cexll/myclaude
1
0
Fork 0
mirror of https://github.com/cexll/myclaude.git synced 2026-02-05 02:30:26 +08:00
Code Issues Packages Projects Releases Wiki Activity

Update Chinese README and requirements-pilot command to align with latest workflow

Browse Source 
- Updated Chinese README to match English version terminology
- Changed from spec-workflow to requirements-pilot throughout documentation
- Updated quality thresholds from 95% to 90% for consistency
- Replaced Sub-Agent terminology with Requirements-Driven workflow
- Updated agent names from spec-* to requirements-* pattern
- Improved requirements-pilot command structure with clearer phase separation
- Added mandatory user approval gate before implementation phase

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
ben chen
2025-08-04 14:23:25 +08:00
parent 1d45442249
commit 18042ae72e
2 changed files with 130 additions and 195 deletions
Download Patch File Download Diff File Expand all files Collapse all files

124
README-zh.md
View File

@@ -15,8 +15,8 @@
**现在**: 一键自动化专家工作流 **现在**: 一键自动化专家工作流
```bash ```bash
/spec-workflow "实现JWT用户认证系统" /requirements-pilot "实现JWT用户认证系统"
# 30分钟自动执行，95%质量门控，零人工干预 # 30分钟自动执行，90%质量门控，零人工干预
``` ```
## 🎯 核心价值主张 ## 🎯 核心价值主张
@@ -30,31 +30,30 @@
## 📋 两种主要使用模式 ## 📋 两种主要使用模式
### 1. 🏭 Sub-Agent 工作流（自动化专家团队） ### 1. 🏭 Requirements-Driven 工作流（自动化专家团队）
**架构**: 并行专家协调与质量门控 **架构**: 需求导向工作流与质量门控
``` ```
spec-generation → spec-executor → spec-validation → (≥95%?) → spec-testing requirements-generate → requirements-code → requirements-review → (≥90%?) → requirements-testing
↑ ↓ (<95%) ↑ ↓ (<90%)
←←←←←← 自动优化循环，直到质量达标 ←←←←←← ←←←←←← 自动优化循环 ←←←←←←
``` ```
**使用方法**: **使用方法**:
```bash ```bash
# 一条命令完成完整开发工作流 # 一条命令完成完整开发工作流
/spec-workflow "构建用户管理系统，支持RBAC权限控制" /requirements-pilot "构建用户管理系统，支持RBAC权限控制"
# 高级多阶段工作流 # 高级多阶段工作流
先使用 spec-generation，然后 spec-executor，再用 spec-validation， 先使用 requirements-generate，然后 requirements-code，再用 requirements-review，
如果评分 ≥95% 则使用 spec-testing，最后用 optimize 优化 如果评分 ≥90% 则使用 requirements-testing
``` ```
**质量评分体系** (总分100%): **质量评分体系** (总分100%):
- 需求符合度 (30%) - 功能性 (40%)
- 代码质量 (25%) - 集成性 (25%)
- 安全性 (20%) - 代码质量 (20%)
- 性能 (15%) - 性能 (15%)
- 测试覆盖 (10%)
### 2. 🎛️ 自定义命令（手动编排） ### 2. 🎛️ 自定义命令（手动编排）
@@ -95,7 +94,7 @@ spec-generation → spec-executor → spec-validation → (≥95%?) → spec-tes
**完整功能开发**: **完整功能开发**:
```bash ```bash
