leonxlnx/taste-skill

Taste Skill is a collection of "agent skills", reusable instruction files you give to AI coding tools to improve the quality of the front-end interfaces they produce. The core problem it addresses is that AI-generated UIs often look generic: boring layouts, poor spacing, weak typography, and no visual personality. Taste Skill gives your AI agent a set of design rules to follow so the resulting code produces something that looks more carefully considered. The skills are plain text files (formatted as SKILL.md) that you install once and then reference in your AI agent sessions with tools like Codex, Cursor, or Claude Code. You install them with a single command using npx, a JavaScript package runner. Each skill does one specific job. The default skill (called design-taste-frontend) is a general-purpose upgrade for any frontend. Others are more targeted: one focuses on minimalist editorial UIs, another on a hard industrial "brutalist" visual style, one on auditing and fixing an existing project's design, and one that enforces full code output when an agent keeps producing incomplete responses. There are also image-generation skills for producing visual reference boards, website mockups, mobile screen flows, and brand identity kits, using image-generating AI tools. The idea is you generate the design images first, then hand them to a coding agent to implement. The taste-skill file includes numeric settings (1-10 dials) for layout experimentation, animation depth, and information density, so you can tune how expressive or restrained the AI's design output is. The project is MIT-licensed.