document

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

/document - Documentation Agent

/document - 文档Agent

Model: haiku (templated documentation work)
模型: haiku(模板化文档生成工作)

Command Flags

命令标志

FlagShortDescription
--help
-h
Show available commands and options
--version
-v
Show workflow skills version
标志简写说明
--help
-h
显示可用命令和选项
--version
-v
显示工作流技能版本

Flag Handling

标志处理

On
-h
or
--help
:
/document - Documentation Agent

Usage:
  /document {ID}                     Update docs for a task
  /document -h, --help               Show this help message
  /document -v, --version            Show version

Arguments:
  {ID}    Task ID (number) or task filename (e.g., 001-auth-jwt)

Creates/Updates:
  - docs/features/{feature}.md       Technical documentation
  - docs/guides/{feature}.md         User guides (if user-facing)
  - CLAUDE.md                        Project patterns (if needed)

Examples:
  /document 1                        # Document task #1
  /document 001-auth-jwt             # Using task filename

Next: /ship {ID}
On
-v
or
--version
:
Display:
Workflow Skills v1.5.1
https://github.com/eljun/workflow-skills

当使用
-h
--help
时:
/document - 文档Agent

使用方法:
  /document {ID}                     更新指定任务的文档
  /document -h, --help               显示本帮助信息
  /document -v, --version            显示版本号

参数:
  {ID}    任务ID(数字)或任务文件名(例如:001-auth-jwt)

创建/更新的文件:
  - docs/features/{feature}.md       技术文档
  - docs/guides/{feature}.md         用户指南(面向用户时生成)
  - CLAUDE.md                        项目模式文档(按需更新)

示例:
  /document 1                        # 为任务#1生成文档
  /document 001-auth-jwt             # 使用任务文件名调用

下一步: /ship {ID}
当使用
-v
--version
时:
显示内容:
Workflow Skills v1.5.1
https://github.com/eljun/workflow-skills

When to Use

适用场景

Invoke
/document {ID}
when:
  • Task has passed testing (
    /test
    returned PASS)
  • User has approved the implementation
  • Ready to update project documentation
Example:
/document 1
or
/document 001-dashboard-redesign
当满足以下条件时调用
/document {ID}
  • 任务已通过测试(/test返回PASS)
  • 用户已批准实现方案
  • 准备更新项目文档
示例:
/document 1
/document 001-dashboard-redesign

Task ID Resolution

任务ID解析规则

The
{ID}
can be:
  • Numeric ID:
    1
    ,
    2
    ,
    3
    → Looks up in TASKS.md, finds matching task document
  • Padded ID:
    001
    ,
    002
    → Same as numeric
  • Full filename:
    001-dashboard-redesign
    → Direct file reference
{ID}
可以是以下类型:
  • 数字ID:
    1
    ,
    2
    ,
    3
    → 在TASKS.md中查找匹配的任务文档
  • 补零ID:
    001
    ,
    002
    → 与数字ID处理逻辑相同
  • 完整文件名:
    001-dashboard-redesign
    → 直接关联对应文件

Workflow

工作流

/document {ID}
1. Resolve task ID → find task document
2. Read task document for context
3. Check Automation field (manual | auto)
4. Read test report for verification
5. Update/create feature documentation
6. Update/create user guide (if user-facing)
7. Update CLAUDE.md files if needed
8. Write retrospective → update LEARNINGS.md
9. Move task to "Approved" in TASKS.md
┌─── Automation Mode? ───┐
│                        │
▼ Manual                 ▼ Auto
Notify user              Invoke /ship {ID}
Ready for /ship
/document {ID}
1. 解析任务ID → 找到任务文档
2. 读取任务文档获取上下文
3. 检查自动化字段(manual | auto)
4. 读取测试报告进行验证
5. 更新/创建功能文档
6. 更新/创建用户指南(面向用户时)
7. 按需更新CLAUDE.md文件
8. 编写回顾内容 → 更新LEARNINGS.md
9. 将任务移至TASKS.md中的「已批准」板块
┌─── 是否启用自动化模式? ───┐
│                        │
▼ 手动模式                ▼ 自动模式
通知用户准备好执行/ship    自动调用/ship {ID}
等待执行/ship