/spec-workflow "实现OAuth2认证，支持刷新令牌" /requirements-pilot "实现OAuth2认证，支持刷新令牌"
``` ```
**手动开发流程**: **手动开发流程**:
@@ -108,10 +107,11 @@ spec-generation → spec-executor → spec-validation → (≥95%?) → spec-tes
### 3. 预期输出 ### 3. 预期输出
**自动化工作流结果**: **自动化工作流结果**:
- ✅ 完整规格说明 (requirements.md, design.md, tasks.md) - ✅ 需求确认，90+质量评分
- ✅ 生产就绪代码，遵循安全最佳实践 - ✅ 实现就绪技术规格
- ✅ 全面测试套件 (单元 + 集成 + 安全测试) - ✅ 生产就绪代码，遵循最佳实践
- ✅ 95%+ 质量验证评分 - ✅ 全面测试套件 (单元 + 集成 + 功能测试)
- ✅ 90%+ 质量验证评分
## 🏗️ 架构概览 ## 🏗️ 架构概览
@@ -126,10 +126,12 @@ spec-generation → spec-executor → spec-validation → (≥95%?) → spec-tes
- **运维工具**: `/deploy-check`, `/cicd` - **运维工具**: `/deploy-check`, `/cicd`
#### **Agents 目录** (`/agents/`) #### **Agents 目录** (`/agents/`)
- **spec-generation**: 自动化规格说明工作流 - **requirements-generate**: 为代码生成优化的技术规格生成
- **spec-executor**: 带进度跟踪的实现协调器 - **requirements-code**: 最小架构开销的直接实现智能体
- **spec-validation**: 多维度质量评分 (0-100%) - **requirements-review**: 专注功能性和可维护性的实用代码审查
- **spec-testing**: 全面测试策略协调 - **requirements-testing**: 专注功能验证的实用测试智能体
- **bugfix**: 分析和修复软件缺陷的错误解决专家
- **bugfix-verify**: 客观评估的修复验证专家
- **code**: 直接实现的开发协调器 - **code**: 直接实现的开发协调器
- **debug**: UltraThink系统化问题分析 - **debug**: UltraThink系统化问题分析
- **optimize**: 性能优化协调 - **optimize**: 性能优化协调
@@ -137,14 +139,14 @@ spec-generation → spec-executor → spec-validation → (≥95%?) → spec-tes
### 多智能体协调系统 ### 多智能体协调系统
**4个核心专家**: **4个核心专家**:
1. **规格生成专家** - 需求、设计、实现规划 1. **需求生成专家** - 实现就绪的技术规格
2. **实现执行专家** - 带任务跟踪的代码开发 2. **代码实现专家** - 直接、实用的代码实现
3. **质量验证专家** - 多维度评分与可执行反馈 3. **质量审查专家** - 实用质量审查与评分
4. **测试协调专家** - 全面测试策略与执行 4. **测试协调专家** - 功能验证和测试
**关键特性**: **关键特性**:
- **独立上下文**: 专家间无上下文污染 - **独立上下文**: 专家间无上下文污染
- **质量门控**: 95%阈值自动进展判断 - **质量门控**: 90%阈值自动进展判断
- **迭代改进**: 自动优化循环 - **迭代改进**: 自动优化循环
- **可追溯性**: 完整规格 → 代码 → 测试追溯链 - **可追溯性**: 完整规格 → 代码 → 测试追溯链
@@ -154,27 +156,27 @@ spec-generation → spec-executor → spec-validation → (≥95%?) → spec-tes
**输入**: **输入**:
```bash ```bash
/spec-workflow "企业JWT认证系统，支持RBAC，500并发用户，集成现有LDAP" /requirements-pilot "企业JWT认证系统，支持RBAC，500并发用户，集成现有LDAP"
``` ```
**自动化执行过程**: **自动化执行过程**:
1. **第1轮** (质量: 83/100) - 基础实现 1. **需求确认** (质量: 92/100) - 交互式澄清
- 问题: JWT密钥硬编码，缺少密码复杂度验证 - 功能清晰度、技术特定性、实现完整性
- **决策**: <95%，重新开始并改进 - **决策**: ≥90%，进入实现阶段
2. **第2轮** (质量: 91/100) - 安全改进 2. **第1轮** (质量: 83/100) - 基础实现
- 问题: 异常处理不完善，性能未优化 - 问题: 错误处理不完整，集成问题
- **决策**: <95%，继续优化 - **决策**: <90%，重新开始并改进
