Configuration

Claude Kit works out of the box after installing the plugin. Run /claudekit:init to scaffold project-level configuration for rules, modes, hooks, and MCP servers.

The Init Wizard

/claudekit:init

The wizard walks you through four steps, asking one question at a time:

Step 1: Rules

Rules enforce standards when Claude works on specific file types:

Rule	Applies To	What It Enforces
`api.md`	API routes, controllers	Input validation, error responses, rate limiting
`frontend.md`	`.tsx`, `.jsx`, components	PascalCase, Server Components, accessibility
`migrations.md`	Migration files	Reversible migrations, indexes, transactions
`security.md`	All files	No hardcoded secrets, parameterized queries, no `eval()`
`testing.md`	Test files	Naming conventions, coverage thresholds, no `.only()`

Rules are installed to .claude/rules/.

Step 2: Modes

Modes change how Claude communicates and solves problems:

Mode	Best For
`default`	General tasks
`brainstorm`	Design, ideation
`implementation`	Code-focused, minimal prose
`review`	Critical analysis
`token-efficient`	High-volume work, cost savings
`deep-research`	Investigation, audits
`orchestration`	Multi-agent coordination

Modes are installed to .claude/modes/. Switch with natural language: “switch to brainstorm mode”.

Step 3: Hooks

Hooks run automatically in response to Claude Code events:

Hook	Event	What It Does
`auto-format`	After Write/Edit	Runs ruff (Python) or eslint (JS/TS) on changed files
`block-dangerous-commands`	Before Bash	Blocks `rm -rf /`, force push to main, `DROP TABLE`, etc.
`notify`	Notification	Cross-platform desktop notifications

Hooks are installed to .claude/hooks/ with config in settings.local.json (gitignored).

Step 4: MCP Servers

MCP servers extend Claude with external tools:

Server	Purpose
Context7	Real-time library documentation
Sequential Thinking	Structured step-by-step reasoning
Playwright	Browser automation for E2E testing
Memory	Persistent knowledge graph
Filesystem	Secure file operations

MCP servers are configured in .mcp.json with automatic platform detection (Windows vs macOS/Linux).

Install Everything

Skip all prompts and install everything:

/claudekit:init --all

Project-Level CLAUDE.md

After running init, you may want to create your own .claude/CLAUDE.md for project-specific instructions. This is independent of Claude Kit — it’s a standard Claude Code feature:

# My SaaS Project

## Tech Stack
- **Backend**: FastAPI + PostgreSQL
- **Frontend**: Next.js 14 + Tailwind
- **Auth**: Clerk
- **Payments**: Stripe

## Code Conventions
- Python: PEP 8, type hints required
- TypeScript: Strict mode, Zod for validation

## Testing
- Python: pytest with 80% coverage minimum
- Frontend: vitest + Playwright

Agent Behavior Overrides

You can customize agent behavior in your CLAUDE.md:

## Agent Behavior Overrides

### claudekit:planner
- Break tasks into 15-60 minute chunks
- Always identify testing requirements

### claudekit:code-reviewer
- Enforce strict typing
- Security-first reviews
- Check for test coverage

### claudekit:tester
- Use pytest for Python, vitest for TypeScript
- Generate edge case tests

Next Steps

Workflows — See how skills work together
Skills Reference — Browse all 44 skills
Creating Skills — Build your own