Auto Mode Behavior

自动模式行为

When task document has
Automation: auto
:
After documentation is complete, use Task tool to spawn ship agent with model: haiku:
Documentation complete: #{ID} - {Task Title}

Updated files:
- docs/features/{feature}.md
- docs/guides/{feature}.md

[AUTO] Spawning /ship with haiku model...
Task({ subagent_type: "general-purpose", model: "haiku", prompt: "/ship {ID}" })
当任务文档中包含
Automation: auto
字段时:
文档生成完成后,使用任务工具启动ship agent,模型指定为haiku:
文档生成完成:#{ID} - {任务标题}

已更新文件:
- docs/features/{feature}.md
- docs/guides/{feature}.md

[自动模式] 正在启动/ship,使用haiku模型...
Task({ subagent_type: "general-purpose", model: "haiku", prompt: "/ship {ID}" })

Pre-Documentation Checklist

文档生成前检查清单

1. Gather Context (Primary Sources Only)

1. 收集上下文(仅参考主要来源)

Read these files — they contain all the context you need:
docs/task/{ID}-{task-name}.md        - Implementation details
docs/testing/{ID}-{task-name}.md     - Test results & verification
IMPORTANT — Context Efficiency: These two documents were created by the
/task
,
/implement
, and
/test
agents which already analyzed the codebase. Do NOT perform broad codebase exploration. Only read specific source files if you need to verify implementation details for documentation accuracy.
读取以下文件即可获取全部所需上下文:
docs/task/{ID}-{task-name}.md        - 实现细节
docs/testing/{ID}-{task-name}.md     - 测试结果与验证信息
重要提示 - 上下文效率: 上述两份文档由/task、/implement和/test agent生成,它们已完成对代码库的分析。请勿进行大范围代码库探索,仅在需要验证文档准确性时才读取特定源文件。

2. Identify Documentation Needs

2. 确定文档需求

Change TypeDocumentation Needed
New featureFeature doc + User guide
EnhancementUpdate existing docs
Bug fixUpdate troubleshooting sections
API changeUpdate API reference
变更类型需要生成的文档
新功能功能文档 + 用户指南
功能增强更新现有文档
Bug修复更新故障排查章节
API变更更新API参考文档

3. Use Templates

3. 使用模板

Follow the templates defined in this document:
  • Feature Documentation Structure - Technical feature docs (see below)
  • User Guide Structure - User-facing guides (see below)

请遵循本文档中定义的模板:
  • 功能文档结构 - 技术类功能文档(详见下文)
  • 用户指南结构 - 面向用户的操作指南(详见下文)

Documentation Types

文档类型

1. Feature Documentation

1. 功能文档

Location:
docs/features/{FEATURE}.md
Audience: Developers Purpose: Technical implementation details
When to create/update:
  • New feature → Create new doc
  • Enhancement → Update existing doc
  • Significant change → Update relevant sections
存储位置:
docs/features/{FEATURE}.md
受众: 开发人员 目的: 记录技术实现细节
创建/更新时机:
  • 新增功能 → 创建新文档
  • 功能增强 → 更新现有文档
  • 重大变更 → 更新相关章节

2. User Guide

2. 用户指南

Location:
docs/guides/{feature}.md
Audience: End users (business owners, customers) Purpose: How to use the feature
When to create/update:
  • User-facing feature → Create/update guide
  • UI changes → Update screenshots/steps
  • New functionality → Add new sections
存储位置:
docs/guides/{feature}.md
受众: 终端用户(企业所有者、客户) 目的: 说明功能使用方法
创建/更新时机:
  • 面向用户的功能 → 创建/更新指南
  • UI变更 → 更新截图/操作步骤
  • 新增功能 → 添加新章节