3. **第3轮** (质量: 97/100) - 生产就绪 3. **第2轮** (质量: 93/100) - 生产就绪
- **决策**: ≥95%，进入全面测试阶段 - **决策**: ≥90%，进入功能测试
**最终交付物**: **最终交付物**:
- 完整EARS格式需求文档 - 质量评估的需求确认
- 安全加固的JWT实现 - 实现就绪技术规格
- 带角色层级的RBAC - 带RBAC的实用JWT实现
- 带错误处理的LDAP集成 - 带正确错误处理的LDAP集成
- 全面测试套件（单元 + 集成 + 安全） - 专注关键路径的功能测试套件
### API网关开发 ### API网关开发
@@ -224,10 +226,10 @@ spec-generation → spec-executor → spec-validation → (≥95%?) → spec-tes
## 🎯 效益与影响 ## 🎯 效益与影响
| 维度 | 传统Slash命令 | Sub-Agent自动工作流 | | 维度 | 手动命令 | Requirements-Driven工作流 |
|------|-------------|------------------| |------|-------------|------------------|
| **复杂度** | 手动触发每个步骤 | 一键启动完整流水线 | | **复杂度** | 手动触发每个步骤 | 一键启动完整流水线 |
| **质量** | 主观评估 | 95%客观评分 | | **质量** | 主观评估 | 90%客观评分 |
| **上下文** | 污染，需要/clear | 隔离，无污染 | | **上下文** | 污染，需要/clear | 隔离，无污染 |
| **专业性** | AI角色切换 | 专注的专家 | | **专业性** | AI角色切换 | 专注的专家 |
| **错误处理** | 手动发现/修复 | 自动优化 | | **错误处理** | 手动发现/修复 | 自动优化 |
@@ -239,7 +241,7 @@ spec-generation → spec-executor → spec-validation → (≥95%?) → spec-tes
每个智能体在独立上下文中专注各自领域专业知识，避免角色切换导致的质量下降。 每个智能体在独立上下文中专注各自领域专业知识，避免角色切换导致的质量下降。
### 2. **智能质量门控** ### 2. **智能质量门控**
95%客观评分，自动决策工作流进展或优化循环。 90%客观评分，自动决策工作流进展或优化循环。
### 3. **完全自动化** ### 3. **完全自动化**
一条命令触发端到端开发工作流，最少人工干预。 一条命令触发端到端开发工作流，最少人工干预。
@@ -267,9 +269,9 @@ spec-generation → spec-executor → spec-validation → (≥95%?) → spec-tes
## 📖 命令参考 ## 📖 命令参考
### 规格说明工作流 ### Requirements 工作流
- `/spec` - 交互式需求 → 设计 → 任务 - `/requirements-pilot` - 完整的需求驱动开发工作流
- `/spec-workflow` - 自动化端到端规格 + 实现 - 交互式需求确认 → 技术规格 → 实现 → 测试
### 开发命令 ### 开发命令
- `/ask` - 架构咨询（不修改代码） - `/ask` - 架构咨询（不修改代码）
@@ -309,12 +311,12 @@ MIT许可证 - 详见[LICENSE](LICENSE)文件。
准备好转换你的开发工作流了吗？从这里开始： 准备好转换你的开发工作流了吗？从这里开始：
```bash ```bash
/spec-workflow "在这里描述你的第一个功能" /requirements-pilot "在这里描述你的第一个功能"
``` ```
看着你的一行请求变成完整、经过测试、生产就绪的实现，95%质量保证。 看着你的一行请求变成完整、经过测试、生产就绪的实现，90%质量保证。
**记住**: 好的软件来自好的流程，好的流程来自专业的团队。Sub-agents让你拥有一个永不疲倦、始终专业的虚拟开发团队。 **记住**: 专业软件来自专业流程。需求驱动工作流为您提供一个永不疲倦、始终专业的虚拟开发团队。
*让专业的AI做专业的事 - 开发从此变得优雅而高效。* *让专业的AI做专业的事 - 开发从此变得优雅而高效。*
@@ -335,9 +337,9 @@ MIT许可证 - 详见[LICENSE](LICENSE)文件。
5. /optimize 性能优化 → 手动优化 5. /optimize 性能优化 → 手动优化
``` ```
**Sub-Agents方式** (30分钟自动): **Requirements-Driven方式** (30分钟自动):
```bash ```bash
/spec-workflow "企业用户管理系统，500人规模，RBAC权限，OA系统集成" /requirements-pilot "企业用户管理系统，500人规模，RBAC权限，OA系统集成"
``` ```
**自动化执行结果**: **自动化执行结果**:
@@ -401,13 +403,13 @@ AI会深入询问：
```bash ```bash
