Build工作流编排器：AI驱动的开发流程自动化工具，支持绿地/棕地/TDD/重构 | SkillsMD

Build工作流编排器：AI驱动的开发流程自动化工具，支持绿地/棕地/TDD/重构

build by parcadei/continuous-claude-v3

204 周安装量

3,600 GitHub Stars

GitHub

安装命令

npx skills add https://github.com/parcadei/continuous-claude-v3 --skill build

开发项目管理自动化

🇨🇳中文介绍

Build - 工作流编排器

你是一个工作流编排器，负责串联现有技能以进行功能开发。你协调多个技能的顺序执行，在它们之间传递交接，并在阶段边界暂停以进行人工检查点。

调用方式

/build <mode> [options] [description]

问题流程（无参数）

如果用户只输入 /build 而没有参数或参数不完整，请通过此问题流程引导他们以推断正确的配置。每个阶段使用 AskUserQuestion。

阶段 0：工作流选择

question: "你想做什么？"
header: "工作流"
options:
  - label: "帮我选择（推荐）"
    description: "我会问几个问题来挑选合适的工作流"
  - label: "绿地开发 - 新功能"
    description: "链式流程：发现 → 规划 → 验证 → 实施 → 提交 → PR"
  - label: "棕地开发 - 现有代码"
    description: "链式流程：熟悉代码库 → 研究 → 规划 → 验证 → 实施"
  - label: "TDD - 测试先行"
    description: "链式流程：规划 → 测试驱动开发 → 实施"
  - label: "重构 - 改进结构"
    description: "链式流程：影响分析 → 规划 → TDD → 实施"

映射关系：

"帮我选择" → 继续第 1-4 阶段的问题
"绿地开发" → 设置 mode=greenfield，跳转到阶段 5（描述）
"棕地开发" → 设置 mode=brownfield，跳转到阶段 5（描述）
"TDD" → 设置 mode=tdd，跳转到阶段 5（描述）
"重构" → 设置 mode=refactor，跳转到阶段 5（描述）

如果答案不明确（通过"其他"选项）：

广告位招租

在这里展示您的产品或服务

触达数万 AI 开发者，精准高效

联系我们

相关 Skills

FlyClaw：零登录航班聚合查询工具，Python实现多源航班信息与价格搜索

4,000,000 周安装

find-skills 技能搜索工具 - Vercel Labs 开源智能体技能包管理器

827,200 周安装

Vercel React 最佳实践指南 | 58条Next.js性能优化规则与代码重构

273,700 周安装

Vercel Web界面规范检查工具 - 自动检测代码是否符合Web设计指南

221,200 周安装

question: "我想了解你的工作流需求。你的意思是..."
header: "澄清"
options:
  - label: "帮我选择"
    description: "不确定哪个工作流 - 请通过问题引导我"
  - label: "绿地开发 - 新功能"
    description: "在没有现有代码的情况下构建新功能"
  - label: "棕地开发 - 现有代码"
    description: "在现有代码库中添加或修改功能"
  - label: "都不是 - 让我换个方式解释"
    description: "我会描述我想要做什么"

question: "这是一个新功能还是对现有代码的修改？"
header: "背景"
options:
  - label: "从头开始的新功能"
    description: "没有需要集成的现有代码"
  - label: "添加到现有代码库"
    description: "需要先理解当前代码"
  - label: "重构现有代码"
    description: "在不改变行为的情况下进行改进"

question: "我想确认我理解了你的意思。你的意思是..."
header: "澄清"
options:
  - label: "从头开始的新功能"
    description: "在没有现有代码的情况下构建新功能"
  - label: "添加到现有代码库"
    description: "与已存在的代码集成"
  - label: "重构现有代码"
    description: "在不改变行为的情况下改进结构"
  - label: "都不是 - 让我换个方式解释"
    description: "我会提供更多细节"

question: "你的需求有多明确？"
header: "需求"
options:
  - label: "我有明确的规格/描述"
    description: "确切知道要构建什么"
  - label: "我有个大概的想法"
    description: "需要帮助完善细节"
  - label: "只是在探索可能性"
    description: "想了解有哪些可能性"

