策略
else if (priority === 'parallel') {
  await executeBackendAPI();       // fullstack (必须先完成)

  // Web和Mobile并行开发
  await Promise.all([
    executeWebFrontend(),          // fullstack
    executeMobileApp()             // mobile
  ]);
}

⚠️ TDD Enforcement: 所有开发必须遵循TDD：

RED: 先写失败的测试
GREEN: 写最少的代码通过测试
REFACTOR: 重构清理

Step 8.7: Update task plan progress

Edit: Replace "- [ ] 阶段4: 开发实现（并行：Web+Mobile）"
With:  "- [x] 阶段4: 开发实现（并行：Web+Mobile）"
Edit: Update "## 当前阶段" to "阶段5: 测试验证"

Proceed to Step 9 (Phase 5).

═════════════════════════════════════════════════════════════
✅ Phase 4 完成 - 开发实现
═════════════════════════════════════════════════════════════

🎯 开发平台: {platforms}
📊 开发策略: {priority}
💻 后端代码: {WORKTREE_PATH}/backend/
🌐 Web代码: {WORKTREE_PATH}/web/ (如适用)
📱 Mobile代码: {WORKTREE_PATH}/mobile/ (如适用)
📋 下一步: Phase 5 - 测试验证（测试工程师-并行修复）

Step 9: Execute Phase 5 - 测试验证（并行修复）

🆕 ENHANCED in v6.0: Integrate Superpowers dispatching-parallel-agents for independent test failures.

⚠️ MANDATORY: Read Previous Phase Outputs

Before executing this phase, you MUST read all previous outputs:

Read file: .claudedocs/ceo-system-architect_result.md
Read file: .claudedocs/ceo-fullstack-developer_result.md (if exists)

This ensures you have complete context from architecture and development.

Execution Steps

Step 9.1: Update task plan current phase

Replace: "## 当前阶段\n阶段4: 开发实现"
With: "## 当前阶段\n阶段5: 测试验证"

Step 9.2: Call Test Engineer agent

⚠️ IMPORTANT: Execute the following steps in order to ensure the agent completes BEFORE checking results.

Step 1: Launch agent

Task tool: Launch the ceo-test-engineer agent with the following task:

## 任务
1. 编写单元测试（覆盖率≥80%）
2. 编写集成测试
3. 编写E2E测试
4. 运行测试套件并报告结果

### 输出要求
- 输出测试报告到 .claudedocs/ceo-test-engineer_result.md
- 包含：测试结果、覆盖率、发现的缺陷

Save the returned task_id as {TEST_ENGINEER_TASK_ID}

Step 2: Wait for agent completion

TaskOutput: Wait for {TEST_ENGINEER_TASK_ID}
Parameters: block=true, timeout=600000

⚠️ CRITICAL: DO NOT PROCEED until agent completes

Step 3: Verify output file exists

Read file: .claudedocs/ceo-test-engineer_result.md

⚠️ If file doesn't exist, agent failed - inform user and ask what to do

Step 9.3: Check test results

Analyze the test results from the file read above:

If all tests pass → Proceed to Step 10 (Phase 6)

If there are test failures → Proceed to Step 9.4

Step 9.4: Invoke parallel-dispatch skill

Use the Task tool to fix failures in parallel:

Task tool: Invoke the parallel-dispatch skill (from superpowers)

Execute the skill and after all tests pass, proceed to Step 10 (Phase 6).

The skill will:
1. Group test failures by independent domains
2. Launch parallel ceo-test-engineer agents for each domain
3. Wait for all agents to complete
4. Review and integrate fixes
5. Re-test until all pass

Step 9.5: Update task plan progress After all tests pass:

Edit: Replace "- [ ] 阶段5: 测试验证（测试工程师-并行修复）"
With:  "- [x] 阶段5: 测试验证（测试工程师-并行修复）"
Edit: Update "## 当前阶段" to "阶段6: 交付部署"

Proceed to Step 10 (Phase 6).

═════════════════════════════════════════════════════════════
✅ Phase 5 完成 - 测试验证
═════════════════════════════════════════════════════════════

📊 测试报告: .claudedocs/ceo-test-engineer_result.md
✅ 所有测试通过
📋 下一步: Phase 6 - 交付部署（市场营销师）

Step 10: Execute Phase 6 - 交付部署

⚠️ MANDATORY: Read Previous Phase Outputs

Before executing this phase, you MUST read all previous outputs:

Read file: .claudedocs/ceo-test-engineer_result.md

This ensures you have complete context from testing phase.

Update task plan current phase

Use Edit tool to update task_plan.md to "阶段6: 交付部署"

Call Marketing Specialist agent

⚠️ IMPORTANT: Execute the following steps in order to ensure the agent completes BEFORE displaying results.

Step 1: Launch agent

Task tool: Launch the ceo-marketing-specialist agent with the following context:

## CEO任务上下文

### 用户输入
{USER_INPUT}

### 所有前期阶段输出
[使用Read工具读取所有前期输出文件]

### 用户回答
{USER_ANSWERS}

### 你的任务
1. 编写用户文档
2. 编写部署文档
3. 编写营销文案
4. 准备推广方案

### ⚠️ 关键约束 - 提问规则（如需确认）
- **最多提问5个问题**：降低用户认知负担
- **分批提问**：如果问题超过5个，分多次提问，每次最多5个
- **优先级排序**：先问最重要的交付相关问题

Save the returned task_id as {MARKETING_TASK_ID}

Step 2: Wait for agent completion

TaskOutput: Wait for {MARKETING_TASK_ID}
Parameters: block=true, timeout=300000

⚠️ CRITICAL: DO NOT PROCEED until agent completes

Step 3: Verify output file exists

Read file: .claudedocs/ceo-marketing-specialist_result.md

⚠️ If file doesn't exist, agent failed - inform user and ask what to do

After agent completes and output is verified:

Use Edit tool to update task_plan.md:
- Mark Phase 6 as completed
- Update current phase to "完成"

Display completion message:

📦 阶段6完成 - 交付部署
📄 交付文档: .claudedocs/ceo-marketing-specialist_result.md

Proceed to Step 11 (Final Summary)

═════════════════════════════════════════════════════════════
✅ Phase 6 完成 - 交付部署
═════════════════════════════════════════════════════════════

📦 交付文档: .claudedocs/ceo-marketing-specialist_result.md
📋 下一步: 生成最终交付物索引

Step 11: Final Summary

After all phases complete:

Create deliverable index using Write tool

Create .claudedocs/deliverable.md with the following content:

# 交付物索引

## 文档
1. [阶段0设计文档](.claudedocs/phase0-design.md)
2. [产品需求文档](.claudedocs/ceo-product-manager_result.md)
3. [设计文档](.claudedocs/ceo-ui-ux-designer_result.md)
4. [架构文档](.claudedocs/ceo-system-architect_result.md)
5. [代码实现](.claudedocs/ceo-fullstack-developer_result.md)
6. [测试报告](.claudedocs/ceo-test-engineer_result.md)
7. [部署文档](.claudedocs/ceo-marketing-specialist_result.md)

## 状态
- [task_plan.md](.claudedocs/task_plan.md)
- [notes.md](.claudedocs/notes.md)

## 代码
{WORKTREE_PATH}/src/ (应用代码)

Display completion message to user

═════════════════════════════════════════════════════════════
🎉 项目完成！
═════════════════════════════════════════════════════════════

