fullstack-feature
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseResources
相关资源
scripts/
validate-feature-workflow.sh
references/
phase-templates.mdscripts/
validate-feature-workflow.sh
references/
phase-templates.mdFullstack Feature Orchestration
全栈功能编排
This skill defines the end-to-end workflow for implementing complete features that span multiple layers of the stack. It orchestrates specialized agents through 7 distinct phases, from requirements clarification through delivery.
此技能定义了实现跨技术栈多层完整功能的端到端工作流。它通过7个不同阶段编排专业Agent,从需求澄清一直到交付完成。
When to Use This Skill
何时使用此技能
Use fullstack-feature when the user requests:
- A complete feature that requires database, API, and UI changes
- End-to-end implementation with testing and review
- A new user-facing capability that touches multiple components
- Anything described as: "build X", "add X feature", "implement X capability"
Do NOT use this skill for:
- Single-layer changes (just API, just UI)
- Bug fixes (use bugfix or hotfix skills)
- Code review only (use review-scoring skill)
- Refactoring without new functionality
当用户提出以下请求时,使用fullstack-feature技能:
- 需要修改数据库、API和UI的完整功能
- 包含测试和评审的端到端实现
- 涉及多个组件的新用户面向能力
- 任何描述为“构建X”、“添加X功能”、“实现X能力”的需求
请勿将此技能用于:
- 单一层级的修改(仅API或仅UI)
- Bug修复(使用bugfix或hotfix技能)
- 仅代码评审(使用review-scoring技能)
- 无新功能的重构
Overview
概述
The fullstack feature workflow consists of 7 phases:
- Understand - Clarify requirements, load skills, identify affected layers
- Foundation (sequential) - Database schema/migrations, type generation
- Core Implementation (parallel) - API endpoints, UI components, state management
- Integration (sequential) - Wire UI to API, state to UI, verify data flow
- Quality (parallel) - Tests, security check, accessibility check
- Review (WRFC) - Full code review, fix issues, re-review until score >= 9.5
- Commit + Log - Git commit, update goodvibes memory/logs, report to user
全栈功能工作流包含7个阶段:
- 需求理解 - 澄清需求、加载相关技能、识别受影响的技术层
- 基础构建(顺序执行)- 数据库架构/迁移、类型生成
- 核心实现(并行执行)- API端点、UI组件、状态管理
- 集成联调(顺序执行)- 将UI与API关联、状态与UI关联、验证数据流
- 质量保障(并行执行)- 测试、安全检查、无障碍性检查
- 代码评审(WRFC循环)- 完整代码评审、修复问题、重新评审直至得分≥9.5
- 提交与日志 - Git提交、更新goodvibes记忆/日志、向用户汇报
Phase 1: Understand
阶段1:需求理解
Purpose
目标
Ensure you have complete clarity on what to build before spawning work agents. This phase prevents wasted effort from misunderstandings.
在启动工作Agent前,确保完全明确要构建的内容。此阶段可避免因误解导致的无效工作。
Steps
步骤
-
Clarify requirements with the user:
- What is the exact functionality requested?
- Who are the users and what are their goals?
- What are the acceptance criteria (how do we know it's done)?
- Are there specific constraints (performance, accessibility, browser support)?
- Are there existing patterns to follow or avoid?
-
Identify affected layers:
- Database: New tables/columns? Migrations?
- API: New endpoints? Modify existing?
- Types: New type definitions?
- UI: New components? Modify existing?
- State: New state management?
- Tests: Unit, integration, e2e?
-
Load relevant outcome skills:
- Backend: ,
trpc,prisma,postgresql,clerk,nextauth,graphqlrest-api-design - Frontend: ,
nextjs,react,tailwindcss,shadcn-uiframer-motion - Protocol: ,
discover-plan-batch,review-scoringgoodvibes-memory
- Backend:
-
Check goodvibes memory:
- - Architectural choices
.goodvibes/memory/decisions.json - - Coding patterns
.goodvibes/memory/patterns.json - - Past failures to avoid
.goodvibes/memory/failures.json
-
Create implementation plan:
- Decompose feature into sub-tasks
- Identify dependencies between sub-tasks
- Estimate scope (how many files/components)
- Map sub-tasks to phases (Foundation, Core, Integration)
-
与用户澄清需求:
- 请求的具体功能是什么?