question: "我想确认我理解了你的需求状态。你的意思是..."
header: "澄清"
options:
  - label: "我有明确的规格/描述"
    description: "准备实施 - 不需要发现阶段"
  - label: "我有个大概的想法"
    description: "需要一些帮助来定义细节"
  - label: "只是在探索可能性"
    description: "还不确切知道什么是可能的"
  - label: "都不是 - 让我换个方式解释"
    description: "我会更好地描述我的情况"

question: "我应该如何着手开发？"
header: "方法"
options:
  - label: "直接实施"
    description: "标准实施流程"
  - label: "先写测试（TDD）"
    description: "测试驱动开发"
  - label: "编码前验证计划"
    description: "在实施前审查计划"

question: "我想确认我理解了你偏好的方法。你的意思是..."
header: "澄清"
options:
  - label: "直接实施"
    description: "标准开发 - 先实施后测试"
  - label: "先写测试（TDD）"
    description: "测试驱动 - 先测试后实施"
  - label: "编码前验证计划"
    description: "首先使用 validate-agent 审查计划"
  - label: "都不是 - 让我换个方式解释"
    description: "我心中有不同的工作流"

question: "实施后应该做什么？"
header: "完成"
multiSelect: true
options:
  - label: "自动提交更改"
    description: "完成后创建 git 提交"
  - label: "创建 PR 描述"
    description: "生成 PR 摘要"
  - label: "只保留文件更改"
    description: "我自己处理 git"

question: "我想了解你在实施后需要什么。哪些适用？"
header: "澄清"
multiSelect: true
options:
  - label: "自动提交更改"
    description: "我将用你的更改创建一个 git 提交"
  - label: "创建 PR 描述"
    description: "我将为你生成一个 PR 摘要"
  - label: "只保留文件更改"
    description: "不进行 git 操作 - 你自己处理"
  - label: "都不是 - 让我换个方式解释"
    description: "我有不同的实施后需求"

question: "描述你想要构建什么（1-2 句话）："
header: "功能"
options: []  # 通过"其他"选项进行自由文本输入

Based on your answers, I'll run:

**Mode:** brownfield
**Chain:** onboard → research-codebase → plan-agent → implement_plan
**Options:** --skip-commit
**Description:** "Add user authentication with OAuth"

Proceed? [Yes / Adjust settings]

模式	链式流程	使用场景
`greenfield`	discovery-interview -> plan-agent -> validate-agent -> implement_plan -> commit -> describe_pr	从头开始的新功能
`brownfield`	onboard -> research-codebase -> plan-agent -> validate-agent -> implement_plan	现有代码库中的功能
`tdd`	plan-agent -> test-driven-development -> implement_plan	测试先行的实施
`refactor`	tldr-code (impact) -> plan-agent -> test-driven-development -> implement_plan	带有影响分析的安全重构

选项	效果
`--skip-discovery`	跳过访谈阶段（使用现有规格或描述）
`--skip-validate`	跳过验证阶段（信任现有计划）
`--skip-commit`	实施后不自动提交
`--skip-pr`	不创建 PR 描述
`--parallel`	并行运行独立的研究代理

/build greenfield --skip-validate Add user authentication
       ^mode      ^options        ^description

CHAINS = {
    "greenfield": ["discovery-interview", "plan-agent", "validate-agent", "implement_plan", "commit", "describe_pr"],
    "brownfield": ["onboard", "research-codebase", "plan-agent", "validate-agent", "implement_plan"],
    "tdd": ["plan-agent", "test-driven-development", "implement_plan"],
    "refactor": ["tldr-impact", "plan-agent", "test-driven-development", "implement_plan"]
}

创建交接目录：

SESSION="build-$(date +%Y%m%d)-<kebab-description>"
mkdir -p "thoughts/shared/handoffs/$SESSION"

创建编排状态文件：

cat > "thoughts/shared/handoffs/$SESSION/orchestration.yaml" << EOF
session: $SESSION
mode: <mode>
options: [<options>]
description: "<description>"
started: $(date -u +"%Y-%m-%dT%H:%M:%SZ")
chain: [<skill1>, <skill2>, ...]
current_phase: 0
phases:
  - skill: <skill1>
    status: pending
  - skill: <skill2>
    status: pending
  ...
EOF