# 第一阶段：基础功能 # 第一阶段：基础功能
/spec-workflow "用户注册登录基础功能" /requirements-pilot "用户注册登录基础功能"
# 第二阶段：权限管理 # 第二阶段：权限管理
/spec-workflow "在现有基础上添加RBAC权限系统" /requirements-pilot "在现有基础上添加RBAC权限系统"
# 第三阶段：系统集成 # 第三阶段：系统集成
/spec-workflow "集成LDAP和SSO单点登录" /requirements-pilot "集成LDAP和SSO单点登录"
``` ```
### 3. 质量优先策略 ### 3. 质量优先策略
@@ -439,7 +441,7 @@ AI扮演产品经理 → 架构师 → 开发者 → 测试工程师 → 优化
每个决策点都可能中断，需要重新组织思路 每个决策点都可能中断，需要重新组织思路
``` ```
### Sub-Agents解决方案 ### Requirements-Driven解决方案
**专业化隔离**: 每个专家在独立上下文中工作 **专业化隔离**: 每个专家在独立上下文中工作
``` ```
@@ -449,18 +451,18 @@ AI扮演产品经理 → 架构师 → 开发者 → 测试工程师 → 优化
**自动化决策**: 基于客观指标的自动流程控制 **自动化决策**: 基于客观指标的自动流程控制
``` ```
质量评分 ≥95% → 自动进入下一阶段 质量评分 ≥90% → 自动进入下一阶段
质量评分 <95% → 自动返回优化，无需人工判断 质量评分 <90% → 自动返回优化，无需人工判断
``` ```
## 🚀 开始你的AI工厂 ## 🚀 开始你的AI工厂
从手工作坊升级到自动化工厂，只需要： 从手工作坊升级到自动化工厂，只需要：
1. **配置一次**: 设置Sub-Agents和自定义命令 1. **配置一次**: 设置Requirements-Driven智能体和自定义命令
2. **使用一生**: 每个项目都能享受专业AI团队服务 2. **使用一生**: 每个项目都能享受专业AI团队服务
3. **持续改进**: 工作流模式不断优化，开发效率持续提升 3. **持续改进**: 工作流模式不断优化，开发效率持续提升
**记住**: 在AI时代，Sub-Agents让你拥有了一个永不疲倦、始终专业的虚拟开发团队。 **记住**: 在AI时代，Requirements-Driven工作流让你拥有了一个永不疲倦、始终专业的虚拟开发团队。
*让专业的AI做专业的事，开发从此变得优雅而高效。* *让专业的AI做专业的事，开发从此变得优雅而高效。*

201
commands/requirements-pilot.md
View File

@@ -1,16 +1,6 @@
## Usage ## Usage
`/requirements-pilot <FEATURE_DESCRIPTION>` `/requirements-pilot <FEATURE_DESCRIPTION>`
**⚡ IMMEDIATE EXECUTION REQUIRED ⚡**
You must immediately start the requirements confirmation process for: [$ARGUMENTS]
**DO THIS NOW:**
1. Extract feature name from [$ARGUMENTS]
2. Assess requirement quality (100-point system)
3. Generate clarification questions for unclear aspects
4. Start interactive clarification with user
## Context ## Context
- Feature to develop: $ARGUMENTS - Feature to develop: $ARGUMENTS
- Pragmatic development workflow optimized for code generation - Pragmatic development workflow optimized for code generation
@@ -22,69 +12,59 @@ You are the Requirements-Driven Workflow Orchestrator managing a streamlined dev
You adhere to core software engineering principles like KISS (Keep It Simple, Stupid), YAGNI (You Ain't Gonna Need It), and SOLID to ensure implementations are robust, maintainable, and pragmatic. You adhere to core software engineering principles like KISS (Keep It Simple, Stupid), YAGNI (You Ain't Gonna Need It), and SOLID to ensure implementations are robust, maintainable, and pragmatic.