- 目标用户是谁,他们的需求是什么?
- 验收标准是什么(如何判断功能完成)?
- 是否有特定约束(性能、无障碍性、浏览器支持)?
- 是否有需要遵循或避免的现有模式?
-
识别受影响的技术层:
- 数据库:是否需要新表/字段?是否需要迁移?
- API:是否需要新端点?是否需要修改现有端点?
- 类型:是否需要新的类型定义?
- UI:是否需要新组件?是否需要修改现有组件?
- 状态:是否需要新的状态管理?
- 测试:是否需要单元测试、集成测试、端到端测试?
-
加载相关技能:
- 后端:,
trpc,prisma,postgresql,clerk,nextauth,graphqlrest-api-design - 前端:,
nextjs,react,tailwindcss,shadcn-uiframer-motion - 协议:,
discover-plan-batch,review-scoringgoodvibes-memory
- 后端:
-
检查goodvibes记忆:
- - 架构决策
.goodvibes/memory/decisions.json - - 编码模式
.goodvibes/memory/patterns.json - - 需要避免的过往错误
.goodvibes/memory/failures.json
-
创建实现计划:
- 将功能拆解为子任务
- 识别子任务之间的依赖关系
- 评估范围(涉及的文件/组件数量)
- 将子任务映射到对应阶段(基础构建、核心实现、集成联调)
Output
输出
At the end of Phase 1, you must have:
- Requirements document (in memory/current-task.md)
- Affected layers list (database, API, UI, tests)
- Implementation plan (decomposed sub-tasks with dependencies)
- Loaded skills (relevant outcome skills available)
阶段1结束时,必须完成以下内容:
- 需求文档(存储在内存/current-task.md中)
- 受影响技术层列表(数据库、API、UI、测试)
- 实现计划(带依赖关系的拆解子任务)
- 已加载的相关技能
Mode-Specific Behavior
模式特定行为
Vibecoding: Confirm with user before proceeding to Phase 2.
Justvibes: Auto-proceed if requirements are clear. (Pause for clarification if ambiguous.)
Vibecoding模式:进入阶段2前与用户确认。
Justvibes模式:若需求明确则自动推进(若需求模糊则暂停澄清)。
Phase 2: Foundation (Sequential)
阶段2:基础构建(顺序执行)
Purpose
目标
Establish the data model and type foundation before building any upstream code. This ensures type safety and consistency across layers.
