Skip to main content
Glama

🍳 skillet

A package manager for AI agent skills

Find, install, version and share SKILL.md skills β€” from a Git-backed registry. No server, no account, no lock-in. Just npx skillet add <skill>.

npx skillet add pdf

Agent Skills are taking over β€” a SKILL.md folder that teaches an agent a new capability (read PDFs, build slide decks, scrape the web…). But sharing them is a mess: you copy-paste from random repos, pin nothing, and have no way to discover what exists.

skillet is npm/brew for skills. One command to install a skill into your project, a lockfile so it's reproducible, and a registry that's just a JSON file in a Git repo β€” so there's nothing to host and anyone can contribute with a PR.

npx skillet search pdf            # discover
npx skillet add pdf               # install into .claude/skills/
npx skillet list                  # see what's installed
npx skillet new my-skill          # scaffold your own

Why skillet

  • Skills are files, not dependencies. Like shadcn/ui, skillet copies the skill into your repo (.claude/skills/<name>/) where you can read and tweak it β€” not into an opaque node_modules.

  • Reproducible. Every install records the exact commit SHA in skillet.lock.json. Commit it, and your whole team gets byte-identical skills with skillet install (the npm ci of skills). Pin a single install with add owner/repo#<sha>; skillet update re-resolves a branch/tag to its latest.

  • Zero infrastructure. The registry is a JSON index in a Git repo, served over raw GitHub. No backend, no database, no API keys. Adding a skill is a PR.

  • Install from anywhere. A registry name, any owner/repo[/path][#ref], or a local folder.

  • Zero dependencies. Pure Node built-ins + your system git. The whole CLI is a few hundred readable lines.

Related MCP server: skill4agent MCP Server

Install targets

npx skillet add pdf                              # from the registry
npx skillet add anthropics/skills/skills/pptx    # any GitHub repo + subpath
npx skillet add owner/repo#v2.1.0                # a tag/branch
npx skillet add owner/repo#<commit-sha>          # pin to an exact commit
npx skillet add ./skills/my-local-skill          # a local folder

Skills install into .claude/skills/ by default (the common 2026 convention). Change it per-project with skillet init or --dir.

Use it from Claude (MCP)

skillet speaks the Model Context Protocol, so Claude Desktop / Claude Code can search and install skills for you β€” "find a PDF skill and install it" just works. Add to claude_desktop_config.json (or a project .mcp.json):

{
  "mcpServers": {
    "skillet": {
      "command": "npx",
      "args": ["-y", "skillet", "mcp"]
    }
  }
}

Tools exposed: skillet_search, skillet_install (registry-only, name-validated), skillet_list. Zero dependencies β€” a few hundred lines of JSON-RPC over stdio.

Browse the registry

npx skillet gallery        # builds a static, searchable HTML gallery β†’ site/

skillet gallery renders registry/index.json into a single self-contained page (search, copy-to-install, links) β€” zero backend. The included GitHub Pages workflow rebuilds and publishes it automatically whenever the registry changes, so the registry has a shareable home.

Authoring a skill

npx skillet new web-scraper      # creates web-scraper/SKILL.md from a template
# edit it…
npx skillet validate ./web-scraper

A skill is just a folder with a SKILL.md:

---
name: web-scraper
description: Scrape pages and extract structured data; use when the user wants web content.
version: 0.1.0
license: MIT
keywords: [scrape, http]
---

# web-scraper
Instructions the agent reads… plus any supporting scripts in the same folder.

Push it to GitHub, then open a PR adding one entry to registry/index.json β€” see docs/SPEC.md.

Commands

skillet search [query]

search the registry

skillet add <ref>

install a skill (registry name / owner/repo[/path][#ref] / ./local)

skillet install

install all locked skills at their pinned commits (npm ci-style)

skillet list

list installed skills

skillet remove <name>

uninstall

skillet update [name]

re-install tracked skill(s) at the latest ref

skillet new <name>

scaffold a new skill

skillet validate [path]

validate a SKILL.md

skillet gallery

build a static, searchable registry gallery

skillet mcp

run as an MCP server (stdio) for Claude/agents

skillet init

write skillet.json config

Flags: --force, --dir <path>, --registry <url|path>, --json.

How it works

  skillet add pdf
        β”‚  resolve name in registry index (raw GitHub JSON)
        β–Ό
  git clone --depth 1 anthropics/skills      ← your system git, partial clone
        β”‚  copy skills/pdf/ β†’ .claude/skills/pdf/
        β–Ό
  pin commit SHA in skillet.lock.json

The "registry" is one JSON file. That's the whole backend.

Compatibility

Works with anything that reads SKILL.md skill folders (Claude Code / Claude Agent Skills and compatible runtimes). skillet is just discovery + install + versioning around the open SKILL.md format β€” it doesn't lock you to a runtime.

Status

Early MVP β€” discovery, install (registry / GitHub / local), lockfile pinning, authoring and validation all work today. Star/watch to follow along; PRs and new registry entries are the most useful contribution right now.

Sibling projects

Part of a small, local-first, zero-dependency toolkit for building AI agents:

  • 🍳 skillet β€” a package manager for agent skills (this repo)

  • πŸ”­ tracelet β€” local DevTools to debug agent runs

  • 🧠 engram β€” a local, private memory layer for agents (and you)

License

MIT β€” see LICENSE. (Skills installed through skillet keep their own licenses.)

A
license - permissive license
-
quality - not tested
B
maintenance

Maintenance

–Maintainers
–Response time
–Release cycle
–Releases (12mo)
Commit activity

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/jnMetaCode/skillet'

If you have feedback or need assistance with the MCP directory API, please join our Discord server