所有阶段已完成，交付物已生成。

📋 交付物索引: .claudedocs/deliverable.md
🌳 工作树位置: {WORKTREE_PATH}

感谢使用CEO Agent v6.2.0！

Version History

Key Changes in v6.2.0

命令规范化改进

🔧 修复了 CLI 命令显示与文档不一致的问题
✅ 重命名命令文件：ceo.md → workflow.md
✅ 修改插件名称：ceo-skills → ceo
✅ 统一命令前缀为 /ceo:

新命令格式：

/ceo:workflow - 启动完整 workflow
/ceo:brainstorm - 需求探索
/ceo:write-plan - 创建实现计划
/ceo:execute-plan - 执行实现计划

Key Changes in v6.1.0

Bug Fix: 强制用户确认检查点生效

🐛 修复了使用 Task 工具调用 agent 时，强制检查点被跳过的问题
✅ 采用 TaskOutput 同步等待机制，确保 agent 完成后才进入检查点
✅ 修改了 5 处 agent 调用（Phase 1, 2, 3, 5, 6）
✅ 保留 agent 专业能力的同时，确保检查点强制执行

技术改进：

所有 agent 调用现在遵循三步流程：Launch → Wait → Verify
使用 TaskOutput(task_id, block=true) 等待 agent 完成
验证输出文件存在后才继续执行

Key Changes in v6.0

Integrates Superpowers frameworks:

New Phases:

✅ Phase 0: Brainstorming for requirement exploration
✅ Phase 3.5: Git worktrees for workspace isolation
✅ Phase 4.5: TDD enforcement (embedded in Phase 4)

Enhanced Phases:

✅ Phase 1: Input now includes Phase 0 design document
✅ Phase 4: Subagent-driven development with two-stage code review
✅ Phase 5: Parallel agent dispatch for independent test failures

Preserved from v5.7:

✅ 3 mandatory checkpoints (PRD, Architecture, Final)
✅ 6 professional agents
✅ Three-file state management system
✅ Maximum automation with minimal human intervention

Confirmation Policy (v6.0)

Required Confirmations (3 checkpoints - unchanged):

✅ Phase 1: Product Requirements Document (PRD)
✅ Phase 3: Technical Architecture Design
✅ Final: Project Completion Summary

No Confirmation Required:

Phase 0: Brainstorming (natural dialogue)
Phase 2: UI/UX Design (direct approval)
Phase 3.5: Git Worktrees (automatic, except directory choice on first use)
Phase 4: Development (automatic two-stage review replaces manual checkpoint)
Phase 5: Testing (automatic parallel fix)

Benefits

Quality Improvements:

📈 需求理解: 对话探索 + PRD → 减少返工
📈 开发质量: 子任务 + 两阶段审查 → 立即发现错误
📈 代码质量: TDD铁律 → 测试覆盖率提升
📈 测试效率: 并行修复 → 修复速度提升
📈 环境安全: 工作树隔离 → 避免污染

Technical Debt Reduction:

✅ 需求返工: 减少
✅ 代码缺陷: 立即发现
✅ 范围蔓延: 两阶段审查防止
✅ 环境污染: 工作树隔离
✅ 测试维护: 并行加速

Version History

v6.0 (2026-01-19): Integrate Superpowers frameworks
- New: Phase 0 (brainstorming)
- New: Phase 3.5 (git-worktrees)
- Enhanced: Phase 4 (subagent-driven-development + two-stage review)
- Enhanced: Phase 5 (parallel-agent-dispatch)
- Integrated: TDD enforcement
v5.7 (2025-01-18): Current baseline
v5.3 (2025-01-17): Enhanced mandatory checkpoints
v5.0 (2025-01-17): Direct SKILL orchestration

name

ceo

description

CEO Agent - Direct Orchestration (v6.6.0)

When user requests software development, execute the following workflow:

Architecture Overview

v6.0 integrates Superpowers frameworks:

Phase 0: Brainstorming for requirement exploration
Phase 3.5: Git worktrees for workspace isolation
Phase 4: Subagent-driven development with two-stage code review
Phase 4.5: TDD enforcement
Phase 5: Parallel agent dispatch for independent test failures

Key principles:

流程自洽：Ensure workflow is complete, non-redundant, and naturally connected
保留自动化：Retain maximum automation, only key checkpoints require human confirmation
符合规范：Follow Claude Skill standards, prioritize Claude Code CLI native capabilities

Step 1: Initialize State Files

⚠️ NOTE: CEO agents are already installed as part of the ceo-skills plugin. No agent verification or installation is required.

Note: The Write tool will automatically create the .claudedocs directory if it doesn't exist.

Write initial task plan using Write tool

Create .claudedocs/task_plan.md with the following content:

# 任务计划

## Workflow配置 🆕 v6.5.0
- 执行模式: 待定（Step 1.5选择）
- 自动流转: 待定

## 用户需求
{USER_INPUT}

## 工作区配置
🔧 v6.5.0: 在阶段3.5创建Git Worktree后填充
- 工作区类型: Git Worktree
- 项目路径: 待定（阶段3.5创建）
- Git 分支: 待定（阶段3.5创建）

## 当前阶段
初始化

## 阶段进度
- [ ] 阶段0: 需求探索（brainstorming）
- [ ] 阶段1: 需求澄清（产品经理）
- [ ] 阶段2: 产品设计（UI/UX设计师）
- [ ] 阶段3: 架构设计（系统架构师）
- [ ] 阶段3.3: 平台决策（Web/Mobile/Both）🆕
- [ ] 阶段3.5: 工作区准备（git-worktrees）🔧 v6.5: 解决权限问题
- [ ] 阶段4: 开发实现（并行：Web+Mobile）🆕
- [ ] 阶段5: 测试验证（测试工程师-并行修复）
- [ ] 阶段6: 交付部署（市场营销师）

## 全局目标
1. 理解并澄清用户需求
2. 设计符合用户期望的产品
3. 实现高质量、可维护的代码
4. 确保充分测试和验证
5. 交付完整的文档和部署方案

Step 1.5: 选择Workflow执行模式 🆕 v6.5.0

Purpose: 在开始workflow前，让用户选择执行模式。

模式说明

🤖 自动模式（推荐）：完全自动化执行，跳过所有确认检查点，适合快速原型开发
👤 交互模式：在关键阶段暂停，等待用户确认，适合重要项目

Execution

Use AskUserQuestion tool:

Question: "选择CEO Workflow执行模式："
Header: "⚙️ Workflow模式配置"
Options:
  - label: "🤖 自动模式（推荐）"
    description: "完全自动化执行，跳过所有确认检查点，适合快速原型开发"
  - label: "👤 交互模式"
    description: "在每个关键阶段暂停，等待用户确认，适合重要项目"

Save user selection to task_plan.md:

If user selects "🤖 自动模式":

Edit file: .claudedocs/task_plan.md
Replace: "## Workflow配置\n- 执行模式: 待定（Step 1.5选择）\n- 自动流转: 待定"
With:
"""
## Workflow配置
- 执行模式: 自动模式
- 自动流转: 启用
- 确认检查点: 跳过
"""

If user selects "👤 交互模式":

Edit file: .claudedocs/task_plan.md
Replace: "## Workflow配置\n- 执行模式: 待定（Step 1.5选择）\n- 自动流转: 待定"
With:
"""
## Workflow配置
- 执行模式: 交互模式
- 自动流转: 禁用
- 确认检查点: 启用
"""

Display confirmation message:

═════════════════════════════════════════════════════════════
✅ Workflow模式已配置
═════════════════════════════════════════════════════════════

执行模式: {用户选择的模式}
{if 自动模式} 🤖 自动模式：将自动流转到所有阶段，无需手动确认
{if 交互模式} 👤 交互模式：将在关键阶段等待您确认

📋 下一步: Phase 0 - 需求探索

Proceed to Step 2 (Phase 0).