**EXECUTE IMMEDIATELY**: Upon receiving this command, immediately begin Phase 1 Requirements Confirmation for the feature: [$ARGUMENTS] ## Workflow Overview
## Automated Workflow Execution ### Phase 1: Requirements Confirmation (Starts Automatically)
Upon receiving this command, immediately begin the requirements confirmation process for: [$ARGUMENTS]
**START NOW**: Begin requirements confirmation immediately for the feature: [$ARGUMENTS] ### 🛑 CRITICAL STOP POINT: User Approval Gate 🛑
**IMPORTANT**: After achieving 90+ quality score, you MUST STOP and wait for explicit user approval before proceeding to Phase 2.
**CRITICAL**: You must start the confirmation process right now. Do not wait or ask what to do next. ### Phase 2: Implementation (Only After Approval)
Execute the sub-agent chain ONLY after the user explicitly confirms they want to proceed.
### Execution Steps: ## Phase 1: Requirements Confirmation Process
1. **Extract feature name** from [$ARGUMENTS] (convert to kebab-case)
2. **Create directory**: `./.claude/specs/{feature_name}/`
3. **Assess requirement quality** using 100-point scoring system
4. **Generate clarification questions** for any unclear aspects
5. **Interact with user** until quality score ≥ 90 points
6. **Save confirmation** to `./.claude/specs/{feature_name}/requirements-confirm.md`
7. **Present summary** to user and ask for explicit approval to proceed
8. **Execute sub-agent chain** only after user confirms to proceed
Execute the following workflow using Claude Code's sub-agent syntax: Start this phase immediately upon receiving the command:
``` ### 1. Input Validation & Length Handling
After requirements confirmation (≥90 quality score) AND explicit user approval to proceed, first use the requirements-generate sub agent to create implementation-ready technical specifications for confirmed requirements, then use the requirements-code sub agent to implement the functionality based on specifications, then use the requirements-review sub agent to evaluate code quality with practical scoring, then if score ≥90% use the requirements-testing sub agent to create functional test suite, otherwise use the requirements-code sub agent again to address review feedback and repeat the review cycle. - **If input > 500 characters**: First summarize the core functionality and ask user to confirm the summary is accurate
``` - **If input is unclear or too brief**: Request more specific details before proceeding
## Requirements Confirmation Process ### 2. Feature Name Generation & Setup
- Extract feature name from [$ARGUMENTS] using kebab-case format
- Create directory: `./.claude/specs/{feature_name}/`
- Initialize confirmation tracking
**IMMEDIATE ACTION REQUIRED**: Start Phase 1 requirements confirmation now for: [$ARGUMENTS] ### 3. Requirements Quality Assessment (100-point system)
- **Functional Clarity (30 points)**: Clear input/output specs, user interactions, success criteria
- **Technical Specificity (25 points)**: Integration points, technology constraints, performance requirements
- **Implementation Completeness (25 points)**: Edge cases, error handling, data validation
- **Business Context (20 points)**: User value proposition, priority definition
**BEGIN CONFIRMATION NOW**: ### 4. Interactive Clarification Loop
1. Analyze the feature description: [$ARGUMENTS] - **Quality Gate**: Continue until score ≥ 90 points (no iteration limit)
2. Generate quality assessment questions immediately - Generate targeted clarification questions for missing areas
3. Start interactive clarification with the user - Document confirmation process and save to `./.claude/specs/{feature_name}/requirements-confirm.md`
- Include: original request, clarification rounds, quality scores, final confirmed requirements
**Phase 1: Requirements Analysis & Confirmation (Orchestrator Level)** ## 🛑 User Approval Gate (Mandatory Stop Point) 🛑
Before executing the sub-agent chain, perform requirements confirmation: **CRITICAL: You MUST stop here and wait for user approval**
1. **Feature Name Generation & Setup** After achieving 90+ quality score:
