claude-code/claude/commands/update-codemaps.md

Update Codemaps

Analyze the codebase structure and generate token-lean architecture documentation.

Step 1: Scan Project Structure

1. Identify the project type (monorepo, single app, library, microservice)
2. Find all source directories (src/, lib/, app/, packages/)
3. Map entry points (main.ts, index.ts, app.py, main.go, etc.)

Step 2: Generate Codemaps

Create or update codemaps in docs/CODEMAPS/ (or .reports/codemaps/):

File	Contents
architecture.md	High-level system diagram, service boundaries, data flow
backend.md	API routes, middleware chain, service → repository mapping
frontend.md	Page tree, component hierarchy, state management flow
data.md	Database tables, relationships, migration history
dependencies.md	External services, third-party integrations, shared libraries

Codemap Format

Each codemap should be token-lean — optimized for AI context consumption:

# Backend Architecture

## Routes
POST /api/users → UserController.create → UserService.create → UserRepo.insert
GET  /api/users/:id → UserController.get → UserService.findById → UserRepo.findById

## Key Files
src/services/user.ts (business logic, 120 lines)
src/repos/user.ts (database access, 80 lines)

## Dependencies
- PostgreSQL (primary data store)
- Redis (session cache, rate limiting)
- Stripe (payment processing)

Step 3: Diff Detection

1. If previous codemaps exist, calculate the diff percentage
2. If changes > 30%, show the diff and request user approval before overwriting
3. If changes <= 30%, update in place

Step 4: Add Metadata

Add a freshness header to each codemap:

<!-- Generated: 2026-02-11 | Files scanned: 142 | Token estimate: ~800 -->

Step 5: Save Analysis Report

Write a summary to .reports/codemap-diff.txt:

• Files added/removed/modified since last scan
• New dependencies detected
• Architecture changes (new routes, new services, etc.)
• Staleness warnings for docs not updated in 90+ days

Tips

• Focus on high-level structure, not implementation details
• Prefer file paths and function signatures over full code blocks
• Keep each codemap under 1000 tokens for efficient context loading
• Use ASCII diagrams for data flow instead of verbose descriptions
• Run after major feature additions or refactoring sessions