在构建任何上游代码前,先建立数据模型和类型基础。确保跨层的类型安全性和一致性。
Steps
步骤
-
Database schema (if required):
- Spawn database agent to design schema
- Create migration files
- Run migrations (dev environment)
- Verify migration success
-
Type generation (if using Prisma/Drizzle/etc):
- Run ORM type generation command
- Verify types are generated
- Commit generated types (if necessary)
-
Shared type definitions (if not using ORM):
- Create files
src/types/[...].ts - Define entities, DTOs, request/response types
- Export from barrel
index.ts
- Create
-
数据库架构(若需要):
- 启动数据库Agent设计架构
- 创建迁移文件
- 在开发环境中运行迁移
- 验证迁移成功
-
类型生成(若使用Prisma/Drizzle等ORM):
- 运行ORM类型生成命令
- 验证类型已生成
- 提交生成的类型(若必要)
-
共享类型定义(若未使用ORM):
- 创建文件
src/types/[...].ts - 定义实体、DTO、请求/响应类型
- 通过桶文件导出
index.ts
- 创建
Agent Instructions
Agent指令
Database agent:
Protocol skills (required for all agents): discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- Use protocol
discover-plan-batch - Follow existing schema patterns (check memory/decisions.json)
- Add appropriate indexes for query performance
- Add foreign key constraints
- Use timestamps (,
createdAt)updatedAt - Test migration before completing
Engineer agent (type generation):
Protocol skills (required for all agents): discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- Run type generation command
- Verify types are correct (spot check generated code)
- Run to confirm no type errors
tsc
数据库Agent:
所有Agent必须使用的协议技能:discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- 使用协议
discover-plan-batch - 遵循现有架构模式(查看memory/decisions.json)
- 为查询性能添加适当的索引
- 添加外键约束
- 使用时间戳(,
createdAt)updatedAt - 完成前测试迁移
工程师Agent(类型生成):
所有Agent必须使用的协议技能:discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- 运行类型生成命令
- 验证类型正确性(抽查生成的代码)
- 运行确认无类型错误
tsc
Clear Definition of Done
明确的完成标准
- Database schema exists, migrations applied
- Types generated and exported
- passes
npm run typecheck - All files committed
- 数据库架构已存在,迁移已应用
- 类型已生成并导出
- 执行通过
npm run typecheck - 所有文件已提交
Mode-Specific Behavior
模式特定行为
Vibecoding: Checkpoint after this phase. Commit with message .
Justvibes: Auto-proceed to Phase 3.
checkpoint: foundation - schema and typesVibecoding模式:此阶段结束时设置检查点,提交信息为。
Justvibes模式:自动推进到阶段3。
checkpoint: foundation - schema and typesPhase 3: Core Implementation (Parallel)
阶段3:核心实现(并行执行)
Purpose
目标
Implement the main functionality across API and UI layers in parallel. This maximizes throughput and minimizes wait time.
并行实现API和UI层的主要功能,最大化吞吐量并减少等待时间。
Sub-Phases
子阶段
These sub-phases run in parallel:
- API endpoints (API agent)
- UI components (UI agent)
- State management (State agent)
这些子阶段并行执行:
- API端点(API Agent)
- UI组件(UI Agent)
- 状态管理(状态Agent)
Agent Instructions
Agent指令
API agent:
Protocol skills (required for all agents): discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- Implement endpoints using existing patterns (check decisions.json for API style)
- Add input validation (use Zod or similar)
- Add error handling
- Add authentication/authorization checks
- Use ORM for database access
- Add logging at appropriate levels
- Use protocol
discover-plan-batch
UI agent:
Protocol skills (required for all agents): discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- Create components using existing patterns (check decisions.json for UI libraries)
- Follow accessibility best practices (ARIA attributes, keyboard navigation)
- Implement responsive design
- Add loading and error states
- Optimize performance (,
useMemo)useCallback - Use protocol
discover-plan-batch
State agent:
Protocol skills (required for all agents): discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- Implement state management using existing patterns (Zustand, Redux, etc)
- Define state shape and actions
- Add selectors for derived state
- Add persistence if required (localStorage, sessionStorage)
- Use protocol
discover-plan-batch
API Agent:
所有Agent必须使用的协议技能:discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- 遵循现有模式实现端点(查看decisions.json了解API风格)
- 添加输入验证(使用Zod或类似工具)
- 添加错误处理
- 添加认证/授权检查
- 使用ORM访问数据库
- 在适当级别添加日志
- 使用协议
discover-plan-batch
UI Agent:
所有Agent必须使用的协议技能:discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- 遵循现有模式创建组件(查看decisions.json了解UI库)
- 遵循无障碍最佳实践(ARIA属性、键盘导航)
- 实现响应式设计
- 添加加载和错误状态
- 优化性能(,
useMemo)useCallback - 使用协议
discover-plan-batch
状态Agent:
所有Agent必须使用的协议技能:discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- 遵循现有模式实现状态管理(Zustand、Redux等)
- 定义状态结构和操作
- 添加派生状态的选择器
- 若需要则添加持久化(localStorage、sessionStorage)
- 使用协议
discover-plan-batch
Clear Definition of Done
明确的完成标准
- All API endpoints implemented and validated
- All UI components created
- State management implemented
- passes
npm run typecheck - passes
npm run lint - All files committed
- 所有API端点已实现并验证
- 所有UI组件已创建
- 状态管理已实现
- 执行通过
npm run typecheck - 执行通过
npm run lint - 所有文件已提交
Mode-Specific Behavior
模式特定行为
Vibecoding: Checkpoint after this phase. Commit with message .