- Extract feature name from [$ARGUMENTS] using kebab-case format 1. Present final requirements summary with quality score
- Create directory: `./.claude/specs/{feature_name}/` 2. Display the confirmed requirements clearly
- Initialize confirmation tracking 3. Ask explicitly: **"Requirements are now clear (90+ points). Do you want to proceed with implementation? (Reply 'yes' to continue or 'no' to refine further)"**
4. **WAIT for user response**
5. **Only proceed if user responds with**: "yes", "确认", "proceed", "continue", or similar affirmative response
6. **If user says no or requests changes**: Return to clarification phase
2. **Requirements Quality Assessment (100-point system)** ## Phase 2: Implementation Process (After Approval Only)
- **Functional Clarity (30 points)**: Clear input/output specs, user interactions, success criteria
- **Technical Specificity (25 points)**: Integration points, technology constraints, performance requirements
- **Implementation Completeness (25 points)**: Edge cases, error handling, data validation
- **Business Context (20 points)**: User value proposition, priority definition
3. **Interactive Clarification Loop** **ONLY execute this phase after receiving explicit user approval**
- **Quality Gate**: Continue until score ≥ 90 points (no iteration limit)
- Generate targeted clarification questions for missing areas
- Document confirmation process and save to `./.claude/specs/{feature_name}/requirements-confirm.md`
- Include: original request, clarification rounds, quality scores, final confirmed requirements
4. **User Approval Gate** Execute the following sub-agent chain:
- Present final requirements summary with quality score
- Ask user: "Requirements are now clear (90+ points). Do you want to proceed with implementation?"
- Wait for explicit user confirmation before starting sub-agent chain
- Only proceed to Phase 2 after user approves
**Phase 2: Sub-Agent Chain Process**
After requirements confirmation (≥90 quality score), execute the following chain:
``` ```
First use the requirements-generate sub agent to create implementation-ready technical specifications for confirmed requirements, then use the requirements-code sub agent to implement the functionality based on specifications, then use the requirements-review sub agent to evaluate code quality with practical scoring, then if score ≥90% use the requirements-testing sub agent to create functional test suite, otherwise use the requirements-code sub agent again to address review feedback and repeat the review cycle. First use the requirements-generate sub agent to create implementation-ready technical specifications for confirmed requirements, then use the requirements-code sub agent to implement the functionality based on specifications, then use the requirements-review sub agent to evaluate code quality with practical scoring, then if score ≥90% use the requirements-testing sub agent to create functional test suite, otherwise use the requirements-code sub agent again to address review feedback and repeat the review cycle.
@@ -94,47 +74,36 @@ First use the requirements-generate sub agent to create implementation-ready tec
## Workflow Logic ## Workflow Logic
### Phase Transitions
1. **Start → Phase 1**: Automatic upon command receipt
2. **Phase 1 → Approval Gate**: Automatic when quality ≥ 90 points
3. **Approval Gate → Phase 2**: ONLY with explicit user confirmation
4. **Approval Gate → Phase 1**: If user requests refinement
### Requirements Quality Gate ### Requirements Quality Gate
- **Requirements Score ≥90 points**: Present summary and request user approval - **Requirements Score ≥90 points**: Move to approval gate
- **Requirements Score <90 points**: Continue interactive clarification - **Requirements Score <90 points**: Continue interactive clarification
- **No iteration limit**: Quality-driven approach ensures requirement clarity - **No iteration limit**: Quality-driven approach ensures requirement clarity
### User Approval Gate ### Code Quality Gate (Phase 2 Only)
- **After 90+ quality score**: Present requirements summary to user
- **User Approval Required**: Ask "Do you want to proceed with implementation?"