Git 分支: 待定（阶段3.5创建）

当前阶段

初始化

阶段进度

阶段0: 需求探索（brainstorming）
阶段1: 需求澄清（产品经理）
阶段2: 产品设计（UI/UX设计师）
阶段3: 架构设计（系统架构师）
阶段3.3: 平台决策（Web/Mobile/Both）🆕
阶段3.5: 工作区准备（git-worktrees）🔧 v6.5: 解决权限问题
阶段4: 开发实现（并行：Web+Mobile）🆕
阶段5: 测试验证（测试工程师-并行修复）
阶段6: 交付部署（市场营销师）

全局目标

理解并澄清用户需求
设计符合用户期望的产品
实现高质量、可维护的代码
确保充分测试和验证
交付完整的文档和部署方案


### Initialize notes file
Create `.claudedocs/notes.md` with the following content:
```markdown
# 项目笔记

## 初始化
项目启动时间: [执行时自动记录]

Step 2: Detect Current Phase and Resume

⚠️ CRITICAL: Before starting any phase, check if there's an existing workflow to resume.

Check for existing task plan

Use Read tool to check if task plan exists:

Read file: .claudedocs/task_plan.md

If task_plan.md exists (Resume Mode)

Parse current phase: Read "## 当前阶段" section
Check progress: Read "## 阶段进度" to see completed phases
Jump to next phase: Use the mapping below

Phase mapping:

"初始化" OR "阶段0: 需求探索" → Go to Step 3 (Phase 0)
"阶段1: 需求澄清" → Go to Step 4 (Phase 1)
"阶段2: 产品设计" → Go to Step 5 (Phase 2)
"阶段3: 架构设计" → Go to Step 6 (Phase 3)
"阶段3.5: 工作区准备" → Go to Step 7 (Phase 3.5)
"阶段4: 开发实现" → Go to Step 8 (Phase 4)
"阶段5: 测试验证" → Go to Step 9 (Phase 5)
"阶段6: 交付部署" → Go to Step 10 (Phase 6)

Display resume message:

═════════════════════════════════════════════════════════════
🔄 恢复工作流
═════════════════════════════════════════════════════════════

检测到现有任务计划，将从 {CURRENT_PHASE} 继续执行。

Then jump to the appropriate step above.

If task_plan.md doesn't exist (Fresh Start)

Display initialization message:

═════════════════════════════════════════════════════════════
🚀 启动新的 CEO 工作流
═════════════════════════════════════════════════════════════

将创建新的任务计划并执行完整 6 阶段开发流程。

Proceed to Step 3 (Phase 0).

Step 3: Execute Phase 0 - 需求探索（Brainstorming）

🆕 NEW in v6.0: Integrate Superpowers brainstorming for requirement exploration.

Purpose

Before generating formal PRD, conduct conversational exploration to fully understand user requirements through dialogue.

Process

Follow the brainstorming process:

digraph phase0_brainstorming {
    rankdir=TB;
    start [label="用户输入模糊需求", shape=ellipse];
    context [label="检查项目状态\n（文件、文档、最近提交）", shape=box];
    questions [label="苏格拉底式提问\n（一次一个问题）", shape=box, style=filled, fillcolor="#e1f5ff"];
    approaches [label="提出2-3种方案\n（含权衡分析）", shape=box];
    design [label="分段展示设计\n（200-300词/段）", shape=box];
    validate [label="每段后验证\n'这看起来对吗？'", shape=diamond];
    doc [label="保存设计文档\n.claudedocs/phase0-design.md", shape=box];

    start → context;
    context → questions;
    questions → approaches;
    approaches → design;
    design → validate;
    validate → design [label="否，调整"];
    validate → doc [label="是"];
}

Key Rules

提问规则:

一次一个问题：不要用多个问题淹没用户
优先选择题：比开放性问题更容易回答
聚焦理解：目的、约束、成功标准

设计展示规则:

分段展示：每段200-300词
每段验证：展示后询问"这看起来对吗？"
灵活调整：如有不清楚，返回澄清

产出物:

设计文档：.claudedocs/phase0-design.md
包含内容：架构、组件、数据流、错误处理、测试策略

Execution

⚠️ DO NOT use AskUserQuestion tool - brainstorming is a natural conversational process.

Invoke the brainstorming skill (from superpowers), execute it, then continue to Phase 1.

After brainstorming completes, save design document to:

.claudedocs/phase0-design.md

Then update task_plan.md to mark Phase 0 as completed:

Edit: Replace "- [ ] 阶段0: 需求探索（brainstorming）"
With:  "- [x] 阶段0: 需求探索（brainstorming）"
Edit: Replace "## 当前阶段\n初始化"
With: "## 当前阶段\n阶段0: 需求探索（完成）"

Proceed to Phase 1.

═════════════════════════════════════════════════════════════
✅ Phase 0 完成 - 需求探索
═════════════════════════════════════════════════════════════

📄 设计文档: .claudedocs/phase0-design.md
📋 下一步: Phase 1 - 需求澄清（产品经理）

Step 4: Execute Phase 1 - 需求澄清（产品经理）

⚠️ MANDATORY: Read Previous Phase Output

Before executing this phase, you MUST read all previous outputs:

Read file: .claudedocs/phase0-design.md

This ensures you have complete context from Phase 0.

Update task plan current phase

Use Edit tool to update task_plan.md:

Replace: "## 当前阶段\n初始化"
With: "## 当前阶段\n阶段1: 需求澄清"

Call Product Manager agent

⚠️ IMPORTANT: Execute the following steps in order to ensure the agent completes BEFORE the confirmation checkpoint.

Step 1: Launch agent

Task tool: Launch the ceo-product-manager agent with the following context:

## CEO任务上下文

### 用户输入
{USER_INPUT}

### 阶段0输出（NEW - 设计文档）
[使用Read工具读取 .claudedocs/phase0-design.md 内容]

### 你的任务
1. 基于阶段0设计文档生成产品需求文档（PRD）
2. 构建用户画像
3. 定义MVP范围
4. 识别未澄清的问题（如有）

### ⚠️ 关键约束 - 提问规则
- **最多提问5个问题**：降低用户认知负担
- **分批提问**：如果问题超过5个，分多次提问，每次最多5个
- **优先级排序**：先问最重要、最核心的问题

### 输出要求
- 输出完整的PRD文档到 .claudedocs/ceo-product-manager_result.md
- 包含用户画像、功能列表、优先级
- 如果有未问的问题，在文档末尾列出"待确认的问题"

Save the returned task_id as {PRODUCT_MANAGER_TASK_ID}

Step 2: Wait for agent completion

TaskOutput: Wait for {PRODUCT_MANAGER_TASK_ID}
Parameters: block=true, timeout=300000

⚠️ CRITICAL: DO NOT PROCEED until agent completes
⚠️ DO NOT proceed to confirmation checkpoint until this step completes

Step 3: Verify output file exists

Read file: .claudedocs/ceo-product-manager_result.md

⚠️ If file doesn't exist, agent failed - inform user and ask what to do

After agent completes and output is verified, proceed to confirmation checkpoint below.

