Agent-Skills by Addy Osmani: Production-Grade Engineering Workflows for AI Coding Agents

Production-Grade Engineering Skills for AI Coding Agents
https://github.com/addyosmani/agent-skills

If you’ve ever worked with an AI coding agent, you’ve probably felt this. It writes code fast. Impressively fast. But sometimes it cuts corners. No tests. No clear spec. Security checks… forgotten. It’s like working with a brilliant junior developer who’s always in a hurry.

That’s exactly the problem Addy Osmani’s “agent-skills” project is solving.

At its core, this repository packages production-grade engineering workflows into structured “skills” that AI agents can follow. Not vague advice. Not fluffy prompts. Real, opinionated processes that senior engineers actually use when shipping software that needs to survive in the real world.

And it’s thoughtfully designed.

There are 7 slash commands mapped to the development lifecycle. When triggered, they activate the right skills automatically. Designing an API? The api-and-interface-design skill kicks in. Building UI? You get frontend engineering best practices. It’s contextual, which makes it powerful.

Under the hood, there are 19 structured skills, each with clear steps, verification gates, and even anti-rationalization checks. I love that part. Because let’s be honest, both humans and AI tend to justify shortcuts when deadlines loom.

What makes this special is how deeply it embeds real engineering culture. You’ll see concepts like Hyrum’s Law in API design, the Test Pyramid, trunk-based development, feature flags, and even Chesterton’s Fence. These aren’t just mentioned. They’re operationalized. Built directly into the workflow an agent must follow.

Installation is flexible too. You can integrate it with Cursor, Gemini, Windsurf, Copilot, or basically any agent that accepts system prompts. The skills are plain Markdown, which keeps things simple and portable.

If AI agents are going to be part of modern engineering teams, they can’t just optimize for speed. They need discipline. Process. Judgment.

This project moves us closer to that future, one structured workflow at a time. And honestly, it feels like the kind of foundation we’ll look back on and say, yes… this is when AI coding started growing up.