Setup

Set up Claude Code optimization for this project. Analyze the codebase and generate project-specific files.

Steps

1. Detect Project Stack

Read these files (skip if not found):

• package.json → Node.js/JS/TS project
• pyproject.toml or requirements.txt → Python
• go.mod → Go
• Cargo.toml → Rust
• Gemfile → Ruby
• pom.xml or build.gradle → Java
• docker-compose.yml → Docker setup
• turbo.json or pnpm-workspace.yaml → Monorepo

2. Find Entry Points

# Look for main/index files
Glob("**/index.{ts,js,tsx,jsx,py}")
Glob("**/main.{ts,js,py,go,rs}")
Glob("**/app.{ts,js,py}")
Glob("**/server.{ts,js}")

3. Find Route/API Registration

Grep("router|app\.(get|post|put|delete|use)|@(Get|Post|Controller)|urlpatterns|HandleFunc")

4. Detect Test Framework

Glob("**/*.test.*")
Glob("**/*.spec.*")
Glob("**/tests/**")
Glob("**/__tests__/**")

5. Detect Config Files

Glob("**/.eslintrc*")
Glob("**/.prettierrc*")
Glob("**/tsconfig.json")
Glob("**/ruff.toml")
Glob("**/.editorconfig")

6. Check Git Info

git log --oneline -5
git branch --show-current

7. Generate CLAUDE.md

Write CLAUDE.md to the project root with:

# Project: [detected name or $ARGUMENTS]

## Commands
- Build: [detected from package.json scripts or Makefile]
- Test: [detected test command]
- Lint: [detected lint command]
- Dev: [detected dev command]

## Entry Points
- [detected entry files with descriptions]

## Request Flow
[detected from framework: e.g., Client → Express → Middleware → Route → Controller → Service → DB]

## Data Flow
[detected from ORM/database setup]

## Key Decisions
- [fill in — prompt user to complete this section]

## Things That Look Wrong But Aren't
- [fill in — prompt user to complete this section]

## Conventions
- [detected from linter config, existing code patterns]
- [detected test conventions]

## Code Annotations
- Search `@claude-entry` for system entry points
- Search `@claude-pattern` for examples to follow
- Search `@claude-warning` for files not to modify

## Workflow
- New feature → /plan first, then /smart-edit to implement
- Bug fix → /debug-error or /fix-issue [number]
- Before editing unfamiliar code → /explore-area [dir]
- After all changes → /review then /commit
- Ready to merge → /create-pr
- New developer → /onboard

8. Generate .claudeignore

Write .claudeignore based on detected stack:

• Always include: node_modules, .git, dist, build, .log, .env, coverage, *.min.js
• Node.js: + package-lock.json, yarn.lock, pnpm-lock.yaml, .next, .nuxt
• Python: + pycache, *.pyc, .venv, .mypy_cache, .ruff_cache, *.egg-info
• Go: + vendor, bin
• Rust: + target
• Assets: + *.png, *.jpg, *.gif, *.ico, *.woff, *.woff2, *.ttf, *.mp4, *.pdf

9. Generate .claude/settings.json (if not exists)

Only if .claude/settings.json doesn't already exist, copy it from the global settings:

{
  "env": {
    "MAX_THINKING_TOKENS": "10000"
  },
  "hooks": {
    // ... all hooks pointing to ~/.claude/hooks/
  }
}

10. Summary

Tell the user:

Setup complete!

Created:
  + CLAUDE.md — edit Key Decisions and Things That Look Wrong sections
  + .claudeignore — [X] patterns for [stack]
  + .claude/settings.json — hooks wired, thinking tokens capped

Auto-detected:
  - Stack: [framework, language, database]
  - Entry points: [count] files
  - Test framework: [name]
  - Commands: build, test, lint, dev

Next: fill in the [bracketed sections] in CLAUDE.md with your project-specific details.