All tools
HarnessCommunity732

claudekit

Updated Jul 5, 2026

An npm-installed kit of smart guardrails, workflow commands, and AI subagents for Claude Code. Real-time hooks catch TypeScript errors, lint violations, leaked secrets, and test failures as Claude works. Slash commands cover code review (/code-review with 6 parallel agents), git workflows, spec creation, and checkpoint management. Ships 30+ domain-expert subagents.

View on GitHub

What it does

  • claudekit setup

    Interactive setup wizard that installs selected hooks, commands, and agents into your project.

  • /code-review

    6 parallel specialized agents analyze architecture, security, performance, testing, quality, and documentation simultaneously.

  • file-guard hook

    Blocks AI access to sensitive files using 195+ patterns — detects risky bash constructs like find-pipe-xargs-cat before execution.

  • /checkpoint:restore

    Restore the working tree to the last auto-saved git checkpoint — one-command undo for any Claude session.

  • /spec:create

    Research the codebase and generate a comprehensive spec document ready for /spec:execute.

  • claudekit-hooks profile

    Profile all configured hooks for execution time and output size — identifies slow hooks and those approaching Claude's context limits.

Files (1)

OVERVIEW.mdprimary · markdown · 2.7 KB
## What it does

claudekit is a smart guardrails and workflow automation kit for Claude Code. It acts as a safety net: hooks catch errors in real time as Claude edits, commands automate routine workflows, and domain-expert subagents handle specialized analysis in parallel.

**Real-time hooks (the core value):**

- **file-guard** — Blocks AI access to sensitive files (.env, AWS credentials, SSH keys, etc.) using 195+ patterns across 12 categories. Advanced bash command analysis detects risky constructs like `find -name '*.env' | xargs cat`.
- **typecheck-changed** — Runs TypeScript type checking on changed files immediately after edits.
- **lint-changed** — Runs Biome/ESLint on changed files before they accumulate.
- **test-changed** — Runs relevant tests for changed files so failures surface immediately.
- **check-any-changed** — Blocks `any` types in TypeScript files as Claude edits.
- **commit-quality** — Pre-commit hook blocking debugger statements and hardcoded secrets.
- **create-checkpoint** — Auto-saves a git checkpoint when Claude stops (enables easy rollback).
- **codebase-map** — Injects a structured project map once per session so Claude navigates without file-by-file exploration.

**Slash commands:**

- /code-review — 6 parallel specialized agents (architecture, security, performance, testing, quality, documentation) with prioritized findings.
- /git:commit — Smart commit following your project's conventions.
- /git:status — Intelligent git analysis grouped by change type.
- /git:checkout — Branch creation with conventional naming.
- /git:ignore-init — AI-safe .gitignore patterns for 195+ sensitive file types.
- /spec:create — Research codebase and generate comprehensive specifications.
- /spec:execute — Implement specifications via iterative 6-phase workflow.
- /checkpoint:create / /checkpoint:restore — Manual checkpoint management.
- /validate-and-fix — Run all quality checks and fix issues.
- /research — Deep parallel research with specialized subagents.

**30+ AI subagents** including code-review-expert (the 6-agent parallel reviewer), typescript-expert, react-expert, testing-expert, database-expert, security-expert, oracle (deep debugging), refactoring-expert, ai-sdk-expert, nestjs-expert, and more.

## Install

```bash
npm install -g claudekit
claudekit setup              # interactive setup
claudekit setup --yes        # quick setup with defaults
claudekit setup --all        # install everything including all agents
```

Verify: `claudekit doctor`

## Key config

Hooks are configured in .claude/settings.json and .claudekit/config.json. Run `claudekit list` to see all installed components and `claudekit-hooks profile` to measure hook execution time and output size.