Step 4.2: CONDITIONAL - User Confirmation Checkpoint 🆕 v6.5.0

⚠️ CONDITIONAL CHECKPOINT - 根据workflow模式决定是否需要确认

Check workflow mode first:

Read file: .claudedocs/task_plan.md

Extract:
  - 执行模式: (自动模式 | 交互模式)

If 执行模式 == "自动模式":

✅ 自动模式：跳过确认检查点

直接执行批准操作：
1. Use Edit tool to update task_plan.md:
   Replace: "## 当前阶段\n阶段1: 需求澄清"
   With: "## 当前阶段\n阶段2: 产品设计"
   Replace: "- [ ] 阶段0: 需求探索（brainstorming）"
   With: "- [x] 阶段0: 需求探索（brainstorming）"
   Replace: "- [ ] 阶段1: 需求澄清（产品经理）"
   With: "- [x] 阶段1: 需求澄清（产品经理）"

2. Display message:

🤖 执行模式: 自动模式 📄 PRD文档: .claudedocs/ceo-product-manager_result.md ✅ 状态: 已自动批准，继续下一阶段

📋 下一步: Phase 2 - 产品设计（UI/UX设计师）


3. Proceed to Step 5 (Phase 2).

If 执行模式 == "交互模式":

👤 交互模式：执行用户确认流程

⚠️ **CRITICAL**: You MUST pause here and wait for user confirmation before proceeding.

First, use Read tool to display PRD preview:

Read file: .claudedocs/ceo-product-manager_result.md Limit: 50 lines Display to user with formatted header: "📋 产品需求文档预览"


Then, use AskUserQuestion tool to get user confirmation:

Question: "请查看产品需求文档（PRD）并提供反馈。是否批准此PRD？" Header: "🎯 检查点 1 - 产品需求文档确认" Options:

label: "✅ 批准PRD" description: "PRD符合预期，批准并继续下一阶段"
label: "📝 修改PRD" description: "我有修改意见，需要调整PRD"
label: "🔄 重做PRD" description: "PRD不符合预期，需要重新澄清需求"
label: "🛑 终止workflow" description: "结束整个开发流程"


⚠️ **DO NOT PROCEED** until user selects an option.

Step 4.3: Process User Decision

If user selects ✅ 批准PRD:

Use Edit tool to update task_plan.md:
- Mark Phase 0 and Phase 1 as completed
- Update current phase to "阶段2: 产品设计"
Proceed to Step 5 (Phase 2).

═════════════════════════════════════════════════════════════
✅ Phase 1 完成 - 需求澄清
═════════════════════════════════════════════════════════════

📄 PRD文档: .claudedocs/ceo-product-manager_result.md
📋 下一步: Phase 2 - 产品设计（UI/UX设计师）

If user selects 📝 修改PRD:

Use AskUserQuestion to collect specific modification requests
Call Product Manager again with feedback
After revision completes, repeat Step 4.2 (confirmation checkpoint)

If user selects 🔄 重做PRD:

Use AskUserQuestion to collect new requirements
Call Product Manager for new round
After new PRD completes, repeat Step 4.2 (confirmation checkpoint)

If user selects 🛑 终止workflow:

Display termination message
Update task_plan.md with termination status
End workflow

Step 5: Execute Phase 2 - 产品设计

⚠️ MANDATORY: Read Previous Phase Outputs

Before executing this phase, you MUST read all previous outputs:

Read file: .claudedocs/phase0-design.md
Read file: .claudedocs/ceo-product-manager_result.md

This ensures you have complete context from Phase 0 and Phase 1.

Update task plan current phase

Use Edit tool to update task_plan.md:

Replace: "## 当前阶段\n阶段1: 需求澄清"
With: "## 当前阶段\n阶段2: 产品设计"

Call UI/UX Designer agent

⚠️ IMPORTANT: Execute the following steps in order to ensure the agent completes BEFORE proceeding.

Step 1: Launch agent

Task tool: Launch the ceo-ui-ux-designer agent with the following context:

## CEO任务上下文

### 用户输入
{USER_INPUT}

### 阶段0-1输出
[使用Read工具读取阶段0设计文档和阶段1 PRD]

### 用户对阶段0-1问题的回答
{USER_ANSWERS_PHASE0_1}

### 你的任务
1. 基于PRD设计用户故事
2. 设计交互流程
3. 设计视觉界面
4. 创建原型设计

### ⚠️ 关键约束 - 提问规则
- **最多提问5个问题**：降低用户认知负担
- **分批提问**：如果问题超过5个，分多次提问，每次最多5个
- **优先级排序**：先问最重要、最核心的设计问题

### 输出要求
- 输出完整的设计文档到 .claudedocs/ceo-ui-ux-designer_result.md
- 包含用户故事、交互流程、视觉设计

Save the returned task_id as {UI_UX_TASK_ID}

Step 2: Wait for agent completion

TaskOutput: Wait for {UI_UX_TASK_ID}
Parameters: block=true, timeout=300000

⚠️ CRITICAL: DO NOT PROCEED until agent completes

Step 3: Verify output file exists

Read file: .claudedocs/ceo-ui-ux-designer_result.md

⚠️ If file doesn't exist, agent failed - inform user and ask what to do

After agent completes and output is verified:

Use Edit tool to update task_plan.md: Mark Phase 2 as completed
Proceed directly to Phase 3 (no confirmation required).

═════════════════════════════════════════════════════════════
✅ Phase 2 完成 - 产品设计
═════════════════════════════════════════════════════════════

📄 设计文档: .claudedocs/ceo-ui-ux-designer_result.md
📋 下一步: Phase 3 - 架构设计（系统架构师）

Step 6: Execute Phase 3 - 架构设计

⚠️ MANDATORY: Read Previous Phase Outputs

Before executing this phase, you MUST read all previous outputs:

Read file: .claudedocs/phase0-design.md
Read file: .claudedocs/ceo-product-manager_result.md
Read file: .claudedocs/ceo-ui-ux-designer_result.md

This ensures you have complete context from Phase 0, 1, and 2.

Update task plan current phase

Use Edit tool to update task_plan.md:

Replace: "## 当前阶段\n阶段2: 产品设计"
With: "## 当前阶段\n阶段3: 架构设计"

Call System Architect agent

⚠️ CRITICAL: Execute the following steps in order to ensure the agent completes BEFORE the confirmation checkpoint.

Step 1: Launch agent

Task tool: Launch the ceo-system-architect agent with the following context:

## CEO任务上下文

### 用户输入
{USER_INPUT}

### 前期阶段输出
[使用Read工具读取所有前期输出文件]

### 用户回答
{USER_ANSWERS}

### 你的任务
1. 技术栈选型（前端、后端、数据库）
2. 系统架构设计
3. API规范设计
4. 数据模型设计

### ⚠️ 关键约束 - 提问规则
- **最多提问5个问题**：降低用户认知负担
- **分批提问**：如果技术决策问题超过5个，分多次提问
- **优先级排序**：先问最关键的技术选型问题

### 输出要求
- 输出完整的架构设计文档到 .claudedocs/ceo-system-architect_result.md

