This is an opinionated design system prompt that tries to fix common LLM UI design patterns by forcing randomization and variety. It simulates Python's random.choice() using deterministic seeds to prevent Claude from always defaulting to the same layouts, then enforces rules around typography (specific font stacks, never Inter), GSAP animations, and grid layouts. The core insight is interesting: LLMs do have statistical biases that make them repeat the same hero sections and bento grids. Whether you agree with the aggressive tone and strict rules is another question, but if you're tired of generic Tailwind layouts with "SECTION 01" labels everywhere, this gives you a framework to push for more varied output. Just be ready for a very specific design philosophy.
npx -y skills add nexu-io/open-design --skill gpt-taste --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
leonxlnx/taste-skill
supercent-io/skills-template
supercent-io/skills-template