/learn-taste turns design corrections into rules for your next agent

Emil Kowalski added /learn-taste to Skills for Designers and Engineers. The new skill collects every design correction you made during a session and turns each one into a strict rule with reasoning, so the next agent doesn't make the same mistake.

It finds corrections you never spelled out too, such as edits you made to the agent's files, repeated nudges and reverted changes. A fix on one component becomes a rule for every similar one. It never makes up a reason. When you didn't say why, it proposes one and asks you to confirm it.

The rules go into DESIGN.md, or the skill suggests a lint rule when a check would catch the mistake more reliably. In the skill page's example, a session without it ends with vague rules and a missed duration edit. With it, each rule has a reason, including a duration change the user made in dropdown.tsx.

Run it at the end of a session where you corrected the agent more than once, with a prompt like /learn-taste go through this session and pull out everything I corrected. Other example prompts have it look at your last 3 sessions, or prune the rules when DESIGN.md is getting long. It pairs with the Design Engineering and Prototyping skills.

Install it with npx skills@latest add emilkowalski/skills --skill learn-taste. It's free, as part of the MIT-licensed repository.