ClaudeKit

Definition

ClaudeKit is a toolkit that provides smart guardrails and workflow automation for Claude Code. It catches errors in real time, saves checkpoints, and strengthens AI coding with expert subagents. It stands out for its automatic codebase map, 6-agent parallel code review, and spec-driven development workflow.

Setup is a global npm install, then claudekit setup. The repo is mostly TypeScript, under MIT, at v0.9.5, with 657 stars and 105 forks. Changed files are typechecked, linted, and tested, and a stop hook stores a checkpoint you can restore.

Basic Information

Installation

npm install -g claudekit
claudekit setup

Architecture

claudekit/
├── .claude/          # Claude configs
├── .claudekit/       # Kit configs
├── bin/              # CLI binaries
├── cli/              # CLI source
├── docs/             # Documentation
├── examples/         # Example configs
├── scripts/          # Build scripts
├── specs/            # Spec definitions
├── src/              # Core source (hooks, agents, commands)
└── tests/            # Test suite

Core Features

Instant Codebase Navigation

Claude sees the entire project structure automatically, skipping most discovery searches.

• Navigate directly to code
• No trial-and-error
• Confident code access
• See relationships & dependencies
• Automatic setup, updates as you code

Comprehensive Code Review

6 parallel agents analyze the code:

Git Checkpoint System

• Auto-saves checkpoints when Claude stops
• /checkpoint:restore to undo changes
• List, restore, or clean up checkpoints

Real-time Error Prevention

Enhanced AI Reasoning

• 4 configurable thinking levels (0-3)
• < 5ms overhead
• Session-aware, activates on first prompt

AI Subagents

Iterative Spec Implementation

6-phase workflow: Implementation -> Test Writing -> Code Review -> Iterative Improvement -> Commit -> Progress Tracking

Hooks

CLI Commands

claudekit setup              # Interactive setup
claudekit setup --yes        # Quick setup
claudekit setup --all        # Install everything
claudekit list               # Show components
claudekit list agents        # List agents with token counts
claudekit doctor             # Health check
claudekit-hooks profile      # Profile hook performance
claudekit-hooks disable/enable <hook>  # Session hook control

Slash Commands

Configuration

.claude/settings.json - Hooks config:

{
  "hooks": {
    "PreToolUse": [{"matcher": "Read|Edit|MultiEdit|Write", "hooks": [...]}],
    "PostToolUse": [{"matcher": "Write|Edit|MultiEdit", "hooks": [...]}],
    "Stop": [{"matcher": "*", "hooks": [...]}]
  }
}

.claudekit/config.json - Hook config:

{
  "hooks": {
    "typecheck-changed": { "command": "npm run typecheck" },
    "thinking-level": { "level": 2 },
    "codebase-map": { "include": ["src/**"], "exclude": ["**/*.test.ts"] }
  }
}

Tech Stack

Pros

Cons

When to Use

• TypeScript-heavy projects: Need real-time type checking
• Code quality focus: Want automated guards and reviews
• Spec-driven development: Want a structured implementation workflow
• Team environments: Need consistent quality standards
• Hook profiling: Want to optimize hook performance

Six agents review in parallel. They cover design patterns, security and OWASP, performance, tests, code smells, and documentation. Checkpoints save when Claude stops. You can list them, restore one, or clean them up. Guards block the any type, catch style issues, and stop code from being swapped for comments. Tests run on file changes. File security scans 195 plus patterns in 12 categories. Thinking has four levels from 0 to 3, adds under 5ms, and starts on the first prompt. Subagents include triage, research listed at a 90% time reduction, code search, TypeScript, React, databases, an oracle that needs setup, refactoring, the Vercel AI SDK v5, and Nest.js. Spec work has six phases, from implementation and tests through review, improvement, commit, and progress tracking. The CLI can install everything, list agents with token counts, run doctor, and profile hooks. The stack is Node.js 20 or newer, Vitest, Biome, and ESLint. The Max plan is required for token use. Many hooks can slow the flow, the agent list is smaller than ECC, and there is no cross-session memory.

References:

• carlrannaberg/claudekit
• npm: claudekit

See also: Claudekit Cli · Claudekit Skills · Kit Comparison