- **Only after user confirms**: Execute sub-agent chain
- **User can decline**: Return to clarification or end workflow
### Code Quality Gate Mechanism
- **Review Score ≥90%**: Proceed to requirements-testing sub agent - **Review Score ≥90%**: Proceed to requirements-testing sub agent
- **Review Score <90%**: Loop back to requirements-code sub agent with feedback - **Review Score <90%**: Loop back to requirements-code sub agent with feedback
- **Maximum 3 iterations**: Prevent infinite loops while ensuring quality - **Maximum 3 iterations**: Prevent infinite loops while ensuring quality
### Complete Workflow Execution Steps ## Execution Flow Summary
1. **Requirements Confirmation (Orchestrator)**: Interactive clarification until 90+ point quality score
2. **User Approval Gate**: Present summary and wait for explicit user confirmation to proceed
3. **requirements-generate sub agent**: Create technical specification from confirmed requirements (only after user approval)
4. **requirements-code sub agent**: Implement code based on technical specification
5. **requirements-review sub agent**: Practical quality review with scoring (0-100%)
6. **Code Quality Gate Decision**:
- If ≥90%: Continue to requirements-testing sub agent
- If <90%: Return to requirements-code sub agent with specific feedback
7. **requirements-testing sub agent**: Create functional test suite (final step)
## Expected Workflow Phases ```
1. Receive command
2. Validate input length (summarize if >500 chars)
3. Start requirements confirmation (Phase 1)
4. Iterate until 90+ quality score
5. 🛑 STOP and request user approval
6. Wait for user response
7. If approved: Execute implementation (Phase 2)
8. If not approved: Return to clarification
```
### Requirements Phase ## Key Workflow Characteristics
- **Clarification Rounds**: Variable based on requirement complexity (until 90+ points)
- **Quality Assessment**: Comprehensive evaluation across 4 dimensions
- **Requirements Summary**: Present final requirements with quality score
- **User Approval**: Explicit confirmation required before implementation phase
### Implementation Phase
- **Round 1**: Initial implementation (typically 75-85% quality)
- **Round 2**: Refined implementation addressing feedback (typically 85-95%)
- **Round 3**: Final optimization if needed (90%+ target)
## Key Workflow Differences from SPEC
### Implementation-First Approach ### Implementation-First Approach
- **Direct Technical Specs**: Skip architectural abstractions, focus on concrete implementation details - **Direct Technical Specs**: Skip architectural abstractions, focus on concrete implementation details
@@ -148,62 +117,26 @@ First use the requirements-generate sub agent to create implementation-ready tec
- **Maintainability**: Code that's easy to understand and modify - **Maintainability**: Code that's easy to understand and modify
- **Performance Adequacy**: Reasonable performance for the use case, not theoretical optimization - **Performance Adequacy**: Reasonable performance for the use case, not theoretical optimization
### Streamlined Testing
- **Risk-Based Testing**: Focus on critical paths and likely failure points
- **Functional Validation**: Ensure implemented features work as specified
- **Integration Testing**: Verify components work together correctly
- **Practical Coverage**: Adequate coverage without obsessing over 100%
## Output Format ## Output Format
1. **Requirements Confirmation** - Interactive clarification with quality scoring 1. **Requirements Confirmation** - Interactive clarification with quality scoring
2. **Documentation Creation** - Save confirmation process and requirements 2. **Documentation Creation** - Save confirmation process and requirements
3. **Requirements Summary** - Present final requirements and quality score to user 3. **Requirements Summary** - Present final requirements and quality score to user
4. **User Approval Request** - Ask explicit permission to proceed with implementation 4. **🛑 User Approval Request** - Ask explicit permission to proceed with implementation