Justvibes: Auto-proceed to Phase 4.
checkpoint: core implementationVibecoding模式:此阶段结束时设置检查点,提交信息为。
Justvibes模式:自动推进到阶段4。
checkpoint: core implementationPhase 4: Integration (Sequential)
阶段4:集成联调(顺序执行)
Purpose
目标
Connect UI to API, wire state to UI, and verify data flows end-to-end. This phase must be sequential because it depends on Phase 3 completion.
将UI与API关联、状态与UI关联,并验证端到端数据流。此阶段必须顺序执行,因为它依赖阶段3的完成。
Steps
步骤
-
Wire UI to API:
- Add API client calls to components
- Handle loading states during API calls
- Handle error states from API
- Display API responses in UI
-
Wire state to UI:
- Connect state management to components
- Use selectors to access state
- Dispatch actions on user interactions
- Verify UI updates when state changes
-
End-to-end verification:
- Manually test data flow: UI -> API -> Database -> API -> UI
- Confirm loading states appear
- Confirm error states appear on failure
- Confirm success states appear on success
-
Fix integration issues:
- Fix type mismatches between UI and API
- Fix data mapping issues
- Fix state synchronization issues
-
UI与API关联:
- 在组件中添加API客户端调用
- 处理API调用期间的加载状态
- 处理API返回的错误状态
- 在UI中展示API响应
-
状态与UI关联:
- 将状态管理与组件连接
- 使用选择器访问状态
- 在用户交互时分发操作
- 验证状态变化时UI是否更新
-
端到端验证:
- 手动测试数据流:UI -> API -> 数据库 -> API -> UI
- 确认加载状态正常显示
- 确认失败时错误状态正常显示
- 确认成功时成功状态正常显示
-
修复集成问题:
- 修复UI与API之间的类型不匹配
- 修复数据映射问题
- 修复状态同步问题
Agent Instructions
Agent指令
Integration agent:
Protocol skills (required for all agents): discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- Read both API and UI code to understand interfaces
- Add API client calls in correct lifecycle hooks (, event handlers)
useEffect - Ensure error handling is present
- Ensure loading states are displayed
- Test each integration point manually
- Use protocol
discover-plan-batch
集成Agent:
所有Agent必须使用的协议技能:discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- 阅读API和UI代码以理解接口
- 在正确的生命周期钩子(、事件处理器)中添加API客户端调用
useEffect - 确保存在错误处理
- 确保加载状态正常显示
- 手动测试每个集成点
- 使用协议
discover-plan-batch
Clear Definition of Done
明确的完成标准
- UI components can call API endpoints
- Data flows from UI -> API -> DB -> API -> UI
- State works correctly
- Loading states display
- Error states display
- passes
npm run typecheck - All files committed
- UI组件可调用API端点
- 数据流从UI -> API -> DB -> API -> UI正常流转
- 状态工作正常
- 加载状态正常显示
- 错误状态正常显示
- 执行通过
npm run typecheck - 所有文件已提交
Mode-Specific Behavior
模式特定行为
Vibecoding: Checkpoint after this phase. Commit with message .
Justvibes: Auto-proceed to Phase 5.
checkpoint: integrationVibecoding模式:此阶段结束时设置检查点,提交信息为。
Justvibes模式:自动推进到阶段5。
checkpoint: integrationPhase 5: Quality (Parallel)
阶段5:质量保障(并行执行)
Purpose
目标
Verify the implementation meets quality standards across multiple dimensions. These checks run in parallel.
