Quick Start

# 1. Initialize project
reskill init

# 2. Install a skill
reskill install github:anthropics/skills/frontend-design@latest

# 3. List installed skills
reskill list

What is reskill?

reskill is a Git-based package manager for AI agent skills, similar to npm or Go modules. It provides declarative configuration, version locking, and seamless synchronization for managing skills across projects and teams.

Supports: Cursor, Claude Code, Codex, OpenCode, Windsurf, GitHub Copilot, and more.

Why reskill?

reskill offers fine-grained skill management and synchronization:

Local Experience

• Declarative config — skills.json clearly expresses project dependencies
• Global cache — Avoid redundant downloads, speed up installation
• Local development — Use link to develop and debug skills locally

Engineering-Grade Management

• Version locking — skills.lock ensures team consistency
• Flexible versioning — Support exact versions, semver ranges, branches, and commits
• Git as Registry — No additional services needed, any Git repo is a skill source

Cross-Project Sync

• Version controlled — Commit skills.json and skills.lock to your repo
• CI integration — Run reskill install in CI to verify dependencies
• Multi-registry — Support GitHub, GitLab, and private repositories

Flexible Version Strategy

{
  "skills": {
    "frontend-design": "github:anthropics/skills/frontend-design@latest",
    "code-review": "github:team/code-review@v2.1.0",
    "testing": "github:team/testing@^1.0.0"
  }
}

When running reskill update:

• @latest skills automatically update to the newest tag
• @v2.1.0 stays locked
• @^1.0.0 updates to the latest 1.x.x version

Installation

# Global install
npm install -g reskill

# Or use npx
npx reskill <command>

Usage

Source Formats

# GitHub shorthand
reskill install github:user/skill@v1.0.0

# Full URL
reskill install https://github.com/user/skill

# GitLab
reskill install gitlab:group/skill@latest

# Private registry
reskill install gitlab.company.com:team/skill@v1.0.0

# Default registry (from skills.json)
reskill install user/skill@v1.0.0

Version Specification

Commands

Run reskill <command> --help for detailed options.

Configuration

skills.json

{
  "name": "my-project",
  "skills": {
    "planning": "github:user/planning-skill@v1.0.0",
    "code-review": "gitlab:team/code-review@latest"
  },
  "defaults": {
    "registry": "github",
    "installDir": ".skills"
  },
  "registries": {
    "internal": "https://gitlab.company.com"
  }
}

skills.lock

The lock file records exact versions and commit hashes to ensure reproducible installations across your team.

Multi-Agent Support

reskill works with all major AI coding agents. Skills are installed to the .skills/ directory by default and can be integrated with any agent.

Skill Repository Structure

Each skill repository should follow this structure:

my-skill/
├── skill.json           # Metadata (required)
├── SKILL.md             # Main entry document (required)
├── README.md            # Repository description
└── templates/           # Template files (optional)

skill.json

{
  "name": "my-skill",
  "version": "1.0.0",
  "description": "A skill for ...",
  "author": "Your Name",
  "license": "MIT",
  "entry": "SKILL.md",
  "keywords": ["ai", "skill"]
}

Project Structure

After installation:

my-project/
├── skills.json          # Dependency declaration
├── skills.lock          # Version lock file
└── .skills/             # Installation directory
    ├── planning/
    │   ├── skill.json
    │   └── SKILL.md
    └── code-review/
        ├── skill.json
        └── SKILL.md

Environment Variables

Development

# Install dependencies
pnpm install

# Development mode
pnpm dev

# Build
pnpm build

# Run tests
pnpm test

# Type check
pnpm typecheck

Acknowledgements

reskill was inspired by and references the implementations of these excellent projects:

• add-skill by Vercel Labs
• skild by Peiiii
• openskills by Numman Ali

Thanks to these projects for pioneering the AI agent skills ecosystem!

Related Links

• Agent Skills Specification

License

MIT