Agent Skills: Addy Osmani's Production AI Engineering Skills

TL;DR: addyosmani/agent-skills is Addy Osmani’s curated library of “production-grade engineering skills for AI coding agents”. It is JavaScript-based, MIT-licensed, and slots into Claude Code, Codex, Cursor, and Antigravity through their native skill discovery paths.

What it is

addyosmani/agent-skills carries the tagline “Production-grade engineering skills for AI coding agents.” Topics in the GitHub metadata are agent-skills, antigravity, claude-code, codex, and cursor, which gives a direct read of which runtimes the skills are written for. The repository has 94,836 stars, 10,064 forks, and 128 open issues — by volume this is one of the largest skills repositories on GitHub.

The language is JavaScript, the license is MIT, and the homepage is https://skills.addy.ie. Addy Osmani is the well-known Google engineering lead behind Lighthouse and a long-time writer on web performance, which gives the project a particular point of view: pragmatic, performance-aware, and oriented toward production codebases rather than toy examples.

What it does well

A “skill” in this context is a packaged set of instructions an AI coding agent loads to do a specific kind of work — review a pull request, write a commit message, scaffold a feature, run a benchmark. The interesting question is what agent-skills includes that you couldn’t trivially write yourself.

• Coverage of the major runtimes. Out of the box the repository publishes skills that work in Claude Code, Codex, Cursor, and Antigravity. You are not locked to one vendor’s skill format.
• Production posture. Skills are written for working codebases, not demos. Examples lean on existing patterns (test-first diffs, narrow refactors, structured reviews) rather than the more theatrical “build me an app” prompts.
• A single discovery surface. Each skill is structured the same way, so once you understand the schema you can navigate the rest of the library without reading every README.

How to try it

The pattern across the supported runtimes is roughly:

# Claude Code
git clone https://github.com/addyosmani/agent-skills.git
cp -R agent-skills/skills/* ~/.claude/skills/

# Codex / Cursor / Antigravity follow analogous install paths
# documented per skill.

After the skills are in place, the agent picks them up automatically. You invoke a skill by asking for it the way you would invoke any other agent command, or by selecting it from the runtime’s skill menu.

When it fits, and when it doesn’t

agent-skills is worth picking up if you already use one of the four supported runtimes and you want a starting library you can audit rather than a marketplace you have to trust. It is less useful if you prefer to write skills from scratch, or if your team has settled on a custom skill schema that does not match the conventions here.

Because the library is curated by one author, expect an opinionated point of view. Read the skills before you install them wholesale; some are intentionally stricter than your team might want.

Source and Accuracy Notes

• Project page: https://github.com/addyosmani/agent-skills
• Source repository: addyosmani/agent-skills (verified via GitHub API html_url)
• License: MIT (verified via GitHub API license.spdx_id)
• Stars at time of writing: 94,836
• Language: JavaScript
• Homepage: https://skills.addy.ie