Addy Osmani — a well-known engineer at Google — quietly published something that's already been starred by over 20,000 developers on GitHub. It's a collection of 20 structured instruction sets for AI coding tools. Each one teaches the AI how a careful, experienced engineer would approach a specific part of the work: planning a feature properly, writing tests, checking for security problems, and so on.
Here's the thing about AI coding tools: left to themselves, they take shortcuts. They'll write code that technically works but skips the steps that make it safe, maintainable, and ready for real users. It's like hiring a fast contractor who skips the foundation inspection.
These instruction sets — called SKILL.md files — sit inside the AI tool like a set of standing orders. When you ask it to build something, it follows a proper process instead of rushing to the finish line.
If you're working with a developer or an agency that uses AI to build your website, app, or internal tool, this is exactly the kind of thing you'd want them to be using. It's free, it's been validated by thousands of engineers, and it makes the AI's output more trustworthy.
Next time you're talking to a developer about AI-assisted work, ask them: "Are you using any kind of discipline layer to guide the AI's process?" That question alone will tell you a lot.
AI coding agent — An AI tool (like Claude or GitHub Copilot) that can write, review, or edit code on your behalf.
SKILL.md file — A plain text instruction set you give to an AI tool. It tells the AI what process to follow, step by step — like a recipe card for how to do the work properly.
Production-ready — Code that's safe, tested, and reliable enough to run in front of real customers. The opposite of a rough prototype.
Context — What the AI "knows" at the start of a task. Dropping a SKILL.md into the context is like handing someone a briefing document before a meeting.