agent-skills

Agent skills for Claude Code, Codex and other AI coding agents, from Win Metta. Each skill is a folder with a SKILL.md in the open Agent Skills format, written so one copy works in every agent.

Skills

Skill What it does

skill-maintainer Creates, updates, improves, tests, renames and checks skills so they work in both Claude Code and Codex from one copy. The real files go in .agents/skills/<name>/ and .claude/skills/<name> is a symlink to them. It also writes a Codex agents/openai.yaml, and guides writing, testing against a baseline, grading and description tuning. Includes a scaffolder, a checker and an eval summarizer (Node 20+, no dependencies).

pr-description Writes or updates a pull request or merge request title and description, on any hosting platform, with the sections Why?, Impact, Technical Impact, Changelog (Conventional Commits), What changed (collapsed), Validation, Dev Test, References, Not included / follow-ups and Checklist; only Why?, Changelog and Validation are required. It reports only checks that were actually run, keeps what people added when updating, and includes a template, an example and a checker for the structure (Node 20+, no dependencies).

Skill	What it does
`skill-maintainer`	Creates, updates, improves, tests, renames and checks skills so they work in both Claude Code and Codex from one copy. The real files go in `.agents/skills/<name>/` and `.claude/skills/<name>` is a symlink to them. It also writes a Codex `agents/openai.yaml`, and guides writing, testing against a baseline, grading and description tuning. Includes a scaffolder, a checker and an eval summarizer (Node 20+, no dependencies).
`pr-description`	Writes or updates a pull request or merge request title and description, on any hosting platform, with the sections Why?, Impact, Technical Impact, Changelog (Conventional Commits), What changed (collapsed), Validation, Dev Test, References, Not included / follow-ups and Checklist; only Why?, Changelog and Validation are required. It reports only checks that were actually run, keeps what people added when updating, and includes a template, an example and a checker for the structure (Node 20+, no dependencies).

Install

Pick one. Review a skill before you install it: skills run with your agent's permissions.

Into a project, for every agent (`npx skills`)

npx skills@latest add winmetta/agent-skills --skill skill-maintainer -a claude-code -a codex

This writes one copy to .agents/skills/skill-maintainer/ and links .claude/skills/skill-maintainer to it; commit both. Leave out --skill to choose from the list. Add -g to install for your user instead of the project. Update with npx skills update. This uses vercel-labs/skills.

Claude Code plugin

/plugin marketplace add winmetta/agent-skills
/plugin install agent-skills@winmetta

Skills from a plugin are namespaced, for example /agent-skills:skill-maintainer.

Codex plugin

codex plugin marketplace add winmetta/agent-skills

Then install agent-skills from the plugin list. This channel follows OpenAI's plugin docs but hasn't been checked with the Codex CLI yet; if it fails, use npx skills above, which supports Codex.

Use

Claude Code: ask for what you need ("create a skill that …"), or run /skill-maintainer.
Codex: ask, or mention $skill-maintainer. /skills lists installed skills.

To wire the skill checks into your own repository's hooks or CI, see references/repo-setup.md.

Contributing

Issues and pull requests are welcome. See AGENTS.md for the layout, the rules for portable skills and how releases work. Run npm test and npm run check before opening a PR.

Licence

MIT. See LICENSE.