3. CLAUDE.md — Living Project Brain

3. CLAUDE.md — 项目动态知识库

Location:
/CLAUDE.md
— auto-loaded by Claude Code on every session, zero cost
CLAUDE.md is the most token-efficient document in the entire project. Every agent reads it for free before any tool call happens. That means stale entries in CLAUDE.md actively mislead agents — they'll follow outdated patterns with full confidence.
The discipline: read → prune → update (in that order)
Before adding anything, read the current CLAUDE.md and ask for each existing entry:
  • Is this still true based on what was just built?
  • Did this task rename a file, change a folder structure, migrate a schema, or replace a pattern?
  • If yes → update or remove the stale entry before adding new ones
Never just append. CLAUDE.md that only grows becomes a liability.

The bar for a CLAUDE.md entry:
An entry earns its place if a future agent would either:
  • Make a mistake without it (wrong pattern, wrong file, wrong approach)
  • Waste time discovering it (file location, non-obvious convention)
  • Make a decision already made (architecture choice, tool selection)
If the answer is "probably fine without it" — don't add it.

Maintain this structure:
markdown
undefined
存储位置:
/CLAUDE.md
— Claude Code会在每次会话开始时自动加载该文件,无需额外成本
CLAUDE.md是整个项目中最节省Token的文档。所有agent在调用任何工具前都会优先读取该文件。这意味着CLAUDE.md中的过时内容会严重误导agent — 它们会完全信任并遵循这些过时的模式。
维护原则: 先阅读 → 再清理 → 最后更新(按此顺序执行)
添加任何内容前,先阅读当前的CLAUDE.md,并对每个现有条目进行检查:
  • 基于刚完成的任务,该条目是否仍然有效?
  • 本次任务是否重命名了文件、修改了文件夹结构、迁移了数据架构或替换了现有模式?
  • 如果是 → 在添加新条目之前,先更新或删除过时内容
切勿直接追加内容。只增不减的CLAUDE.md会成为项目负担。

CLAUDE.md条目收录标准:
满足以下任一条件的内容才能被收录:
  • 缺少该条目会导致未来的agent犯错(错误的模式、错误的文件、错误的实现方式)
  • 缺少该条目会让未来的agent浪费时间去发现(文件位置、非显性约定)
  • 缺少该条目会让未来的agent重复已做过的决策(架构选择、工具选型)
如果答案是「没有也没问题」—— 请勿添加。

保持以下结构:
markdown
undefined

{Project Name}

{项目名称}

Stack

技术栈

{Only what's non-obvious or frequently relevant to agents}
  • Framework: Next.js 14 App Router (NOT Pages Router)
  • Database: Supabase (PostgreSQL) — client at src/lib/supabase/client.ts
  • Auth: Supabase Auth — helpers at src/lib/auth.ts
  • State: Zustand — stores at src/store/
  • Styling: Tailwind CSS
{仅记录非显性或与agent频繁相关的内容}
  • 框架:Next.js 14 App Router(非Pages Router)
  • 数据库:Supabase(PostgreSQL)—— 客户端位于src/lib/supabase/client.ts
  • 认证:Supabase Auth —— 工具函数位于src/lib/auth.ts
  • 状态管理:Zustand —— 状态存储位于src/store/
  • 样式:Tailwind CSS

Project Structure

项目结构

{Update this when directories are created, renamed, or removed} src/ app/ # App Router pages and layouts components/ ui/ # Primitive components (Button, Input, Modal) features/ # Feature-specific components lib/ # Integrations (supabase/, stripe/, etc.) hooks/ # Custom React hooks store/ # Zustand stores
{当目录被创建、重命名或删除时更新} src/ app/ # App Router页面与布局 components/ ui/ # 基础组件(Button、Input、Modal) features/ # 功能专属组件 lib/ # 集成工具(supabase/、stripe/等) hooks/ # 自定义React钩子 store/ # Zustand状态存储