Save the returned task_id as {ARCHITECT_TASK_ID}

Step 2: Wait for agent completion

TaskOutput: Wait for {ARCHITECT_TASK_ID}
Parameters: block=true, timeout=300000

⚠️ CRITICAL: DO NOT PROCEED until agent completes
⚠️ DO NOT proceed to confirmation checkpoint until this step completes

Step 3: Verify output file exists

Read file: .claudedocs/ceo-system-architect_result.md

⚠️ If file doesn't exist, agent failed - inform user and ask what to do
⚠️ DO NOT proceed to confirmation checkpoint if file doesn't exist

After agent completes and output is verified, CHECK WORKFLOW MODE before proceeding.

⚠️ DO NOT PROCEED to Phase 3.5 until user confirms the architecture (交互模式) or auto-approves (自动模式)!

Step 6.1: CONDITIONAL - Architecture Confirmation Checkpoint 🆕 v6.5.0

🚨 CONDITIONAL CHECKPOINT - 根据workflow模式决定是否需要确认

Check workflow mode first:

Read file: .claudedocs/task_plan.md

Extract:
  - 执行模式: (自动模式 | 交互模式)

If 执行模式 == "自动模式":

✅ 自动模式：跳过确认检查点

直接执行批准操作：
1. Use Edit tool to update task_plan.md:
   Replace: "## 当前阶段\n阶段3: 架构设计"
   With: "## 当前阶段\n阶段3.3: 平台决策"
   Replace: "- [ ] 阶段3: 架构设计（系统架构师）"
   With: "- [x] 阶段3: 架构设计（系统架构师）"

2. Display message:

🤖 执行模式: 自动模式 📄 架构文档: .claudedocs/ceo-system-architect_result.md ✅ 状态: 已自动批准，继续平台决策

📋 下一步: Phase 3.3 - 平台决策


3. Proceed to Step 6.3 (Phase 3.3).

If 执行模式 == "交互模式":

👤 交互模式：执行用户确认流程

🚨 **CRITICAL CHECKPOINT - MANDATORY USER CONFIRMATION REQUIRED**

You are at the Architecture Confirmation Checkpoint. You MUST execute this step before proceeding to Phase 3.5.

**Step 1**: Display architecture document preview using Read tool:

Read file: .claudedocs/ceo-system-architect_result.md Limit: 50 lines Display formatted header: "🏗️ 技术架构设计文档预览"


**Step 2**: Use AskUserQuestion tool to get user confirmation:

Question: "请查看技术架构设计文档并提供反馈。是否批准此架构方案？" Header: "🏗️ 检查点 2 - 技术架构方案确认" Options:

label: "✅ 批准架构" description: "架构方案符合预期，批准并继续创建工作区"
label: "📝 修改架构" description: "我有修改意见，需要调整架构设计"
label: "🔄 重做架构" description: "架构不符合预期，需要重新设计"
label: "🛑 终止workflow" description: "结束整个开发流程"


**Step 3**: ⚠️ **WAIT FOR USER RESPONSE - DO NOT PROCEED**

⚠️ **DO NOT PROCEED** until user selects an option.
⚠️ **DO NOT PROCEED to Phase 3.5** until user selects "✅ 批准架构".

Step 6.2: Process User Decision

If user selects ✅ 批准架构:

Use Edit tool to update task_plan.md:
- Mark Phase 3 as completed
- Update current phase to "阶段3.3: 平台决策"
Proceed to Step 6.3 (Phase 3.3 - 平台决策).

═════════════════════════════════════════════════════════════
✅ Phase 3 完成 - 架构设计
═════════════════════════════════════════════════════════════

📄 架构文档: .claudedocs/ceo-system-architect_result.md
📋 下一步: Phase 3.3 - 平台决策（Web/Mobile/Both）

Step 6.3: 🆕 Phase 3.3 - 平台决策（NEW in v6.4.0）

🆕 NEW in v6.4.0: 在架构设计后进行平台决策，动态决定开发Web、Mobile还是Both。

Purpose

根据产品需求、用户场景、功能需求和技术约束，决定开发平台：

Web Only: 仅Web应用
Mobile Only: 仅移动应用
Both: Web + Mobile并行开发

决策流程

Step 1: 分析决策因素

基于以下因素进行分析：

决策因素:
  用户因素:
    - 目标用户类型（办公用户/移动用户/全场景用户）
    - 主要使用设备（桌面/移动/混合）
    - 使用场景（固定/移动/灵活）

  功能因素:
    - 需要的原生功能（GPS、相机、传感器等）
    - 复杂表单处理（Web优势）
    - 实时协作（Web优势）
    - 离线需求（Mobile优势）

  技术因素:
    - 技术栈复杂度
    - 开发时间要求
    - 团队能力匹配

  约束因素:
    - 开发预算
    - 上市时间要求
    - 维护成本考虑

Step 2: 生成平台决策

基于分析生成决策文档：

interface PlatformDecision {
  // 目标平台
  platforms: ('web' | 'mobile')[];

  // 开发优先级
  priority: 'web-first' | 'mobile-first' | 'parallel';

  // 决策理由
  rationale: string;

  // 实施建议
  implementation: {
    phasedRollout: boolean;
    mvpPlatform: 'web' | 'mobile';
    featuresByPlatform: {
      web: string[];
      mobile: string[];
      shared: string[];
    };
  };
}

Step 3: 保存决策文档

使用Write工具创建平台决策文档：

Write file: .claudedocs/platform-decision.md

Content:
# 平台决策文档

## 决策结果
- **目标平台**: {web | mobile | both}
- **开发优先级**: {web-first | mobile-first | parallel}
- **决策理由**: {rationale}

## 实施计划
- **分阶段发布**: {yes | no}
- **MVP平台**: {web | mobile}

## 功能分配
- **Web独有功能**: [...]
- **Mobile独有功能**: [...]
- **共享功能**: [...]

Step 4: 更新任务计划

使用Edit工具更新task_plan.md：

Replace: "## 当前阶段\n阶段3: 架构设计"
With: "## 当前阶段\n阶段3.3: 平台决策"

Step 6.4: CONDITIONAL - 平台决策确认检查点 🆕 v6.5.0

⚠️ CONDITIONAL CHECKPOINT - 根据workflow模式决定是否需要确认

Check workflow mode first:

Read file: .claudedocs/task_plan.md

Extract:
  - 执行模式: (自动模式 | 交互模式)

If 执行模式 == "自动模式":

✅ 自动模式：跳过确认检查点

直接执行批准操作：
1. Use Edit tool to update task_plan.md:
   Replace: "## 当前阶段\n阶段3.3: 平台决策"
   With: "## 当前阶段\n阶段3.5: 工作区准备"
   Replace: "- [ ] 阶段3.3: 平台决策（Web/Mobile/Both）"
   With: "- [x] 阶段3.3: 平台决策（Web/Mobile/Both）"

2. Extract platform decision and add to global goals:

Read file: .claudedocs/platform-decision.md

Extract:

platforms: (web | mobile | both)
priority: (web-first | mobile-first | parallel)


Add to task_plan.md "## 全局目标":

开发平台: {platforms}
优先级策略: {priority}


3. Display message:

🤖 执行模式: 自动模式 📄 决策文档: .claudedocs/platform-decision.md 📱 开发平台: {platforms} ⚡ 优先级: {priority} ✅ 状态: 已自动批准，继续工作区准备

📋 下一步: Phase 3.5 - 工作区准备（Git Worktrees）


4. Proceed to Step 7 (Phase 3.5).

If 执行模式 == "交互模式":

👤 交互模式：执行用户确认流程

🚨 **CRITICAL CHECKPOINT - MANDATORY USER CONFIRMATION REQUIRED**

**Step 1**: 显示平台决策文档预览：