从多个维度验证实现是否符合质量标准。这些检查并行执行。
Sub-Phases
子阶段
These sub-phases run in parallel:
- Tests (Tester agent)
- Security check (Security agent)
- Accessibility check (A11y agent)
这些子阶段并行执行:
- 测试(测试Agent)
- 安全检查(安全Agent)
- 无障碍性检查(A11y Agent)
Agent Instructions
Agent指令
Tester agent:
Protocol skills (required for all agents): discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- Write unit tests for API endpoints
- Write component tests for UI
- Write integration tests for data flow
- Target >=80% coverage for new code
- Test both happy path and edge cases
- Use protocol
discover-plan-batch
Security agent:
Protocol skills (required for all agents): discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- Check for input validation on all endpoints
- Check for authentication/authorization checks
- Check for SQL injection vulnerabilities
- Check for XSS vulnerabilities
- Check for exposed secrets
- Check for CORS configuration
- Use protocol
discover-plan-batch
A11y agent:
Protocol skills (required for all agents): discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- Check for ARIA attributes on interactive elements
- Check for keyboard navigation support
- Check for color contrast (WCAG AA)
- Check for focus indicators
- Check for alt text on images
- Use protocol
discover-plan-batch
测试Agent:
所有Agent必须使用的协议技能:discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- 为API端点编写单元测试
- 为UI编写组件测试
- 为数据流编写集成测试
- 新代码的测试覆盖率目标≥80%
- 测试正常路径和边缘情况
- 使用协议
discover-plan-batch
安全Agent:
所有Agent必须使用的协议技能:discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- 检查所有端点的输入验证
- 检查认证/授权检查
- 检查SQL注入漏洞
- 检查XSS漏洞
- 检查是否有暴露的密钥
- 检查CORS配置
- 使用协议
discover-plan-batch
A11y Agent:
所有Agent必须使用的协议技能:discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- 检查交互元素的ARIA属性
- 检查键盘导航支持
- 检查颜色对比度(符合WCAG AA标准)
- 检查焦点指示器
- 检查图片的替代文本
- 使用协议
discover-plan-batch
Clear Definition of Done
明确的完成标准
- Tests written and passing
- Security check complete (no critical issues)
- Accessibility check complete (no critical issues)
- All files committed
- 测试已编写并通过
- 安全检查完成(无严重问题)
- 无障碍性检查完成(无严重问题)
- 所有文件已提交
Mode-Specific Behavior
模式特定行为
Vibecoding: Checkpoint after this phase. Commit with message .
Justvibes: Auto-proceed to Phase 6.
checkpoint: quality checksVibecoding模式:此阶段结束时设置检查点,提交信息为。
Justvibes模式:自动推进到阶段6。
checkpoint: quality checksPhase 6: Review (WRFC Loop)
阶段6:代码评审(WRFC循环)
Purpose
目标
Ensure all code meets production standards through systematic review and fix. Use the Work-Review-Fix-Check loop until score >= 9.5.