Key Conventions

核心约定

{Only conventions that are actively enforced or non-obvious}
  • Components: PascalCase files and exports
  • Hooks: use* prefix, always in src/hooks/
  • DB queries: always go through src/lib/supabase/ helpers, never raw client calls in components
  • {Add when /simplify catches a violation for the second time — that means it's a real pattern}
{仅记录被严格执行或非显性的约定}
  • 组件:文件与导出均使用PascalCase命名
  • 钩子:以use*为前缀,统一存放于src/hooks/
  • 数据库查询:必须通过src/lib/supabase/工具函数执行,禁止在组件中直接调用客户端
  • {当/simplify第二次检测到违规时添加——说明这是需要严格执行的模式}

Important File Locations

关键文件位置

{Files that agents always need to find — update when files move}
  • Supabase client: src/lib/supabase/client.ts
  • Auth helpers: src/lib/auth.ts
  • Route middleware: src/middleware.ts
  • Environment types: src/types/env.d.ts
{agent经常需要查找的文件,文件移动时更新}
  • Supabase客户端:src/lib/supabase/client.ts
  • 认证工具函数:src/lib/auth.ts
  • 路由中间件:src/middleware.ts
  • 环境变量类型:src/types/env.d.ts

Do Not

禁止操作

{Hard rules — each one from a real mistake, not speculation}
  • Do not import Supabase client directly in components — use lib/supabase/ helpers
  • Do not use
    any
    types — caught by /simplify, will fail quality gate
  • {Add only when something was caught by the quality gate or broke in tests}
{硬性规则——每条规则均来自实际错误,而非推测}
  • 禁止在组件中直接导入Supabase客户端——使用lib/supabase/工具函数
  • 禁止使用
    any
    类型——会被/simplify检测到,无法通过质量校验
  • {仅当被质量校验或测试捕获到问题时添加}

Architecture Decisions

架构决策

{Decisions made so future agents don't re-open the debate}
  • Using Zustand over Redux: simpler API for this app's state complexity
  • App Router over Pages Router: required for server components
  • {Date of decision so agents can judge if it's still current}

---

**Triggers for a CLAUDE.md update after this task:**

| What happened | What to update |
|---------------|----------------|
| New directory created | Project Structure |
| File moved or renamed | Important File Locations |
| Database migration ran | Stack section, any schema notes |
| New package installed | Stack section |
| /simplify caught a repeated violation | Do Not |
| Architecture debate resolved | Architecture Decisions |
| Convention violated and then fixed | Key Conventions |
| Old pattern replaced by new one | Remove old entry, add new |

**Skip the update entirely if:** this task was a small bugfix or UI change that didn't touch architecture, conventions, or file structure.
{记录已确定的决策,避免未来的agent重复讨论}
  • 选择Zustand而非Redux:针对本应用的状态复杂度,Zustand API更简洁
  • 选择App Router而非Pages Router:为了支持服务端组件
  • {记录决策日期,方便agent判断是否仍适用}

---

**本次任务触发CLAUDE.md更新的场景:**

| 变更内容 | 需要更新的板块 |
|---------------|----------------|
| 新增目录 | 项目结构 |
| 文件移动或重命名 | 关键文件位置 |
| 执行数据库迁移 | 技术栈章节、架构说明 |
| 安装新依赖 | 技术栈章节 |
| /simplify检测到重复违规 | 禁止操作 |
| 架构讨论达成结论 | 架构决策 |
| 约定被违规后修复 | 核心约定 |
| 旧模式被新模式替代 | 删除旧条目,添加新条目 |

**无需更新的场景:** 本次任务仅为小型Bug修复或UI变更,未涉及架构、约定或文件结构。

4. Retrospective & LEARNINGS.md

4. 回顾与LEARNINGS.md

Locations:
  • docs/learnings/{ID}-{task-name}.md
    — Full retrospective for this task
  • LEARNINGS.md
    — Project-wide knowledge base (read by
    /task
    and
    /implement
    on every run)
Always write after documentation is complete. This is what makes the workflow self-learning — each completed task teaches future agents what to do and what to avoid.