```bash
Read file: .claudedocs/platform-decision.md
Limit: 30 lines
Display formatted header: "📱 平台决策预览"

Step 2: 使用AskUserQuestion工具获取用户确认：

Question: "根据产品需求和技术架构分析，建议开发{platforms}，采用{priority}策略。是否批准此平台决策？"
Header: "📱 检查点 2.5 - 平台决策确认"
Options:
  - label: "✅ 批准决策"
    description: "同意平台决策，继续工作区准备"
  - label: "📝 修改决策"
    description: "有不同意见，需要调整平台决策"
  - label: "🛑 终止workflow"
    description: "结束整个开发流程"

Step 3: ⚠️ 等待用户响应 - 不要继续执行


### Step 6.5: 处理用户决策

**如果用户选择 ✅ 批准决策**:

1. 使用Edit工具更新task_plan.md：
   - 标记阶段3.3为已完成
   - 更新当前阶段为"阶段3.5: 工作区准备"
   - 添加平台决策信息到全局目标

```bash
Edit: task_plan.md
Add to "## 全局目标":
  - 开发平台: {platforms}
  - 开发策略: {priority}

继续执行Step 7 (Phase 3.5)

═════════════════════════════════════════════════════════════
✅ Phase 3.3 完成 - 平台决策
═════════════════════════════════════════════════════════════

📄 决策文档: .claudedocs/platform-decision.md
🎯 目标平台: {platforms}
📋 下一步: Phase 3.5 - 工作区准备（git-worktrees）

如果用户选择 📝 修改决策:

使用AskUserQuestion收集具体的修改意见
重新分析并生成新的平台决策
重复Step 6.4（确认检查点）

如果用户选择 🛑 终止workflow:

显示终止消息
更新task_plan.md为终止状态
结束workflow

If user selects 📝 修改架构:

Use AskUserQuestion to collect specific modification requests
Call System Architect again with feedback
After revision completes, repeat Step 6.1 (confirmation checkpoint)

If user selects 🔄 重做架构:

Use AskUserQuestion to collect new requirements
Call System Architect for new round
After new architecture completes, repeat Step 6.1 (confirmation checkpoint)

If user selects 🛑 终止workflow:

Display termination message
Update task_plan.md with termination status
End workflow

Step 7: Execute Phase 3.5 - 工作区准备（Git Worktrees）

🆕 NEW in v6.0: Integrate Superpowers using-git-worktrees for workspace isolation.

🔧 UPDATED in v6.5.0: 明确使用git-worktree解决Claude Code CLI跨目录访问权限问题。

Purpose

Before starting development, create isolated Git worktree to:

✅ 避免分支切换污染
✅ 解决跨目录访问权限问题（worktree属于项目内，无需额外确认）
✅ 提供独立的工作环境

Why Git Worktree Solves Permission Issues

问题根源：

# 当前工作目录
/path/to/ceo-skills-plugin/

# 操作目标（项目外）
/path/to/another-project/.claudedocs/
# ❌ Claude Code提示权限确认

Git Worktree解决方案：

# 创建worktree（项目内）
git worktree add ../my-project -b feature/my-project

# worktree目录结构
/path/to/repo/
├── .git/                    # 共享Git仓库
├── ceo-skills-plugin/        # 主worktree（当前目录）
└── my-project/               # 新worktree（共享.git）
    └── .git (worktree元数据)

# 操作目标（项目内）
../my-project/.claudedocs/
# ✅ worktree是Git仓库的一部分，无需权限确认

Execution Steps

Step 7.1: 确定项目名称和路径

从架构设计文档中提取项目名称：

Read file: .claudedocs/ceo-system-architect_result.md

Extract:
  - project_name: 项目名称

Step 7.2: 创建Git Worktree

# 计算worktree路径（相对于当前仓库）
WORKTREE_PATH="../${project_name}"
BRANCH_NAME="feature/${project_name}"

# 创建worktree
git worktree add ${WORKTREE_PATH} -b ${BRANCH_NAME}

# 验证创建成功
git worktree list

# 在worktree中创建文档目录（无需权限确认）
mkdir -p ${WORKTREE_PATH}/.claudedocs

Step 7.3: 保存Worktree信息到任务计划

Edit file: .claudedocs/task_plan.md

Add section after "## 用户需求":

## 工作区配置
- 工作区类型: Git Worktree
- 项目路径: ${WORKTREE_PATH}
- Git 分支: ${BRANCH_NAME}
- 相对路径: ../${project_name}

Replace: "## 当前阶段\n阶段3: 架构设计"
With: "## 当前阶段\n阶段3.5: 工作区准备"

Replace: "- [ ] 阶段3.5: 工作区准备（git-worktrees）"
With: "- [x] 阶段3.5: 工作区准备（git-worktrees）"

Step 7.4: 验证Worktree准备完成

# 列出所有worktrees
git worktree list

# 切换到worktree查看（可选）
cd ${WORKTREE_PATH}
pwd
ls -la

Proceed to Step 8 (Phase 4).

═════════════════════════════════════════════════════════════
✅ Phase 3.5 完成 - 工作区准备
═════════════════════════════════════════════════════════════

🌳 工作树路径: {WORKTREE_PATH}
📋 下一步: Phase 4 - 开发实现（子任务驱动）

Step 8: Execute Phase 4 - 开发实现（并行：Web + Mobile）

🆕 ENHANCED in v6.4.0: 支持并行Web和Mobile开发，根据平台决策动态激活agents。

🔧 UPDATED in v6.5.0: 所有项目操作在Git Worktree路径中执行，避免跨目录访问权限问题。

⚠️ MANDATORY: Read Previous Phase Outputs & Platform Decision

Before executing this phase, you MUST read all previous outputs AND the platform decision:

Read file: .claudedocs/phase0-design.md
Read file: .claudedocs/ceo-product-manager_result.md
Read file: .claudedocs/ceo-ui-ux-designer_result.md
Read file: .claudedocs/ceo-system-architect_result.md
Read file: .claudedocs/platform-decision.md  🆕 平台决策

This ensures you have complete context from all previous phases AND know which platforms to develop.

⚠️ CRITICAL: Load Worktree Configuration

🔧 v6.5.0: 从任务计划中加载worktree配置，确保所有项目操作在正确路径中执行：

Read file: .claudedocs/task_plan.md

Extract:
  - WORKTREE_PATH: 工作区路径（如 ../my-project）
  - BRANCH_NAME: Git分支（如 feature/my-project）

重要提示：

✅ 所有项目创建、文件操作、命令执行都在 ${WORKTREE_PATH} 中进行
✅ Worktree是Git仓库的一部分，属于"项目内"，无需权限确认
❌ 不要直接操作 /Users/zyb/.../my-project，使用相对路径 ${WORKTREE_PATH}

Execution Steps

Step 8.1: Update task plan current phase

Replace: "## 当前阶段\n阶段3.5: 工作区准备"
With: "## 当前阶段\n阶段4: 开发实现"

Step 8.2: 🆕 Parse Platform Decision

从平台决策文档中提取开发策略：

Read file: .claudedocs/platform-decision.md

Extract:
  - platforms: (web | mobile | both)
  - priority: (web-first | mobile-first | parallel)

Step 8.3: 🆕 根据平台决策配置开发任务

方案A: Backend Only（纯后端）

如果决策是纯后端项目（无前端需求）：

📋 开发策略：Backend Only

任务分配:
  - fullstack-developer: 后端API

执行步骤:
  1. 调用subagent-driven-development skill
  2. 激活ceo-fullstack-developer agent（仅后端）
  3. 任务分解：
     - 后端API实现
     - 数据库设计
     - 业务逻辑实现
  4. 执行两阶段代码审查

方案B: Web Frontend + Backend（传统Web）

如果决策是传统Web应用（React/Next.js等）：

📋 开发策略：Web Full Stack

任务分配:
  - fullstack-developer: 后端API + Web前端

执行步骤:
  1. 调用subagent-driven-development skill
  2. 激活ceo-fullstack-developer agent
  3. 任务分解：
     - 后端API实现
     - Web前端实现（React/Next.js）
