Taste Skill

Taste Skill is an open-source collection of SKILL.md files that plug into AI coding agents—Cursor, Claude Code, Codex, Gemini CLI, v0, Lovable, and more—to enforce strict frontend design rules and stop generic-looking output. By providing structured constraints that ban common AI-generated patterns, enforce component variants, and require dark mode parity, it guides agents to produce interfaces that look bespoke rather than templated. The v2 (experimental) rewrite, shipped in May 2026, introduces brief inference: the agent reads industry, audience, and mood before generating, then maps to appropriate design systems (Material, shadcn, Tailwind, Radix, Fluent, Carbon, etc.). New rules include hero discipline (max 2-line headline, 20-word subtext), layout variety (≥4 different section families across 8 sections), a dark mode protocol with WCAG AA contrast, and a hard pre-flight check that halts output if any rule fails. The changelog also notes animation skeletons using GSAP (sticky-stack, horizontal-pan, scroll-reveal) and expanded bans like em-dashes in headings and version badges. Taste Skill includes a library of 13 skills: taste-skill (default v2), v1 legacy, GPT/Codex variant, image-to-code, redesign, soft, output, minimalist, brutalist, stitch, and imagegen skills for web, mobile, and brand assets. It is sponsored by Vercel's Open Source Program, MIT-licensed, and installed via a single npm command. Compared to alternatives like Lintrul or manual design system enforcement, Taste Skill is free, community-driven, and instantly integrable with any SKILL.md–compatible agent, but lacks a visual GUI or managed service. Positioned as a lightweight, developer-tooling approach to taming AI-generated UI, it is ideal for agentic frontend work where design consistency matters but a full design system is overkill.