通过系统的评审和修复,确保所有代码符合生产标准。使用Work-Review-Fix-Check循环直至得分≥9.5。
WRFC Loop
WRFC循环
-
Review (Reviewer agent):
- Apply skill to all new code
review-scoring - Score across all 10 dimensions
- Produce structured review with issues and verdict
- Apply
-
Fix (Fix agent):
- Address all Critical and Major issues
- Address Minor issues unless explicitly deprioritized
- Produce fix report
-
Check (Re-reviewer agent):
- Verify all issues were fixed
- Re-score all 10 dimensions
- Identify new issues (if any)
- Determine verdict: PASS (>= 9.5), CONDITIONAL PASS (8.0-9.49), FAIL (< 8.0)
-
Loop:
- If FAIL or CONDITIONAL PASS, repeat from step 1
- If PASS, exit the loop
-
评审(评审Agent):
- 对所有新代码应用技能
review-scoring - 从10个维度打分
- 生成包含问题和结论的结构化评审报告
- 对所有新代码应用
-
修复(修复Agent):
- 解决所有严重和主要问题
- 除非明确优先级降低,否则解决次要问题
- 生成修复报告
-
验证(复评审Agent):
- 验证所有问题是否已修复
- 重新从10个维度打分
- 识别新问题(若有)
- 确定结论:通过(≥9.5)、有条件通过(8.0-9.49)、不通过(<8.0)
-
循环:
- 若不通过或有条件通过,从步骤1重复
- 若通过,退出循环
Agent Instructions
Agent指令
Reviewer agent:
Protocol skills (required for all agents): discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- Use skill
review-scoring - Score objectively, use the rubric literally
- Provide specific FILE:LINE references
- Provide specific fix suggestions
- Categorize issues as Critical/Major/Minor
Fix agent:
Protocol skills (required for all agents): discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- Address all Critical and Major issues
- Document fixes applied
- Document issues not fixed (with reasons)
- Do not claim fixed without actual changes
- Use protocol
discover-plan-batch
Re-reviewer agent:
Protocol skills (required for all agents): discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- Verify each previously flagged issue was fixed
- Re-score from scratch (DO NOT copy previous scores)
- Identify new issues
- Use skill
review-scoring
评审Agent:
所有Agent必须使用的协议技能:discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- 使用技能
review-scoring - 客观打分,严格遵循评分标准
- 提供具体的文件:行号引用
- 提供具体的修复建议
- 将问题分类为严重/主要/次要
修复Agent:
所有Agent必须使用的协议技能:discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- 解决所有严重和主要问题
- 记录已应用的修复
- 记录未修复的问题及原因
- 未实际修改则不要声称已修复
- 使用协议
discover-plan-batch
复评审Agent:
所有Agent必须使用的协议技能:discover-plan-batch, precision-mastery, error-recovery, goodvibes-memory
- 验证每个之前标记的问题是否已修复
- 重新从头打分(不要复制之前的分数)
- 识别新问题
- 使用技能
review-scoring
Clear Definition of Done
明确的完成标准
- Overall score >= 9.5/10
- Verdict: PASS
- All Critical and Major issues resolved
- All files committed
- 整体得分≥9.5/10
- 结论:通过
- 所有严重和主要问题已解决
- 所有文件已提交
Mode-Specific Behavior
模式特定行为
Vibecoding: No checkpoint here. Commit only when PASS verdict reached.
Justvibes: Auto-loop until PASS.
Vibecoding模式:此阶段不设置检查点,仅在结论为通过时提交。
Justvibes模式:自动循环直至通过。
Phase 7: Commit + Log
阶段7:提交与日志
Purpose
目标
Finalize the feature by committing to git and updating project memory.
通过Git提交和更新项目记忆来完成功能。
Steps
步骤
-
Create git commit:
- Stage all new/modified files
- Write clear commit message (follow existing conventions)
- Include
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> - Commit
-
Update goodvibes memory:
- Add to (if new patterns emerged)
.goodvibes/memory/patterns.json - Add to (if new decisions made)
.goodvibes/memory/decisions.json - Use skill
goodvibes-memory
- Add to
-
Update goodvibes logs:
- Log feature completion to
.goodvibes/logs/tasks.jsonl - Include feature name, files changed, final score
- Use skill
goodvibes-memory
- Log feature completion to
-
Report to user:
- Summarize what was built
- List files created/modified
- List commit SHA
- List final review score
- Provide next steps (if any)
-
创建Git提交:
- 暂存所有新增/修改的文件
- 编写清晰的提交信息(遵循现有规范)
- 包含
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> - 执行提交
-
更新goodvibes记忆:
- 若出现新模式,添加到
.goodvibes/memory/patterns.json - 若做出新决策,添加到
.goodvibes/memory/decisions.json - 使用技能
goodvibes-memory
- 若出现新模式,添加到
-
更新goodvibes日志:
- 将功能完成记录到
.goodvibes/logs/tasks.jsonl - 包含功能名称、修改的文件、最终得分
- 使用技能
goodvibes-memory
- 将功能完成记录到
-
向用户汇报:
- 总结已构建的内容
- 列出创建/修改的文件
- 列出提交SHA
- 列出最终评审得分
- 提供后续步骤(若有)
Clear Definition of Done
明确的完成标准
- All files committed to git
- Memory updated
- Logs updated
- User report provided
- 所有文件已提交到Git
- 记忆已更新
- 日志已更新
- 已向用户汇报
Mode-Specific Behavior
模式特定行为
Vibecoding: Confirm commit message with user before committing.
