Agent Skills are a lightweight, open format for extending AI agent capabilities with specialized knowledge and workflows.
This is a portable library of them for Claude Code, Cursor, Codex CLI, Antigravity CLI, and
OpenCode users who want their agent to discover a repo's own conventions rather than have a skill
dictate new ones. Each one started as something built and iterated on while working on a real
application, then got rewritten into a plain SKILL.md: a description that tells an agent when to
reach for it, and a body that tells it what to actually do.
Prerequisites: git, plus the GitHub CLI (gh), logged in with
gh auth login, and Node ≥22.20 for the npx skills install. Some skills (create-pr,
review-code, write-tests, and a few others) read PRs, issues, or repo settings through gh.
Without it they fall back to plain git where they can, and stop where they can't.
To use it in your own project:
Claude Code: install the plugin.
claude plugin marketplace add awcameron/skills
claude plugin install awcameron-skills@awcameron-skillsAny tool: install with the third-party skills CLI,
run in your project:
npx skills add awcameron/skills # all skills
npx skills add awcameron/skills --list # browse before installing
npx skills add awcameron/skills --skill <name> # one skill (repeat --skill for more)It installs each skill's whole folder, references/ included, and records it in
skills-lock.json. Pick tools with --agent <name>, or add -g to install for your user instead
of the project. In Claude Code it skips the plugin's slash commands; run a skill directly as
/<name> instead. Each skill's evals/ folder comes along; agents don't load it.
Cursor, Codex CLI, Antigravity CLI, OpenCode: or clone anywhere, then symlink skills/ into
your project (~/awcameron-skills is just an example path):
git clone https://github.com/awcameron/skills.git ~/awcameron-skills
cd your-project
mkdir -p .agents && ln -s ~/awcameron-skills/skills .agents/skillsAfter installing, open your tool in the project and ask for something a skill matches ("review my changes"); the agent should name the skill. For global installs, tool-specific paths, and how to check it's working, see the per-tool guides: Claude Code, Cursor, Codex CLI, Antigravity CLI, OpenCode.
Or try it in a clone -- no install. The repo ships the symlinks each tool reads, so open your tool in the clone and the skills are already there:
git clone https://github.com/awcameron/skills.git
cd skills
# open Claude Code / Cursor / Codex CLI / Antigravity CLI / OpenCode hereRun a skill directly instead of waiting for it to trigger: /review-code in Claude Code and
Antigravity CLI, $review-code in Codex CLI, or / then the skill name in Cursor's Agent chat.
In OpenCode the agent loads skills itself, so there's no way to run one directly.
Update
Claude Code plugin -- refresh the marketplace, update the plugin, then restart Claude Code:
claude plugin marketplace update awcameron-skills # the marketplace
claude plugin update awcameron-skills # the plugin (same name)skills CLI installs: run npx skills update in the project (add -g for user-level installs).
Any clone (symlinked or tried in place): run git pull in the clone.
Each skill triggers on its own when a request matches it, and the "Stage" column says what part of
the work it acts on. docs/skill-categories.md explains the stages
and how skills hand off to each other.
| Skill | Use it when | Stage |
|---|---|---|
fullstack-feature-slice |
Building one feature across a monorepo's layers (shared contract, backend, frontend), following each layer's own conventions. | Feature building |
create-tdd |
Writing a technical design doc or RFC, greenfield or for existing code, from your template or the repo's own (with a bundled fallback). | Design documentation |
review-code |
Reviewing a diff or PR against the repo's own standards. Reports findings; doesn't fix them. | Code-quality lenses |
ts-best-practices |
Writing or reviewing TypeScript/JavaScript: the judgment calls a linter can't enforce. | Code-quality lenses |
zero-trust-architecture |
Touching auth, tenant isolation, or service-to-service calls: verify identity and ownership at every layer. | Code-quality lenses |
rollout-compatibility |
Changing a schema, API shape, or event contract that something else deploys against. | Code-quality lenses |
diagnose-bug |
Something fails, crashes, flakes, or is slow: find the confirmed root cause, evidence first, without fixing it. | The bug lifecycle |
fix-bug |
The root cause is confirmed: fix the cause, not the symptom, and verify it. | The bug lifecycle |
write-tests |
Adding tests with the repo's own runner and conventions, and proving each new test can fail. | Test authoring |
upgrade-dependency |
Bumping a dependency: read the changelog across the range, check real usage, compare tests against a baseline. | Dependency maintenance |
create-pr |
Taking local changes through the repo's branch, commit, push, and PR conventions, and cleaning up after merge. | Shipping workflow |
doc-fact-check |
Checking a doc's claims (including skill files) against the code and git history. | Documentation integrity |
format-docs |
Formatting Markdown with the repo's own formatter, and flagging structural changes instead of making them. | Documentation integrity |
choose-subagent |
Deciding which subagent type and model to spawn, based on whether the task writes anything. | Agent meta-behavior |
terse-reports |
You ask for terse, fact-dense status reports in this conversation. In Claude Code, the built-in Concise output style (/output-style concise) shortens every session's responses, less tersely than this skill. |
Agent meta-behavior |
There's no imposed spec→plan→build→ship pipeline and no fixed set of gates every change must
pass. Every skill here discovers a repo's own conventions -- its stack, file layout, naming
precedent, test setup -- rather than assuming or imposing them. Larger, more opinionated skill
packs bring their own process and vocabulary; this one is meant to disappear into whatever repo
it's dropped into. See
examples/skill-origin-case-study.md for the concrete
story behind a few of these skills, including a real skill-drift bug this repo's own
doc-fact-check skill was built to catch.
Needs Node ≥22:
npm ci
npm run validate && npm test && npm run eval- Repo conventions (branches, commits, PRs, releases): AGENTS.md, the canonical
contributor doc for humans and agents alike. Release steps are in
docs/releasing.md. - Adding or changing a skill:
docs/skill-anatomy.mdhas the frontmatter/structure rules and contributing checklist. Hold skills to the "discover, don't dictate" principle from Philosophy: read a repo's own conventions doc, check real git/GitHub history, and never hardcode a stack, file layout, or naming convention. - Security and conduct: SECURITY.md covers reporting a vulnerability or a skill whose instructions look designed to make an agent behave harmfully; CODE_OF_CONDUCT.md covers community standards.
MIT. See LICENSE.