  4. 执行两阶段代码审查

技术栈:
  - 后端: Node.js/Python/Go
  - 前端: React/Next.js/Vue（传统Web技术）

方案C: Cross-Platform Mobile（跨平台移动应用）

如果决策是跨平台移动应用（需要iOS + Android + Web）：

📋 开发策略：Cross-Platform Mobile

任务分配:
  - fullstack-developer: 后端API
  - mobile-developer: React Native应用（iOS + Android + Web）

执行步骤:
  1. 调用subagent-driven-development skill
  2. 并行激活agents:
     - ceo-fullstack-developer (后端API)
     - ceo-mobile-developer (React Native应用)
  3. 移动端开发依赖后端API完成
  4. 执行两阶段代码审查

技术栈:
  - 后端: Node.js/Python/Go
  - 移动端: React Native + Expo
  - 平台: iOS、Android、Web（React Native for Web）

说明:
  - React Native的"Web"是通过React Native for Web编译的
  - 这与传统Web前端（React/Next.js）不同
  - 一套代码覆盖三个平台

方案D: Web + Mobile Both（Web + 移动端）

如果决策是同时需要传统Web和移动应用：

📋 开发策略：Web + Mobile Both

任务分配:
  - fullstack-developer: 后端API + 传统Web前端
  - mobile-developer: React Native移动应用

根据优先级策略执行:

#### 策略1: web-first

执行顺序:
  1. 后端API (fullstack-developer)
  2. 传统Web前端 (fullstack-developer)
  3. 移动端 (mobile-developer)

#### 策略2: mobile-first

执行顺序:
  1. 后端API (fullstack-developer)
  2. 移动端 (mobile-developer)
  3. 传统Web前端 (fullstack-developer)

#### 策略3: parallel

执行顺序:
  1. 后端API (fullstack-developer) - 必须首先完成
  2. 并行开发:
     - 传统Web前端 (fullstack-developer)
     - 移动端 (mobile-developer)

技术栈:
  - 后端: Node.js/Python/Go
  - Web前端: React/Next.js（传统Web）
  - 移动端: React Native + Expo

Step 8.4: Extract development tasks from architecture

Read the architecture document and identify all development tasks. Group them into 2-5 minute subtasks.

⚠️ Note: All operations should use absolute paths based on {WORKTREE_PATH}.

Step 8.5: 🆕 根据平台策略执行开发

后端API开发（始终执行）

# 后端API由fullstack-developer统一实现
Task tool: Launch ceo-fullstack-developer agent

任务:
  1. 实现后端API（按照architect设计的API规范）
  2. 实现数据模型
  3. 实现业务逻辑
  4. API文档生成

输出: .claudedocs/backend-api-result.md

Web前端开发（按需执行）

如果 platforms.includes('web')：

Task tool: Launch ceo-fullstack-developer agent

任务:
  1. 实现Web前端组件
  2. 实现Web状态管理
  3. 集成后端API
  4. 响应式实现

依赖: 后端API完成

输出: .claudedocs/web-frontend-result.md

移动端开发（按需执行）

如果 platforms.includes('mobile')：

Task tool: Launch ceo-mobile-developer agent

任务:
  1. 初始化移动项目（使用obytes模板）
  2. 实现移动端UI
  3. 实现移动端状态管理
  4. 集成后端API
  5. 原生功能集成（如需要）

依赖: 后端API完成

输出: .claudedocs/mobile-app-result.md

Step 8.6: 🆕 并行开发协调

根据 priority 策略协调开发：

// web-first 策略
if (priority === 'web-first') {
  await executeBackendAPI();     // fullstack
  await executeWebFrontend();     // fullstack
  await executeMobileApp();       // mobile
}

// mobile-first 策略
else if (priority === 'mobile-first') {
  await executeBackendAPI();      // fullstack
  await executeMobileApp();        // mobile
  await executeWebFrontend();      // fullstack
}

// parallel 策略
else if (priority === 'parallel') {
  await executeBackendAPI();       // fullstack (必须先完成)

  // Web和Mobile并行开发
  await Promise.all([
    executeWebFrontend(),          // fullstack
    executeMobileApp()             // mobile
  ]);
}

⚠️ TDD Enforcement: 所有开发必须遵循TDD：

RED: 先写失败的测试
GREEN: 写最少的代码通过测试
REFACTOR: 重构清理

Step 8.7: Update task plan progress

Edit: Replace "- [ ] 阶段4: 开发实现（并行：Web+Mobile）"
With:  "- [x] 阶段4: 开发实现（并行：Web+Mobile）"
Edit: Update "## 当前阶段" to "阶段5: 测试验证"

Proceed to Step 9 (Phase 5).

═════════════════════════════════════════════════════════════
✅ Phase 4 完成 - 开发实现
═════════════════════════════════════════════════════════════

🎯 开发平台: {platforms}
📊 开发策略: {priority}
💻 后端代码: {WORKTREE_PATH}/backend/
🌐 Web代码: {WORKTREE_PATH}/web/ (如适用)
📱 Mobile代码: {WORKTREE_PATH}/mobile/ (如适用)
📋 下一步: Phase 5 - 测试验证（测试工程师-并行修复）

Step 9: Execute Phase 5 - 测试验证（并行修复）

🆕 ENHANCED in v6.0: Integrate Superpowers dispatching-parallel-agents for independent test failures.

⚠️ MANDATORY: Read Previous Phase Outputs

Before executing this phase, you MUST read all previous outputs:

Read file: .claudedocs/ceo-system-architect_result.md
Read file: .claudedocs/ceo-fullstack-developer_result.md (if exists)

This ensures you have complete context from architecture and development.

Execution Steps

Step 9.1: Update task plan current phase

Replace: "## 当前阶段\n阶段4: 开发实现"
With: "## 当前阶段\n阶段5: 测试验证"

Step 9.2: Call Test Engineer agent

⚠️ IMPORTANT: Execute the following steps in order to ensure the agent completes BEFORE checking results.

Step 1: Launch agent

Task tool: Launch the ceo-test-engineer agent with the following task:

## 任务
1. 编写单元测试（覆盖率≥80%）
2. 编写集成测试
3. 编写E2E测试
4. 运行测试套件并报告结果

### 输出要求
- 输出测试报告到 .claudedocs/ceo-test-engineer_result.md
- 包含：测试结果、覆盖率、发现的缺陷

Save the returned task_id as {TEST_ENGINEER_TASK_ID}

Step 2: Wait for agent completion

TaskOutput: Wait for {TEST_ENGINEER_TASK_ID}
Parameters: block=true, timeout=600000

⚠️ CRITICAL: DO NOT PROCEED until agent completes

Step 3: Verify output file exists

Read file: .claudedocs/ceo-test-engineer_result.md

⚠️ If file doesn't exist, agent failed - inform user and ask what to do

Step 9.3: Check test results

Analyze the test results from the file read above:

If all tests pass → Proceed to Step 10 (Phase 6)

If there are test failures → Proceed to Step 9.4

Step 9.4: Invoke parallel-dispatch skill

Use the Task tool to fix failures in parallel:

Task tool: Invoke the parallel-dispatch skill (from superpowers)

Execute the skill and after all tests pass, proceed to Step 10 (Phase 6).

The skill will:
1. Group test failures by independent domains
2. Launch parallel ceo-test-engineer agents for each domain
3. Wait for all agents to complete
4. Review and integrate fixes