1. Read previous handoff (if exists)
2. Execute skill (spawn agent or invoke directly)
3. Capture skill output/handoff
4. Update orchestration state
5. Human checkpoint (if phase boundary)
6. Continue or handle error

Task(
  subagent_type="discovery-interview",
  prompt="""
  [Contents of discovery-interview SKILL.md]

  ---

  ## Context
  Feature request: <description>
  Handoff directory: thoughts/shared/handoffs/<session>/

  Conduct the interview and create spec.
  """
)

Task(
  subagent_type="onboard",
  prompt="""
  [Contents of onboard SKILL.md]

  ---

  Analyze this codebase and create continuity ledger.
  Handoff directory: thoughts/shared/handoffs/<session>/
  """
)

Task(
  subagent_type="research-codebase",
  prompt="""
  [Contents of research-codebase SKILL.md]

  ---

  Research question: How should we implement <description>?
  Focus areas: [based on spec or description]
  Handoff directory: thoughts/shared/handoffs/<session>/
  """
)

# Run impact analysis on the function/module being refactored
tldr impact <target> src/ --depth 3 > thoughts/shared/handoffs/<session>/impact-analysis.json

# Also run architecture analysis
tldr arch src/ > thoughts/shared/handoffs/<session>/architecture.json

Task(
  subagent_type="plan-agent",
  prompt="""
  [Contents of plan-agent SKILL.md]

  ---

  ## Context
  Feature request: <description>

  [Include spec if exists from discovery-interview]
  [Include research findings if exists]
  [Include impact analysis if refactor mode]

  Handoff directory: thoughts/shared/handoffs/<session>/
  """
)

Plan created: thoughts/shared/plans/PLAN-<name>.md

Please review the plan. Options:
1. Approve and continue to [next phase]
2. Request changes to plan
3. Abort workflow

[Show plan summary]

Task(
  subagent_type="validate-agent",
  prompt="""
  [Contents of validate-agent SKILL.md]

  ---

  Plan to validate: [Plan content]
  Plan path: thoughts/shared/plans/PLAN-<name>.md
  Handoff directory: thoughts/shared/handoffs/<session>/
  """
)

Validation complete with issues:
- [Issue 1]
- [Issue 2]

Options:
1. Proceed anyway (acknowledge risks)
2. Update plan and re-validate
3. Abort workflow

Present TDD guidance to user:

"Entering TDD mode. For each feature:
1. Write failing test first
2. Implement minimal code to pass
3. Refactor while keeping tests green

I'll guide you through each cycle. Starting with first test..."

# Check plan size
if task_count <= 3:
    # Direct implementation
    Follow implement_plan skill directly
else:
    # Agent orchestration mode
    For each task:
        Task(
          subagent_type="implement_task",
          prompt="""
          [Contents of implement_task SKILL.md]

          ---

          Plan: [Plan content]
          Your task: Task N of M: <task description>
          Previous handoff: [Previous task handoff or "First task"]
          Handoff directory: thoughts/shared/handoffs/<session>/
          """
        )

Phase [N] Complete

Automated verification:
- [x] Tests passing
- [x] Type check passed
- [ ] Manual testing required

Please verify:
- [Manual test items from plan]

Continue to next phase? [Y/n]

Follow commit skill:
1. Show git status and diff
2. Present commit plan
3. Execute on user approval
4. Generate reasoning file

Follow describe_pr skill:
1. Create PR if not exists
2. Generate description from changes
3. Update PR with description

# Update orchestration.yaml
phases:
  - skill: plan-agent
    status: complete
  - skill: validate-agent
    status: blocked
    error: "Validation found deprecated library"
    blocker: "Need to replace X with Y"

Workflow blocked at: validate-agent

Issue: Validation found deprecated library
Blocker: Need to replace X with Y

Options:
1. Retry this phase
2. Skip this phase (not recommended)
3. Abort workflow
4. Manual intervention (I'll help you fix it)

Build workflow complete!

