Labsco
Labsco/Repos/leonxlnx/taste-skill
REPO PACKAGE

leonxlnx taste-skill

COMMUNITY
leonxlnx · publisher57,762 repository starsMIT · Freegithub.com/leonxlnx/taste-skill
13skills
5groups
11ready to use

Ten of the thirteen write code or files directly: a default frontend-design skill (now a substantial v2 rewrite, with the old v1 preserved separately for anyone who depends on its exact prior behavior), five narrower visual-style presets — minimalist, industrial-brutalist, a GPT/Codex-tuned variant, a "high-end agency" look, and a Google Stitch design-system generator — a redesign-audit skill for existing codebases, and a skill whose only job is stopping the model from truncating output. The other three only generate reference images — website comps, mobile-app screens, and brand-identity boards — for a separate coding agent to build from afterward.

Eleven of the thirteen need nothing beyond installing them. Two carry a real dependency — one a local tool, one an MCP service — but the material available here does not say which two, so that is worth checking before relying on either.

READ THE FULL ANALYSIS

The v1/v2 split is a real seam. The default install name did not change between versions, so upgrading silently swaps in a rewrite the repository itself still calls experimental; anyone who needs the old, stable behavior has to know to ask for design-taste-frontend-v1 by name instead. Several of the visual-style presets are also close enough in scope — minimalist, industrial-brutalist, the "high-end" preset — that choosing between them means reading all three descriptions rather than trusting the names alone.

Who this is for. Teams generating UI or marketing visuals with an AI agent who are tired of the same purple-gradient, card-in-a-card default look, and who are willing to try a few stylistic presets to find the one that fits their brand. If what you want is one opinionated design skill rather than a style catalogue, a single preset plus the image-to-code skill is closer to it than installing all thirteen at once.

13Grouped into 5 sets by the authors.
11Work with nothing else to set up.
57,762Stars on the GitHub repository, at last check.