5. Re-test until all pass

Step 9.5: Update task plan progress After all tests pass:

Edit: Replace "- [ ] 阶段5: 测试验证（测试工程师-并行修复）"
With:  "- [x] 阶段5: 测试验证（测试工程师-并行修复）"
Edit: Update "## 当前阶段" to "阶段6: 交付部署"

Proceed to Step 10 (Phase 6).

═════════════════════════════════════════════════════════════
✅ Phase 5 完成 - 测试验证
═════════════════════════════════════════════════════════════

📊 测试报告: .claudedocs/ceo-test-engineer_result.md
✅ 所有测试通过
📋 下一步: Phase 6 - 交付部署（市场营销师）

Step 10: Execute Phase 6 - 交付部署

⚠️ MANDATORY: Read Previous Phase Outputs

Before executing this phase, you MUST read all previous outputs:

Read file: .claudedocs/ceo-test-engineer_result.md

This ensures you have complete context from testing phase.

Update task plan current phase

Use Edit tool to update task_plan.md to "阶段6: 交付部署"

Call Marketing Specialist agent

⚠️ IMPORTANT: Execute the following steps in order to ensure the agent completes BEFORE displaying results.

Step 1: Launch agent

Task tool: Launch the ceo-marketing-specialist agent with the following context:

## CEO任务上下文

### 用户输入
{USER_INPUT}

### 所有前期阶段输出
[使用Read工具读取所有前期输出文件]

### 用户回答
{USER_ANSWERS}

### 你的任务
1. 编写用户文档
2. 编写部署文档
3. 编写营销文案
4. 准备推广方案

### ⚠️ 关键约束 - 提问规则（如需确认）
- **最多提问5个问题**：降低用户认知负担
- **分批提问**：如果问题超过5个，分多次提问，每次最多5个
- **优先级排序**：先问最重要的交付相关问题

Save the returned task_id as {MARKETING_TASK_ID}

Step 2: Wait for agent completion

TaskOutput: Wait for {MARKETING_TASK_ID}
Parameters: block=true, timeout=300000

⚠️ CRITICAL: DO NOT PROCEED until agent completes

Step 3: Verify output file exists

Read file: .claudedocs/ceo-marketing-specialist_result.md

⚠️ If file doesn't exist, agent failed - inform user and ask what to do

After agent completes and output is verified:

Use Edit tool to update task_plan.md:
- Mark Phase 6 as completed
- Update current phase to "完成"

Display completion message:

📦 阶段6完成 - 交付部署
📄 交付文档: .claudedocs/ceo-marketing-specialist_result.md

Proceed to Step 11 (Final Summary)

═════════════════════════════════════════════════════════════
✅ Phase 6 完成 - 交付部署
═════════════════════════════════════════════════════════════

📦 交付文档: .claudedocs/ceo-marketing-specialist_result.md
📋 下一步: 生成最终交付物索引

Step 11: Final Summary

After all phases complete:

Create deliverable index using Write tool

Create .claudedocs/deliverable.md with the following content:

# 交付物索引

## 文档
1. [阶段0设计文档](.claudedocs/phase0-design.md)
2. [产品需求文档](.claudedocs/ceo-product-manager_result.md)
3. [设计文档](.claudedocs/ceo-ui-ux-designer_result.md)
4. [架构文档](.claudedocs/ceo-system-architect_result.md)
5. [代码实现](.claudedocs/ceo-fullstack-developer_result.md)
6. [测试报告](.claudedocs/ceo-test-engineer_result.md)
7. [部署文档](.claudedocs/ceo-marketing-specialist_result.md)

## 状态
- [task_plan.md](.claudedocs/task_plan.md)
- [notes.md](.claudedocs/notes.md)

## 代码
{WORKTREE_PATH}/src/ (应用代码)

Display completion message to user

═════════════════════════════════════════════════════════════
🎉 项目完成！
═════════════════════════════════════════════════════════════

所有阶段已完成，交付物已生成。

📋 交付物索引: .claudedocs/deliverable.md
🌳 工作树位置: {WORKTREE_PATH}

感谢使用CEO Agent v6.2.0！

Version History

Key Changes in v6.2.0

命令规范化改进

🔧 修复了 CLI 命令显示与文档不一致的问题
✅ 重命名命令文件：ceo.md → workflow.md
✅ 修改插件名称：ceo-skills → ceo
✅ 统一命令前缀为 /ceo:

新命令格式：

/ceo:workflow - 启动完整 workflow
/ceo:brainstorm - 需求探索
/ceo:write-plan - 创建实现计划
/ceo:execute-plan - 执行实现计划

Key Changes in v6.1.0

Bug Fix: 强制用户确认检查点生效

🐛 修复了使用 Task 工具调用 agent 时，强制检查点被跳过的问题
✅ 采用 TaskOutput 同步等待机制，确保 agent 完成后才进入检查点
✅ 修改了 5 处 agent 调用（Phase 1, 2, 3, 5, 6）
✅ 保留 agent 专业能力的同时，确保检查点强制执行

技术改进：

所有 agent 调用现在遵循三步流程：Launch → Wait → Verify
使用 TaskOutput(task_id, block=true) 等待 agent 完成
验证输出文件存在后才继续执行

Key Changes in v6.0

Integrates Superpowers frameworks:

New Phases:

✅ Phase 0: Brainstorming for requirement exploration
✅ Phase 3.5: Git worktrees for workspace isolation
✅ Phase 4.5: TDD enforcement (embedded in Phase 4)

Enhanced Phases:

✅ Phase 1: Input now includes Phase 0 design document
✅ Phase 4: Subagent-driven development with two-stage code review
✅ Phase 5: Parallel agent dispatch for independent test failures

Preserved from v5.7:

✅ 3 mandatory checkpoints (PRD, Architecture, Final)
✅ 6 professional agents
✅ Three-file state management system
✅ Maximum automation with minimal human intervention

Confirmation Policy (v6.0)

Required Confirmations (3 checkpoints - unchanged):

✅ Phase 1: Product Requirements Document (PRD)
✅ Phase 3: Technical Architecture Design
✅ Final: Project Completion Summary

No Confirmation Required:

Phase 0: Brainstorming (natural dialogue)
Phase 2: UI/UX Design (direct approval)
Phase 3.5: Git Worktrees (automatic, except directory choice on first use)
Phase 4: Development (automatic two-stage review replaces manual checkpoint)
Phase 5: Testing (automatic parallel fix)

Benefits

Quality Improvements:

📈 需求理解: 对话探索 + PRD → 减少返工
📈 开发质量: 子任务 + 两阶段审查 → 立即发现错误
📈 代码质量: TDD铁律 → 测试覆盖率提升
📈 测试效率: 并行修复 → 修复速度提升
📈 环境安全: 工作树隔离 → 避免污染

Technical Debt Reduction:

✅ 需求返工: 减少
✅ 代码缺陷: 立即发现
✅ 范围蔓延: 两阶段审查防止
✅ 环境污染: 工作树隔离
✅ 测试维护: 并行加速

Version History

v6.0 (2026-01-19): Integrate Superpowers frameworks
- New: Phase 0 (brainstorming)
- New: Phase 3.5 (git-worktrees)
- Enhanced: Phase 4 (subagent-driven-development + two-stage review)
- Enhanced: Phase 5 (parallel-agent-dispatch)
- Integrated: TDD enforcement
v5.7 (2025-01-18): Current baseline
v5.3 (2025-01-17): Enhanced mandatory checkpoints
v5.0 (2025-01-17): Direct SKILL orchestration