Session: thoughts/shared/handoffs/<session>/

Artifacts created:
- Spec: thoughts/shared/specs/<name>-spec.md (if greenfield)
- Plan: thoughts/shared/plans/PLAN-<name>.md
- Validation: <session>/validation-<name>.md
- Implementation handoffs: <session>/task-*.md
- PR: #<number> (if --skip-pr not set)

Commit: <hash> (if --skip-commit not set)

Total phases: N completed, M skipped

阶段之后	检查点目的
discovery-interview	验证规格是否捕获了需求
plan-agent	批准实施计划
validate-agent (if issues)	确认验证发现的问题
Each implement task	在继续之前验证阶段工作正常
commit	批准提交信息和文件

/build resume thoughts/shared/handoffs/<session>/

User: /build greenfield Add user authentication with OAuth

Claude: Starting greenfield workflow for "Add user authentication with OAuth"

Creating session: build-20260108-user-auth-oauth
Chain: discovery-interview -> plan-agent -> validate-agent -> implement_plan -> commit -> describe_pr

Phase 1/6: Discovery Interview
[Spawns discovery-interview agent]

Interview questions:
1. What OAuth providers do you need? (Google, GitHub, etc.)
2. What user data should we store after auth?
3. Should sessions persist across browser closes?
...

[After interview completes]

Spec created: thoughts/shared/specs/2026-01-08-user-auth-spec.md

Phase 2/6: Planning
[Spawns plan-agent]

Plan created: thoughts/shared/plans/PLAN-user-auth.md

--- CHECKPOINT ---
Plan Summary:
- Task 1: Set up OAuth provider configuration
- Task 2: Create user model and session storage
- Task 3: Implement OAuth callback endpoints
- Task 4: Add auth middleware
- Task 5: Create login/logout routes

Approve plan and continue? [Y/n]

User: Y

Phase 3/6: Validation
[Spawns validate-agent]

Validation: PASSED
All tech choices are current best practices.

Phase 4/6: Implementation (5 tasks)
[Spawns implementation agents for each task]

Task 1/5: OAuth configuration... Complete
Task 2/5: User model... Complete
Task 3/5: OAuth callbacks... Complete
Task 4/5: Auth middleware... Complete
Task 5/5: Login/logout routes... Complete

--- CHECKPOINT ---
All implementation complete.
Tests: 24 passing
Manual verification needed:
- [ ] Test OAuth flow end-to-end
- [ ] Verify session persistence

Continue to commit? [Y/n]

User: Y

Phase 5/6: Commit
[Shows changes, creates commit]

Phase 6/6: PR Description
[Creates/updates PR]

Build complete!
PR #42: https://github.com/user/repo/pull/42

User: /build tdd Add rate limiting to API endpoints

Claude: Starting TDD workflow for "Add rate limiting to API endpoints"

Phase 1/3: Planning
[Creates plan with test specifications]

--- CHECKPOINT ---
Plan approved? [Y/n]

User: Y

Phase 2/3: TDD Implementation

Let's start with the first test. According to the plan, we need to test:
"Rate limiter returns 429 after 100 requests per minute"

RED - Write failing test:
```python
def test_rate_limiter_returns_429_after_limit():
    limiter = RateLimiter(limit=100, window=60)
    for _ in range(100):
        assert limiter.check("user1") == True
    assert limiter.check("user1") == False

### Refactor Mode

## Error Recovery

### If a phase fails:

### If implementation is blocked:

## Parallel Execution

With `--parallel` option, independent phases run concurrently:

Only truly independent phases run in parallel. Dependencies are respected.

## Configuration

### Default mode preferences

Set in `.claude/settings.json`:

```json
{
  "skills": {
    "build": {
      "default_mode": "brownfield",
      "always_validate": true,
      "auto_commit": false,
      "checkpoint_phases": ["plan-agent", "implement_plan"]
    }
  }
}

问题	解决方案
"未找到连续性记录"	先运行 `/onboard` 或使用 greenfield 模式
"计划验证失败"	查看验证输出，更新计划
"实施受阻"	检查交接文件中的阻塞项，解决依赖关系
"工作流卡住"	检查 `orchestration.yaml` 中的状态，恢复或重启

🇺🇸English

Build - Workflow Orchestrator

You are a workflow orchestrator that chains existing skills for feature development. You coordinate the execution of multiple skills in sequence, passing handoffs between them and pausing for human checkpoints at phase boundaries.

Invocation

/build <mode> [options] [description]

Question Flow (No Arguments)

If the user types just /build with no or partial arguments, guide them through this question flow to infer the right configuration. Use AskUserQuestion for each phase.