Sep 13, 2026

Build a Curated Skill Library with Shelves and Provenance Using ai-agent-skills

The ai-agent-skills CLI turns a flat pile of installed skills into a managed library with shelves, collections, and per-skill notes.

#tutorial#skill-management#agent-tools

Most agents accumulate skills the way most people accumulate browser tabs. ai-agent-skills is an npm CLI that ships one maintainer's curated library — 140 skills across 6 shelves, per its README — and gives you the same tooling to build and manage your own. It works with any Agent Skills-compatible agent, and the repo sits at 1,140 stars on the SkillMap leaderboard.

Why This Skill Matters

Installing skills is easy; knowing why a skill is on your machine is not. This project models a library instead of a pile: skills sit on shelves by kind of work (Frontend, Backend, Mobile, Workflow, Agent Engineering, Marketing), every pick carries provenance — which upstream repo it came from — and a note explaining why it earned a place. Two tiers keep it lean: 17 house copies ship locally with the npm package and install offline, while 123 cataloged upstream skills stay as metadata and install from their source repos on demand. The README's own positioning: use skills.sh for the broad ecosystem, use this when you want a smaller, cared-for library.

Installation

There is nothing to install ahead of time — run it with npx:

npx ai-agent-skills list

Default install targets match common conventions: ~/.claude/skills/ for global installs and .agents/skills/ for project installs. Twelve legacy agent targets (claude, cursor, codex, amp, vscode, copilot, gemini, goose, opencode, letta, kilocode, project) still work through --agent <name>.

Real Workflow: Start Your Own Library

  1. Create a managed workspace and enter it:
npx ai-agent-skills init-library my-library
cd my-library
  1. Add a bundled pick to a shelf, recording why it belongs:
npx ai-agent-skills add frontend-design --area frontend --branch Implementation --why "I want this on my shelf."
  1. Add a skill from an upstream repo, then install, refresh, and rebuild the docs:
npx ai-agent-skills add anthropics/skills --skill webapp-testing --area workflow --branch Testing --why "Browser-level checks in the workspace."
npx ai-agent-skills install frontend-design -p
npx ai-agent-skills sync frontend-design -p
npx ai-agent-skills build-docs
  1. Bootstrap an existing flat repo of skills in place — invalid private-only names are skipped and reported, and low-confidence imports fall back to the workflow shelf with a needs-curation label:
cd ~/projects/my-skills
npx ai-agent-skills init-library . --areas "mobile,workflow,agent-engineering" --import --auto-classify

A second workflow is browsing before committing. npx ai-agent-skills list shows shelves, info <skill> shows provenance, search <query> and collections find candidates, and install <owner/repo> --list previews any upstream repo. The bundled collections are ready-made starter stacks — test-and-debug, docs-and-research, or build-apps — and install --collection <name> installs one wholesale.

Tips

  • Use catalog instead of add when you want an upstream skill recorded without a local copy, and vendor when you explicitly want a house copy.
  • curate <skill> --why "..." rewrites the note on an existing pick; curate review opens the cleanup queue.
  • The library launched December 17, 2025, before a default universal installer existed; the README now describes it as a place to build and manage your own library, while noting it still installs.
  • Health checks exist: doctor, check, and validate [path] cover the workspace.

When Not to Use This

If you want every skill from every publisher, a broad registry like skills.sh serves that goal better — this project deliberately keeps shelves small and curated. And since the bundled library reflects one maintainer's working setup (76 of 140 skills sit on the Marketing shelf), treat it as a starting point to edit, not a canonical catalog.


See the leaderboard for more skills.