Radar
← All projectsThursday, August 27, 2026
Taste Skill
Anti-slop agent skills for frontend: layout, type, motion, and spacing rules plus image-gen boards so coding agents stop shipping generic UI.
Why it matters
Taste Skill is Leonxlnx’s MIT open pack of Agent Skills for premium frontends. The default skill is design-taste-frontend v2: brief inference, design-system mapping, variance/motion/density dials, GSAP skeletons, and a pre-flight checklist. The repo also ships style packs (brutalist, soft, minimalist), redesign and image-to-code skills, Stitch-compatible rules, and image-generation skills for web/mobile/brand reference boards that pair with ChatGPT Images then hand off to a coding agent.
Agent UIs still default to the same safe look. Taste Skill is a portable skill pack (not a component library) that forces a design language, hard checks, and redesign audits before code lands — install once, use in Claude Code, Codex, or Cursor.
How it works
Install the full pack with `npx skills add https://github.com/Leonxlnx/taste-skill`, or one skill by install name such as `design-taste-frontend`. Copy SKILL.md files into a project or paste into a session. Point agents at the skill when building or auditing UI; use the image-gen skills first for frames, then implement with Codex, Cursor, or Claude Code.
Not another UI kit or prompt dump. It is an installable skills.sh-compatible framework that upgrades agent behavior — taste constraints, audit modes, and image-first workflows — at ~80k stars of community pull without locking you into a paid design system.
Capabilities
- Public demo available