Justvibes: Auto-commit with generated message.
Vibecoding模式:提交前与用户确认提交信息。
Justvibes模式:使用生成的信息自动提交。
Mode-Specific Behavior
模式特定行为
Vibecoding Mode
Vibecoding模式
Vibecoding mode is collaborative. The orchestrator:
- Confirms after Phase 1 (Understand) before proceeding
- Checkpoints after Phases 2, 3, 4, 5 (creates git commits)
- Confirms commit message before final commit in Phase 7
- Pauses for approval if uncertainties arise
Checkpoint commit messages:
checkpoint: foundation - schema and typescheckpoint: core implementationcheckpoint: integrationcheckpoint: quality checks
Vibecoding模式为协作模式。编排器:
- 确认:阶段1(需求理解)后进入下一阶段前与用户确认
- 检查点:阶段2、3、4、5结束时设置检查点(创建Git提交)
- 确认:阶段7最终提交前与用户确认提交信息
- 若出现不确定情况则暂停等待批准
检查点提交信息:
checkpoint: foundation - schema and typescheckpoint: core implementationcheckpoint: integrationcheckpoint: quality checks
Justvibes Mode
Justvibes模式
Justvibes mode is autonomous. The orchestrator:
- Auto-proceeds through all phases without user input
- Pauses only if requirements are ambiguous or critical decisions are needed
- Auto-loops in Phase 6 (until PASS verdict)
- Auto-commits in Phase 7 (generates commit message)
Justvibes模式为自主模式。编排器:
- 自动推进:无需用户输入即可完成所有阶段
- 仅在需求模糊或需要关键决策时暂停
- 阶段6自动循环直至通过
- 阶段7自动提交(生成提交信息)
Agent Orchestration Patterns
Agent编排模式
Sequential Phases
顺序阶段
Phases 2, 4, 6, 7 must be sequential:
- Phase 2 (Foundation): Must complete before Phase 3 because API/UI code depends on types
- Phase 4 (Integration): Must come after Phase 3 because it connects API+UI
- Phase 6 (Review): Must come after Phase 5 because it reviews completed work
- Phase 7 (Commit+Log): Must be last because it finalizes everything
阶段2、4、6、7必须顺序执行:
- 阶段2(基础构建):必须在阶段3前完成,因为API/UI代码依赖类型
- 阶段4(集成联调):必须在阶段3后执行,因为它需要连接API和UI
- 阶段6(代码评审):必须在阶段5后执行,因为它评审已完成的工作
- 阶段7(提交与日志):必须最后执行,因为它是最终收尾
Parallel Phases
并行阶段
Phases 3 and 5 spawn multiple agents in parallel:
- Phase 3 (Core): API agent, UI agent, State agent run in parallel
- Phase 5 (Quality): Tester agent, Security agent, A11y agent run in parallel
All agents in a parallel phase must complete before proceeding to the next phase.
