project-docs（项目文档）

Project Documentation Generator

Generate structured project documentation by analysing the codebase. Produces docs that reflect the actual code, not aspirational architecture.

When to Use

• New project needs initial documentation
• Docs are missing or stale
• Onboarding someone to the codebase
• Post-refactor doc refresh

Workflow

1. Detect Project Type

Scan the project root to determine what kind of project this is:

2. Ask What to Generate

Which docs should I generate?
1. ARCHITECTURE.md — system overview, stack, directory structure, key flows
2. API_ENDPOINTS.md — routes, methods, params, response shapes, auth
3. DATABASE_SCHEMA.md — tables, relationships, migrations, indexes
4. All of the above

Only offer docs that match the project. Don't offer API_ENDPOINTS.md for a static site. Don't offer DATABASE_SCHEMA.md if there's no database.

3. Scan the Codebase

For each requested doc, read the relevant source files:

ARCHITECTURE.md — scan:

• package.json / pyproject.toml (stack, dependencies)
• Entry points (src/index.ts, src/main.tsx, src/App.tsx)
• Config files (wrangler.jsonc, vite.config.ts, tsconfig.json)
• Directory structure (top 2 levels)
• Key modules and their exports

API_ENDPOINTS.md — scan:

• Route files (src/routes/, src/api/, or inline in index)
• Middleware files (auth, CORS, logging)
• Request/response types or Zod schemas
• Error handling patterns

DATABASE_SCHEMA.md — scan:

• Drizzle schema files (src/db/schema.ts, src/schema/)
• Migration files (drizzle/, migrations/)
• Raw SQL files if present
• Seed files if present

4. Generate Documentation

Write each doc to docs/ (create the directory if it doesn't exist). If the project already has docs there, offer to update rather than overwrite.

For small projects with no docs/ directory, write to the project root instead.

Document Templates

ARCHITECTURE.md

# Architecture

## Overview
[One paragraph: what this project does and how it's structured]

## Stack
| Layer | Technology | Version |
|-------|-----------|---------|
| Runtime | [e.g. Cloudflare Workers] | — |
| Framework | [e.g. Hono] | [version] |
| Database | [e.g. D1 (SQLite)] | — |
| ORM | [e.g. Drizzle] | [version] |
| Frontend | [e.g. React 19] | [version] |
| Styling | [e.g. Tailwind v4] | [version] |

## Directory Structure
[Annotated tree — top 2 levels with purpose comments]

## Key Flows
### [Flow 1: e.g. "User Authentication"]
[Step-by-step: request → middleware → handler → database → response]

### [Flow 2: e.g. "Data Processing Pipeline"]
[Step-by-step through the system]

## Configuration
[Key config files and what they control]

## Deployment
[How to deploy, environment variables needed, build commands]

API_ENDPOINTS.md

# API Endpoints

## Base URL
[e.g. `https://api.example.com` or relative `/api`]

## Authentication
[Method: Bearer token, session cookie, API key, none]
[Where tokens come from, how to obtain]

## Endpoints

### [Group: e.g. Users]

#### `GET /api/users`
- **Auth**: Required
- **Params**: `?page=1&limit=20`
- **Response**: `{ users: User[], total: number }`

#### `POST /api/users`
- **Auth**: Required (admin)
- **Body**: `{ name: string, email: string }`
- **Response**: `{ user: User }` (201)
- **Errors**: 400 (validation), 409 (duplicate email)

[Repeat for each endpoint]

## Error Format
[Standard error response shape]

## Rate Limits
[If applicable]

DATABASE_SCHEMA.md

# Database Schema

## Engine
[e.g. Cloudflare D1 (SQLite), PostgreSQL, MySQL]

## Tables

### `users`
| Column | Type | Constraints | Description |
|--------|------|-------------|-------------|
| id | TEXT | PK | UUID |
| email | TEXT | UNIQUE, NOT NULL | User email |
| name | TEXT | NOT NULL | Display name |
| created_at | TEXT | NOT NULL, DEFAULT now | ISO timestamp |

### `posts`
[Same format]

## Relationships
[Foreign keys, join patterns, cascading rules]

## Indexes
[Non-primary indexes and why they exist]

## Migrations
- Generate: `npx drizzle-kit generate`
- Apply local: `npx wrangler d1 migrations apply DB --local`
- Apply remote: `npx wrangler d1 migrations apply DB --remote`

## Seed Data
[Reference to seed script if one exists]

Quality Rules

1. Document what exists, not what's planned — read the actual code, don't invent endpoints or tables
2. Include versions — extract from package.json/lock files, not from memory
3. Show real response shapes — copy from TypeScript types or Zod schemas in the code
4. Keep it scannable — tables over paragraphs, code blocks over prose
5. Don't duplicate CLAUDE.md — if architecture info is already in CLAUDE.md, either move it to ARCHITECTURE.md or reference it
6. Flag gaps — if you find undocumented routes or tables without clear purpose, note them with <!-- TODO: document purpose -->

Updating Existing Docs

If docs already exist:

1. Read the existing doc
2. Diff against the current codebase
3. Show the user what's changed (new endpoints, removed tables, updated stack)
4. Apply updates preserving any hand-written notes or sections

Never silently overwrite custom content the user has added to their docs.