If you've used any of the AI tools that build websites for you — or hired someone who does — you've probably seen the result. Purple buttons. A big centered headline. That same clean-but-forgettable font. It all looks like it came from the same factory, because in a way, it did. The AI learned from millions of websites and regurgitated the average.
A developer called Leonxlnx got fed up and built something small and clever: a file called taste-skill. You drop it into your project, and it gives the AI a set of firm opinions about design — things it should never do, and things it should aim for instead. No neon gradients. No centered hero. Considered spacing. Thoughtful typography. The kind of quiet confidence that separates a well-designed thing from a generic one.
Nearly ten thousand developers starred it on GitHub this week. The comment that keeps showing up: this is the solution to the biggest problem with AI-built interfaces.
More and more small studios and agencies are using AI tools to build and iterate on websites faster. That's fine — but speed without taste just produces more noise. This is an early sign that the community is starting to care about what gets built, not just how fast.
If your team uses any of these tools, it's worth asking: what rules are guiding the output?
AI coding agent — A tool that writes and builds software on your behalf, like an intern who types very fast but needs clear instructions.
SKILL.md — A plain text file that tells an AI agent how to behave in a specific situation. Think of it as a house style guide, but for a robot.
Open-source — Software that anyone can read, copy, and improve for free. This project is open-source, meaning thousands of developers can refine it over time.
Vibe coding — The informal term for building software quickly with AI tools, often by describing what you want in plain language rather than writing code yourself.