5. **Sub-Agent Chain Initiation** - Execute sub-agents only after user approval 5. **Sub-Agent Chain Initiation** - Execute sub-agents only after user approval
6. **Progress Tracking** - Monitor each sub-agent completion and decisions 6. **Progress Tracking** - Monitor each sub-agent completion and decisions
7. **Quality Gate Decisions** - Report review scores and iteration actions 7. **Quality Gate Decisions** - Report review scores and iteration actions
8. **Completion Summary** - Final artifacts and practical quality metrics 8. **Completion Summary** - Final artifacts and practical quality metrics
## Key Benefits
- **Implementation-Focused**: Optimized for direct code generation rather than human review
- **Practical Quality Control**: 90% threshold ensures good quality without perfectionism
- **Faster Delivery**: Streamlined workflow reduces development time
- **Code-Generation Friendly**: Specifications designed for automatic implementation
- **Maintainable Solutions**: Focus on long-term maintainability over architectural purity
## Success Criteria ## Success Criteria
- **Clear Requirements**: 90+ quality score before implementation
- **User Control**: Implementation only begins with explicit approval
- **Working Implementation**: Code fully implements specified functionality - **Working Implementation**: Code fully implements specified functionality
- **Quality Assurance**: 90%+ quality score indicates production-ready code - **Quality Assurance**: 90%+ quality score indicates production-ready code
- **Integration Success**: New code integrates seamlessly with existing systems - **Integration Success**: New code integrates seamlessly with existing systems
- **Test Coverage**: Adequate test coverage for reliability and maintenance
- **Development Velocity**: Efficient workflow that doesn't slow down development
## Execution Instructions ## Important Reminders
- **Phase 1 starts automatically** - No waiting needed for requirements confirmation
**IMPORTANT**: You must immediately begin execution upon receiving a feature description. Do not wait for additional instructions. - **Phase 2 requires explicit approval** - Never skip the approval gate
- **Long inputs need summarization** - Handle >500 character inputs specially
### Step 1: Start Requirements Confirmation - **User can always decline** - Respect user's decision to refine or cancel
1. **Extract feature name** from [$ARGUMENTS] and convert to kebab-case - **Quality over speed** - Ensure clarity before implementation
2. **Create specs directory**: `./.claude/specs/{feature_name}/`
3. **Begin quality assessment** of the provided feature description
4. **Generate clarification questions** for any unclear aspects
5. **Present questions to user** and await responses
6. **Iterate until quality score ≥ 90 points**
7. **Save confirmation results** to `./.claude/specs/{feature_name}/requirements-confirm.md`
### Step 2: Execute Sub-Agent Chain
Once requirements are confirmed (≥90 points), immediately execute:
```
First use the requirements-generate sub agent to create implementation-ready technical specifications for confirmed requirements, then use the requirements-code sub agent to implement the functionality based on specifications, then use the requirements-review sub agent to evaluate code quality with practical scoring, then if score ≥90% use the requirements-testing sub agent to create functional test suite, otherwise use the requirements-code sub agent again to address review feedback and repeat the review cycle.
```
## Workflow Summary
Simply provide the feature description and the workflow will:
1. **First** ensure requirement clarity through interactive confirmation (90+ point quality gate)
2. **Second** present summary and request explicit user approval
3. **Then** execute the requirements-driven sub-agent chain for complete development (only after approval)
This three-step approach (clarify → confirm → implement) prevents ambiguous requirements from entering the development pipeline and ensures user control over when implementation begins.
**START IMMEDIATELY**: Begin Phase 1 requirements confirmation as soon as you receive the feature description [$ARGUMENTS].