A collection of specialized, high-quality, and objectively evaluated skills for AI agents. These skills follow the Agent Skills open standard managed by the skills CLI and are compatible with Claude Code, GitHub Copilot, Gemini CLI, Kiro, and other modern AI coding tools.
This repository features two powerful meta-skills designed to help you build and maintain your own agent capabilities:
- Skill Creator: A specialized skill for designing, testing, and objectively evaluating new agent capabilities. It enforces a rigorous cycle of intent capture, pattern discovery, and automated evaluation.
- Skill Evolver: A proactive skill that reflects on task execution to surgically refine other skills through automated friction capture and secure human-in-the-loop review.
The easiest way to install these skills across all your agents simultaneously is using the skills CLI:
# Interactively choose which skills to install from this repository
npx skills add https://github.com/grantmcd/skills
# OR explicitly install a specific skill directly
npx skills add https://github.com/grantmcd/skills --skill skill-evolver| Tool | Installation Path | Command / Prefix |
|---|---|---|
| Claude Code | ~/.config/claude-code/skills/ |
/ (e.g., /create-skill) |
| GitHub Copilot | ~/.copilot/skills/ |
/ (e.g., /evolve-skill) |
| Gemini CLI | ~/.gemini/skills/ |
Auto-detected via frontmatter |
| Kiro | ~/.kiro/skills/ |
Slash commands in panel |
| OpenAI Codex | ~/.codex/skills/ |
$ (e.g., $create-skill) |
Agent skills are more than just system prompts. They are modular, testable, and refined capabilities that embody four key ideals:
- Intuitiveness: Designed with clear triggers and imperative instructions that agents follow reliably.
- Objectivity: Verified against real-world test cases with automated assertions and visual diffs.
- Refined: Surgically updated based on actual friction points and session history.
- Secure: Built with mandates for credential protection and authenticated human-in-the-loop approval.
This repository uses a rigorous automated quality gate to ensure all skills and supporting scripts are production-ready.
- Python Linting/Formatting: Powered by Ruff.
- UI Formatting: Powered by Prettier.
- CI/CD: GitHub Actions runs a unified quality check on every push.
To run quality checks locally:
npm testTo fix formatting and linting issues separately:
npm run format
npm run lint:fixWe welcome community contributions! Please see our Contributing Guide for details on how to add new skills or improve existing ones.