存储位置:
  • docs/learnings/{ID}-{task-name}.md
    — 本次任务的完整回顾
  • LEARNINGS.md
    — 项目级知识库(/task和/implement在每次执行前都会读取)
必须在文档生成完成后编写。 这是工作流的自学习机制——每个完成的任务都会教会未来的agent该做什么、不该做什么。

Feature Documentation Structure

功能文档结构

markdown
undefined
markdown
undefined

Feature: {Feature Name}

功能:{功能名称}

Status: PRODUCTION Last Updated: {Date}
状态: 已上线 最后更新: {日期}

Overview

概述

{Brief 2-3 sentence description}

{2-3句话简要描述}

User Journey

用户流程

For Customers

面向客户

  1. Step 1
  2. Step 2
  1. 步骤1
  2. 步骤2

For Business Owners

面向企业所有者

  1. Step 1
  2. Step 2

  1. 步骤1
  2. 步骤2

Architecture

架构

File Structure

文件结构

{Accurate file tree - VERIFY paths exist}
{准确的文件树——需验证路径是否存在}

Database Schema

数据库架构

{SQL with comments, if applicable}
{带注释的SQL语句(如有)}

API Endpoints

API端点

MethodEndpointDescription

方法端点说明

Implementation Details

实现细节

Key Components

核心组件

ComponentLocationPurpose
组件位置用途

Technical Notes

技术说明

  • Important detail 1
  • Important detail 2

  • 重要细节1
  • 重要细节2

Related Features

相关功能

  • [Link to related feature]

**Target length:** 200-400 lines

---
  • [关联功能链接]

**目标长度:** 200-400行

---

User Guide Structure

用户指南结构

markdown
undefined
markdown
undefined

{Feature Name} Guide

{功能名称}使用指南

{Brief intro sentence}
{一句话简要介绍}

Quick Start

快速开始

{1-2 paragraph overview}

{1-2段概述}

For Customers

面向客户

How to {Primary Action}

如何{核心操作}

  1. Step with context
  2. Step with context
  1. 带上下文的步骤说明
  2. 带上下文的步骤说明

Tips

使用技巧

  • Tip 1
  • Tip 2

  • 技巧1
  • 技巧2

For Business Owners

面向企业所有者

How to {Admin Action}

如何{管理员操作}

  1. Step with context
  1. 带上下文的步骤说明

Settings

设置选项

SettingLocationWhat it does

设置项位置功能说明

FAQ

常见问题

Q: Question? A: Answer.

问:问题? 答:答案。

Troubleshooting

故障排查

Issue: Description Solution: How to fix

问题: 问题描述 解决方案: 修复方法

Related Guides

相关指南

  • [Link]

**Target length:** 100-200 lines

---
  • [链接]

**目标长度:** 100-200行

---

Retrospective Writing

回顾内容编写

This is the self-learning step. Every completed task generates signal — plan vs reality, quality gate findings, test failures, decisions made. Writing it down means future agents don't repeat the same mistakes.
这是自学习环节。每个完成的任务都会产生经验反馈——计划与实际的差异、质量校验结果、测试失败原因、已做决策。将这些内容记录下来,未来的agent就不会重复犯错。

Step 1: Write Task Retrospective

步骤1:编写任务回顾

File:
docs/learnings/{ID}-{task-name}.md
Gather from sources already read (task doc, simplify Implementation Notes, test report):
markdown
undefined
文件:
docs/learnings/{ID}-{task-name}.md
从已读取的来源(任务文档、/simplify实现说明、测试报告)中收集信息:
markdown
undefined

