Canonical Agent Skills library — one skills/ tree for Claude Code, Cursor, Codex, and other agents.

This repository is the source of truth for shared skills: each skill lives at skills/<name>/SKILL.md with YAML frontmatter (name, description). There is no per-agent copy of the catalog in-repo; installers symlink into each agent’s config directory.

Use the Vercel Labs skills CLI to install into detected agents. Prefer Bun and bunx (same idea as npx / pnpm dlx for other runtimes):

# Latest from default branch (all skills, all detected agents)
bunx skills add lgtm-hq/ai-skills -g --all

# Pinned to a release tag (example)
bunx skills add lgtm-hq/ai-skills@v0.1.7 -g --all

# Single agent (e.g. Cursor)
bunx skills add lgtm-hq/ai-skills -a cursor -g

npm / pnpm equivalents (same flags):

npx skills add lgtm-hq/ai-skills -g --all
pnpm dlx skills add lgtm-hq/ai-skills -g --all

Install only the skills you need (names match skills/<name>/ in this repo; see AGENTS.md):

# List available skills without installing
bunx skills add lgtm-hq/ai-skills -l

# Install specific skills globally
bunx skills add lgtm-hq/ai-skills -g --skill lint commit branch -y

# Pin a subset to a release tag
bunx skills add lgtm-hq/ai-skills@v0.1.7 -g --skill lint test -y

The CLI updates by skill name or scope, not by package slug (update lgtm-hq/ai-skills does not match installed skills).

# Update all globally installed skills from their recorded source
bunx skills update -g

# Update only skills you have installed (by name)
bunx skills update lint commit -g

# npm / pnpm equivalents
npx skills update -g
pnpm dlx skills update -g

To move everything to a specific release, reinstall with a tag (replaces symlinks for that install):

bunx skills add lgtm-hq/ai-skills@v0.1.7 -g --all

List or remove installs with bunx skills ls -g and bunx skills remove <name> -g. See the upstream CLI docs for more flags.

• Contributing
• Support
• Security policy
• Code of Conduct

See AGENTS.md for the full list of skills with short descriptions and paths. Regenerate that index after skill changes (see scripts/generate_agents_md.py in CI).

skills/<name>/SKILL.md   # Canonical skill definitions (this is the product)
AGENTS.md                # Human- and agent-readable index
scripts/validate.sh       # Frontmatter, naming, AGENTS sync checks
tests/                    # Pytest wraps for validate script
.github/workflows/      # CI + thin callers into org reusable workflows

flowchart LR
  subgraph repo [ai-skills]
    SK["skills/**/SKILL.md"]
  end
  subgraph cli [skills CLI]
    add["bunx skills add …"]
  end
  subgraph dirs [Agent config]
    c["~/.claude/skills"]
    u["~/.cursor/skills"]
    x["~/.codex/skills …"]
  end
  SK --> add
  add --> c
  add --> u
  add --> x

See CONTRIBUTING.md for org policies, skill layout rules, and full PR expectations. Quick local loop:

git clone https://github.com/lgtm-hq/ai-skills.git
cd ai-skills
uv sync
uv run lintro fmt
uv run lintro chk
uv run lintro tst
bash scripts/validate.sh

Use Conventional Commits for PR titles (squash merge drives release semver). Follow .github/PULL_REQUEST_TEMPLATE.md.

Pull requests and pushes to main run lintro (via the published py-lintro container image) and bash scripts/validate.sh.

Version bumps and CHANGELOG.md updates flow through lgtm-hq/lgtm-ci reusable workflows, called from this repo with full SHA pins on uses: (see .github/workflows/release-version-pr.yml and release-auto-tag.yml). When upstream release behavior changes, bump those SHAs to a commit that exists on GitHub (repos/lgtm-hq/lgtm-ci/commits/<sha>).

Baseline note: lgtm-ci#138 is merged; the current caller pins match lgtm-hq/lgtm-ci main at 79444626c1b3afa4d959b5840b4b5310a46a4095 (re-verify when bumping).

MIT — see LICENSE.