阶段3和5会启动多个Agent并行执行:
- 阶段3(核心实现):API Agent、UI Agent、状态Agent并行执行
- 阶段5(质量保障):测试Agent、安全Agent、A11y Agent并行执行
并行阶段的所有Agent必须全部完成后,才能推进到下一阶段。
Agent Communication
Agent通信
Agents should NOT communicate directly. Instead:
- Agents write results to filesystem
- Orchestrator reads agent outputs
- Orchestrator passes relevant info to next agents
Agent之间不应直接通信。而是:
- Agent将结果写入文件系统
- 编排器读取Agent的输出
- 编排器将相关信息传递给下一个Agent
Integration with Other Protocols
与其他协议的集成
Discover-Plan-Batch (DPB)
Discover-Plan-Batch(DPB)
All work agents must follow the DPB loop:
- Discover: Use tool to run parallel grep/glob/symbol queries
discover - Plan: Identify files to create/modify, order of operations, batch opportunities
- Batch: Group operations into batched precision_engine calls
- Loop: Return to discovery if assumptions change
See skill for details.
discover-plan-batch所有工作Agent必须遵循DPB循环:
- 发现:使用工具运行并行grep/glob/符号查询
discover - 计划:确定要创建/修改的文件、操作顺序、批量处理机会
- 批量处理:将操作分组为批量的precision_engine调用
- 循环:若假设发生变化则返回发现步骤
详情请查看技能。
discover-plan-batchReview-Scoring (WRFC)
Review-Scoring(WRFC)
Phase 6 uses the Work-Review-Fix-Check loop:
- Work: Phases 2-5 produce the work
- Review: Reviewer agent scores using review-scoring rubric
- Fix: Fix agent addresses issues
- Check: Re-reviewer agent verifies fixes
- Loop until score >= 9.5
See skill for details.
review-scoring阶段6使用Work-Review-Fix-Check循环:
- 工作:阶段2-5生成工作成果
- 评审:评审Agent使用review-scoring评分标准打分
- 修复:修复Agent解决问题
- 验证:复评审Agent验证修复
- 循环直至得分≥9.5
详情请查看技能。
review-scoringGoodvibes Memory
Goodvibes记忆
Use skill to:
goodvibes-memory- Read in Phase 1
.goodvibes/memory/{decisions,patterns,failures}.json - Write to in Phase 7
.goodvibes/logs/tasks.jsonl - Update memory files with new patterns/decisions in Phase 7
使用技能:
goodvibes-memory- 阶段1中读取
.goodvibes/memory/{decisions,patterns,failures}.json - 阶段7中写入
.goodvibes/logs/tasks.jsonl - 阶段7中用新模式/决策更新记忆文件
Common Pitfalls
常见陷阱
Starting Implementation Too Early
过早启动实现
Do not spawn work agents until Phase 1 is complete. Clarify requirements first.
在阶段1完成前不要启动工作Agent。先澄清需求。
Parallelizing Sequential Phases
并行顺序阶段
Do not run Phases 2 and 3 in parallel. Phase 3 depends on Phase 2 types.
不要并行执行阶段2和3。阶段3依赖阶段2生成的类型。
Skipping Quality Phase
跳过质量保障阶段
Do not skip Phase 5. Tests are not optional. Security is not optional.
不要跳过阶段5。测试不是可选的。安全不是可选的。
Accepting Low Review Scores
接受低评审得分
Do not exit Phase 6 until score >= 9.5. Loop until standards are met.
阶段6得分未达≥9.5前不要退出。循环直至符合标准。
Committing Without Review
未评审就提交
Do not commit in Phase 7 until Phase 6 is complete. All code must be reviewed.
阶段6完成前不要在阶段7提交。所有代码必须经过评审。
Forgetting to Update Memory
忘记更新记忆
Do not skip memory updates in Phase 7. Future agents rely on this context.
不要跳过阶段7的记忆更新。未来的Agent依赖此上下文。
Validation
验证
Use to verify your orchestration follows this workflow. See for detailed templates and examples.
scripts/validate-feature-workflow.shreferences/phase-templates.md使用验证你的编排是否遵循此工作流。查看获取详细模板和示例。
scripts/validate-feature-workflow.shreferences/phase-templates.md