Retrospective: #{ID} - {Task Title}

回顾:#{ID} - {任务标题}

Date: {date} Quality Gate: PASS on attempt {N} | PASS first try Test cycles: {N} — {0 = passed first time, 1-3 = retried}
日期: {日期} 质量校验: 第{N}次尝试通过 | 首次尝试通过 测试周期: {N} — {0=首次通过,1-3=重试}

Plan vs Reality

计划与实际差异

What was planned

计划内容

{2-3 sentences from task doc Overview — what was the intent}
{任务文档概述中的2-3句话——原目标}

What was delivered

实际交付

{What actually changed — from git diff or Implementation Notes}
{实际变更内容——来自git diff或实现说明}

Deviations from plan

与计划的偏差

{None | Minor: {description} | Medium: {description}}
{无 | 轻微:{描述} | 中等:{描述}}

What the Quality Gate Caught

质量校验捕获的问题

{Issues /simplify flagged — if none, write "None — implementation met standards on first pass"}
These are direct lessons: if the quality gate caught something, future agents doing similar work should know to avoid it upfront.
{/simplify标记的问题——如果没有,填写「无——首次实现即符合标准」}
这些都是直接的经验教训:如果质量校验捕获到问题,未来执行类似任务的agent需要提前知道如何避免。

What Tests Revealed

测试暴露的问题

{What failed on first test run and why — if tests passed first time, write "Tests passed on first run"}
Root cause of any test failures is the most valuable lesson here.
{首次测试失败的内容及原因——如果首次测试通过,填写「首次测试即通过」}
测试失败的根本原因是最有价值的经验。

Key Lessons

核心经验

  1. {Specific, actionable lesson — concrete enough to apply to a future task}
  2. {Another lesson if applicable}
  1. {具体、可落地的经验——需足够具体,可应用于未来任务}
  2. {如有其他经验可补充}

Applies To

适用场景

  • Task type: {feature | bugfix | enhancement}
  • Tech areas: {e.g., "Supabase auth", "Next.js App Router", "React state management"}
undefined
  • 任务类型:{功能开发 | Bug修复 | 功能增强}
  • 技术领域:{例如:Supabase认证、Next.js App Router、React状态管理}
undefined

Step 2: Update LEARNINGS.md

步骤2:更新LEARNINGS.md

File:
LEARNINGS.md
(project root)
Extract the most reusable insight from the retrospective and append it. This file is read by
/task
and
/implement
at the START of every run — keep each entry concise and actionable.
If LEARNINGS.md doesn't exist, create it:
markdown
undefined
文件:
LEARNINGS.md
(项目根目录)
从回顾内容中提取最具复用性的经验并追加到文件中。该文件会在/task和/implement执行前被读取——请保持每条内容简洁且可落地。
如果LEARNINGS.md不存在,创建如下内容:
markdown
undefined

Project Learnings

项目经验库

Auto-updated by /document after each completed task. Read by /task and /implement before starting any work.

由/document在每个任务完成后自动更新。 /task和/implement在开始工作前会读取该文件。

Architecture & Decisions

架构与决策

DecisionRationaleDateTask

决策理由日期关联任务

Established Coding Patterns

已验证的编码模式

{Patterns that have proven to work in this codebase}

{在本代码库中被证明有效的模式}

Common Mistakes to Avoid

需避免的常见错误

{Things the quality gate or tests caught — don't repeat these}

{被质量校验或测试捕获的问题——请勿重复}

Testing Patterns

测试模式

{What works and what doesn't when testing this project}

{在本项目中有效的和无效的测试方法}

Tech Stack Notes

技术栈说明

{Non-obvious things about the stack that future agents should know}

**Append the lesson(s) to the relevant section:**

```markdown
{技术栈的非显性细节,未来的agent需要了解}

**将经验追加到对应章节:**

```markdown

Common Mistakes to Avoid

需避免的常见错误

  • {Mistake in 5 words}: {What to do instead — 1 sentence} Source: Task #{ID} — {task name} ({date})

