addyosmani/agent-skills: The Best Free AI Coding Tool on GitHub

Key Takeaways

• addyosmani/agent-skills has 98k stars on GitHub
• 9 slash commands that map to the dev lifecycle
• Works with Claude Code, Codex, Cursor, and any AI coding agent
• MIT licensed — completely free
• Created by Addy Osmani, ex-Google Chrome team, now at Anthropic

What Is This

addyosmani/agent-skills is a free open source GitHub project that provides production-grade engineering skills for AI coding tools. It has 98.9k stars and 10.4k forks. MIT licensed. Totally free.

The creator, Addy Osmani, spent 14 years at Google leading Chrome DevEx. He worked on DevTools, Lighthouse, and Core Web Vitals. He's now at Anthropic working on Claude Code. When someone like Addy builds a tool for developers, you should pay attention.

The repo packages skills that encode the workflows, quality gates, and best practices that senior engineers use when building software. These ones are packaged so AI agents follow them consistently across every phase of development.

The 9 Commands

The repo provides 9 slash commands that map to the dev lifecycle. Each command activates the right skills automatically. Here's what each one does:

• /spec — Define what to build. Spec before code. This is the most important command. Most devs jump straight into coding without defining what they're building. The /spec command makes you write a clear spec before writing any code. It includes acceptance criteria, constraints, and success metrics.
• /plan — Plan how to build it. Small, atomic tasks. Once you know what you're building, the /plan command breaks it down into small, manageable tasks. Each task is atomic — it can be completed independently. This makes it easier to track progress and catch issues early.
• /build — Build incrementally. One slice at a time. The /build command implements the plan incrementally. You don't build everything at once. You build one slice, test it, and move to the next. This reduces risk and makes it easier to iterate based on feedback.
• /test — Prove it works. Tests are proof. The /test command ensures the code actually works. Not just "it compiles" — it passes the tests. Tests are proof that the code does what it's supposed to do. No tests, no merge.
• /constraints — Set the quality bar. Decide it once, enforce it everywhere. The /constraints command sets the quality standards for the project. Coding conventions, performance targets, security requirements. Decide these once, enforce them everywhere. No exceptions.
• /review — Review before merge. Improve code health. The /review command catches issues before they merge. Code style, potential bugs, performance issues. Review before merge. This is where senior engineers add the most value — they catch things AI agents miss.
• /webperf — Audit web performance. Measure before you optimize. The /webperf command audits web performance. It measures before optimizing. You can't improve what you don't measure. This command makes sure the code isn't just correct — it's fast.
• /code-simplify — Simplify the code. Clarity over cleverness. The /code-simplify command simplifies the code. Clever code is hard to maintain. Clear code is easy to maintain. This command removes unnecessary complexity and makes the code readable.
• /ship — Ship to production. Faster is safer. The /ship command ships the code to production. Faster is safer — the longer code sits in a branch, the more likely it is to have conflicts or issues. Ship early, ship often.

## Why It Matters

AI coding tools are getting popular fast. Claude Code, GitHub Copilot, Cursor — they all help developers write code faster. But they don't always follow best practices. They skip tests. They ignore performance. They ship code that works but isn't maintainable.

Agent-skills fixes this. It forces AI agents to follow structured workflows. Every phase of development has a command. Every command has a clear purpose. No guessing, no shortcuts.

The Anthropic 2026 Agentic Coding Trends Report calls this "repository intelligence" — AI that grasps not just lines of code but the relationships and intent behind them. Agent-skills gives agents that intelligence.

My Honest Take

I've been using AI coding tools for a while. They're powerful but inconsistent. Some sessions produce great code. Others produce code that works but is hard to maintain.

Agent-skills makes them more consistent. The /spec command alone is worth it — it forces you to define what you're building before writing a single line. Most developers skip this step. Agent-skills makes it mandatory.

The /review command is also solid. It catches issues before they merge. The /test command ensures tests actually pass, not just exist. The /webperf command makes sure the code isn't just correct — it's fast.

The /review command is also solid. It catches issues before they merge. The /test command ensures tests actually pass, not just exist.

Who This Is For

• Developers using AI coding tools daily
• Teams that want consistent code quality
• Open source contributors who want to review code faster
• Anyone tired of AI agents skipping best practices
• Engineering managers who want to enforce coding standards
• Students learning to code — agent-skills teaches good habits from day one
• Teams that want consistent code quality

Common Questions

Q: Does this work with any AI coding agent?

A: Yes. Agent-skills works with any AI coding agent that accepts Markdown instructions. Claude Code, Codex, Cursor, and others all support it.

Q: Is it really free?

A: Yes. MIT licensed. No subscription. No hidden fees. Just clone the repo and use it.

## Bottom Line

addyosmani/agent-skills is one of the best free open source tools for AI coding tools. 98k stars doesn't lie. The 9 commands cover the entire dev lifecycle. It's free, open source, and actually useful.

If you use AI coding tools, install this. It's the best $0 you'll spend all year. The /spec command alone will change how you approach coding tasks. The /review command will catch issues you'd miss. The /ship command will get your code to production faster.

Addy Osmani built this for developers. He's been in the trenches — Chrome DevTools, Lighthouse, Core Web Vitals. He knows what developers need. And this is it.

If you use AI coding tools, install this. It's the best $0 you'll spend all year.