agent-skillet 0.2.0

Install, version, and sync agent skills in your repository to your favorite agents (Claude, Cursor, Copilot, and similar).

Agent Skillet

Prepare and serve agent skills!

Agent Skillet helps teams install, version, and sync agent skills inside a repository.

Install

From PyPI (recommended)

uvx agent-skillet init
uv tool install agent-skillet

From source (development)

uv pip install -e .

If you do not already have uv, run:

zsh install.sh

Quick Start

1. Initialize Skillet in your project

   skillet init
   

   This command prompts which agentic coding tool you're using, then

   • Seeds current project managed by skillet with every bundled skill in skills folder
   • Materializes them under .skillet/skills/<skill_name>/, and mirrors into configured agent folders

2. Add new skills from Github (owner/repo or owner/repo/path[@ref]):

   skillet add anthropics/skills/frontend-design@main
   

Configuration

1. To configure which agentic coding tool(s) skills show up for (Cursor, Claude Code, and other agents each have their own mirror paths):

   skillet config    # update your agentic coding tool
   skillet sync      # re-apply tracked sources and refresh native mirrors
   

skillet sync is a local reconciliation command. It reads .skillet/config/sources.json, materializes/removes skills under .skillet/skills/, and updates native mirror directories. It is not a "check for latest remote updates" command.

Browse skills

1. Browse skills publicly with skillet find <query> or local ones with skillet search <query>.

How It Works

• Tracks installed skill sources in .skillet/config/sources.json.
• Materializes installed skills into .skillet/skills/<name>/SKILL.md.
• Mirrors enabled skills into agent-native directories (for example .cursor/skills/ and .claude/skills/).
• Supports local sources and GitHub specs (owner/repo, owner/repo/subpath, owner/repo/subpath@ref).
• GitHub installs download GitHub’s full repository archive tarball for the resolved ref, then copy to the skill folder from your path segment
• skillet sync reconciles to tracked source specs and local managed state; it should not be assumed to always pull newer upstream content when a remote ref moves.

Example sources.json

Each entry maps a skill name to its source. The kind field is either "local" or "github".

{
  "git-os": {
    "kind": "local",
    "source": "git-os"
  },
  "python-design-patterns": {
    "kind": "github",
    "source": "wshobson/agents/python-design-patterns@main"
  },
  "skill-creator": {
    "kind": "github",
    "source": "anthropics/skills/skill-creator"
  }
}

skillet add writes these entries for you — you rarely need to edit the file directly.

Common Commands Examples

# Add a local skill directory (must contain SKILL.md)
skillet add ./team-skills/checkout-flow

# Add all skills from a GitHub repo (owner/repo)
skillet add wshobson/agents

# Add a single skill from a GitHub repo (owner/repo/subpath)
skillet add anthropics/skills/skill-creator

# Pin to a specific branch or tag (owner/repo/subpath@ref)
skillet add wshobson/agents/python-design-patterns@main

# Reconcile tracked sources after editing .skillet/config/sources.json
skillet sync

# Find skills on skills.sh
skillet find <query>

# Search local skills
skillet search <query>

# List installed skills
skillet list

# Remove a skill
skillet remove skill-creator

Tip: skillet.lock records origins with a github: prefix (e.g. github:anthropics/skills/skill-creator). skillet add accepts both forms, so you can copy-paste a lock origin directly as a spec.

Bundled Skills

• find-skills: Discover and install skills from the Agent Skillet ecosystem
• git-os: Conventional commits, atomic changes, and GIT-OS workflow
• sprint: Ticket-to-PR automation with branch and description templates
• deploy-checklist: Pre/post deployment verification checklist

Documentation

• Authoring skills
• Development guide
• Releasing
• TODO

Contributing

Contributions are welcome and encouraged.

• Open an issue first for bug reports, feature requests, or design discussion.
• Keep pull requests focused and small; include clear context in the description.
• Add or update tests when behavior changes.
• Run local checks before opening a PR:

uv sync
ruff check
pytest

• Be respectful and collaborative in reviews so we can keep the project healthy and active!

License

MIT