**Rules for LEARNINGS.md entries:**
- One entry = one actionable lesson, not a summary of the task
- If the lesson is "nothing went wrong," write a pattern that worked well instead
- Never duplicate — check if a similar lesson already exists before appending
- Keep entries scannable: bold the key phrase, explain in one sentence

---
  • {5字以内的错误总结}:{正确做法——1句话} 来源:任务#{ID} — {任务名称}({日期})

**LEARNINGS.md条目规则:**
- 每条条目对应一个可落地的经验,而非任务总结
- 如果经验是「一切顺利」,记录有效的模式
- 切勿重复——追加前检查是否已有类似经验
- 保持内容易读:用粗体标记核心短语,用1句话说明

---

Documentation Checklist

文档校验清单

Feature Doc

功能文档

  • Follows template structure
  • File paths verified to exist
  • API endpoints are accurate
  • Schema matches database
  • Status set to PRODUCTION
  • Last Updated date set
  • Under 400 lines
  • 遵循模板结构
  • 文件路径已验证存在
  • API端点准确
  • 数据架构与数据库匹配
  • 状态设置为「已上线」
  • 已填写最后更新日期
  • 内容不超过400行

User Guide

用户指南

  • Follows template structure
  • Clear step-by-step instructions
  • FAQ answers common questions
  • Troubleshooting section included
  • Under 200 lines
  • 遵循模板结构
  • 操作步骤清晰易懂
  • FAQ覆盖常见问题
  • 包含故障排查章节
  • 内容不超过200行

CLAUDE.md Updates

CLAUDE.md更新

  • Read existing CLAUDE.md before touching it
  • Pruned any entries made stale by this task (moved files, changed patterns, replaced approaches)
  • Stack updated if new dependency added
  • Project Structure updated if directories created or renamed
  • Important File Locations updated if key files moved
  • Key Conventions updated only if a pattern is now actively enforced
  • "Do Not" added only for real mistakes caught by /simplify or tests
  • Skipped entirely if task was a small change with no architectural impact
  • 更新前已读取当前CLAUDE.md
  • 清理了因本次任务而过时的条目(移动的文件、变更的模式、替代的实现方式)
  • 安装新依赖后已更新技术栈章节
  • 创建/重命名目录后已更新项目结构
  • 关键文件移动后已更新关键文件位置
  • 仅当约定被严格执行时才更新核心约定
  • 仅当/simplify或测试捕获到实际错误时才添加禁止操作条目
  • 若任务仅为小型变更且未影响架构,跳过更新

Retrospective & LEARNINGS.md

回顾与LEARNINGS.md

  • docs/learnings/{ID}-{task-name}.md
    created
  • Quality gate findings documented
  • Test cycle count and root cause documented
  • Key lesson(s) appended to
    LEARNINGS.md
  • No duplicate entries in LEARNINGS.md

  • 已创建
    docs/learnings/{ID}-{task-name}.md
  • 已记录质量校验结果
  • 已记录测试周期数与失败根本原因
  • 已将核心经验追加到LEARNINGS.md
  • LEARNINGS.md中无重复条目

Update TASKS.md

更新TASKS.md

Move task to "Approved" section:
markdown
undefined
将任务移至「已批准」板块:
markdown
undefined

Approved

已批准

IDTaskTask DocFeature DocTest ReportApproved
1Quick Actions Redesign001-quick-actions.mdfeature001-quick-actions.mdJan 25

---
ID任务任务文档功能文档测试报告批准日期
1快速操作重设计001-quick-actions.md功能文档001-quick-actions.md1月25日

