This is a frontend design skill that tries to force Claude to break out of its statistical defaults. The author noticed LLMs tend to generate the same layouts repeatedly, so they built a system that simulates Python randomization to pick different hero architectures, typography stacks, and component patterns before generating code. It's opinionated about what makes designs look cheap versus expensive, bans certain fonts and patterns outright, and pushes for GSAP animations and tight spacing. If you've ever noticed Claude generating similar looking interfaces over and over, this attempts to solve that by literally forcing variety through simulated random selection. The directive is extremely detailed and prescriptive about grid math, motion, and asset usage.
npx skills add https://github.com/leonxlnx/taste-skill --skill gpt-taste