---

What to Include

应包含的内容

  • Current state only - Document what exists now
  • Accurate file paths - Verify paths exist
  • Working examples - Only code that matches production
  • Clear user journeys - What users actually do
  • 仅记录当前状态 - 仅描述现有内容
  • 准确的文件路径 - 验证路径存在
  • 可运行的示例 - 仅使用与生产环境一致的代码
  • 清晰的用户流程 - 描述用户实际操作

What to Exclude

应排除的内容

RemoveWhy
Development logsMove to changelogs
Before/after comparisonsOnly document current state
"Lessons learned"Dev notes, not docs
Speculative future phasesKeep minimal
Full component codeCode lives in codebase

需移除的内容原因
开发日志移至变更日志
前后对比仅记录当前状态
「经验教训」属于开发笔记,非正式文档
未来阶段的设想尽量简化
完整组件代码代码存放在代码库中

Handoff to /ship

移交至/ship

Check the task document for
Automation: auto
field.
检查任务文档中的
Automation: auto
字段。

Manual Mode

手动模式

Documentation updated for: #{ID} - {Task Title}

Updated files:
- docs/features/{feature}.md (created/updated)
- docs/guides/{feature}.md (created/updated)
- CLAUDE.md (if updated)
- docs/learnings/{ID}-{task-name}.md (retrospective)
- LEARNINGS.md (lesson appended)

Task moved to "Approved" in TASKS.md

Next Steps:
  /ship {ID}              # e.g., /ship 1
  /ship {ID}-{task-name}  # e.g., /ship 001-auth-jwt
文档已更新:#{ID} - {任务标题}

已更新文件:
- docs/features/{feature}.md(创建/更新)
- docs/guides/{feature}.md(创建/更新)
- CLAUDE.md(若已更新)
- docs/learnings/{ID}-{task-name}.md(回顾内容)
- LEARNINGS.md(追加经验)

任务已移至TASKS.md中的「已批准」板块

下一步:
  /ship {ID}              # 示例:/ship 1
  /ship {ID}-{task-name}  # 示例:/ship 001-auth-jwt

Auto Mode

自动模式

Documentation updated for: #{ID} - {Task Title}

Updated files:
- docs/features/{feature}.md (created/updated)
- docs/guides/{feature}.md (created/updated)
- CLAUDE.md (if updated)
- docs/learnings/{ID}-{task-name}.md (retrospective)
- LEARNINGS.md (lesson appended)

Task moved to "Approved" in TASKS.md

[AUTO] Spawning /ship with haiku model...
Use Task tool:
Task({ subagent_type: "general-purpose", model: "haiku", prompt: "/ship {ID}" })

文档已更新:#{ID} - {任务标题}

已更新文件:
- docs/features/{feature}.md(创建/更新)
- docs/guides/{feature}.md(创建/更新)
- CLAUDE.md(若已更新)
- docs/learnings/{ID}-{task-name}.md(回顾内容)
- LEARNINGS.md(追加经验)

任务已移至TASKS.md中的「已批准」板块

[自动模式] 正在启动/ship,使用haiku模型...
使用任务工具:
Task({ subagent_type: "general-purpose", model: "haiku", prompt: "/ship {ID}" })

Related Skills

相关技能

SkillWhen to Use
/implement
If issues found during doc review
/ship
After documentation complete
技能适用场景
/implement
文档审核中发现问题时
/ship
文档生成完成后

Recommended Plugins (Install Separately)

推荐插件(需单独安装)

These plugins must be installed separately. Once installed, they MUST be invoked — do not skip them:
PluginInstall FromWhen to Invoke
vercel-react-best-practices
vercel-labs/agent-skillsReference React patterns for docs
supabase-postgres-best-practices
supabase/agent-skillsReference database patterns for docs
以下插件需单独安装。安装完成后必须调用——请勿跳过:
插件安装来源调用时机
vercel-react-best-practices
vercel-labs/agent-skills编写文档时参考React最佳实践
supabase-postgres-best-practices
supabase/agent-skills编写文档时参考数据库最佳实践