Leonxlnx/taste-skill: 13 skills against generic AI design

What is Leonxlnx/taste-skill?

Leonxlnx/taste-skill is a Claude Code plugin built entirely out of skills: 13 of them, with no agents, commands, MCP servers, or hooks alongside them. Flowy files it under the Development category, and the repository carries 80,462 GitHub stars, one of the more heavily starred entries in the index. The repository's own pitch is blunt: stop an AI from generating output that looks boring and generic, and each of the 13 skills approaches that goal from a different angle, whether that is typography rules, a named aesthetic, or image generation instead of code.

Which skill should I reach for first?

taste-skill is the flagship, the one the repository is named after. It reads the brief, works out a design direction on its own, and aims to ship interfaces that do not look templated, building a real design system where one applies and running an audit first when the job is a redesign rather than a fresh build. taste-skill-v1 sits next to it as the original version, kept for projects that depend on its exact earlier behavior; its own description notes that the current default has since moved to a separate, experimental v2 skill elsewhere, so v1 is a compatibility choice rather than the recommended starting point. Two more skills work at the same level. gpt-tasteskill pairs UX and UI rules with GSAP animation: a strict AIDA page structure, wide editorial type that bans cramped multi-line wraps, gapless bento grids, and scroll-triggered pinning, stacking, and scrubbing. soft-skill skips the animation focus and sticks to the basics of an agency-built site instead: specific fonts, spacing, shadows, and card structures, with the common defaults that make an AI-built page look cheap blocked outright.

Is there a skill for a specific look, like minimalist or brutalist?

Yes, two of the thirteen skills each commit to one aesthetic rather than inferring one. minimalist-skill produces clean, editorial interfaces: a warm monochrome palette, typographic contrast instead of color, flat bento grids, and muted pastels, with gradients and heavy shadows ruled out. brutalist-skill goes the other way, pairing Swiss print typography with a military terminal look: rigid grids, extreme contrast between type sizes, a utilitarian color range, and analog degradation effects, aimed at data-heavy dashboards, portfolios, or editorial sites that want a raw, official feel rather than a soft one.

What if I am redesigning an existing site instead of starting fresh?

That is what redesign-skill is for. It audits whatever is already there, points out the generic patterns that mark it as AI-built, and applies the same high-end standards as the rest of the plugin without breaking what already works. It is built to work with any CSS framework, or with vanilla CSS, rather than assuming one specific stack.

Are any of these skills built for one specific tool?

Two are. stitch-skill targets Google Stitch: it generates agent-friendly DESIGN.md files that lock down typography, color, asymmetric layout, and constant micro-motion for that tool specifically. image-to-code-skill is written for Codex: on a visually important task it generates the design image itself first, studies it closely, and only then builds the site to match it, rather than jumping straight to markup.

What do the image-generation skills produce?

Three of the 13 skills generate images rather than code. imagegen-frontend-web produces design-reference images for a landing page, with a strict rule attached: one separate horizontal image per section, so an eight-section page produces eight images rather than one combined mockup. imagegen-frontend-mobile does the same job for iOS, Android, and cross-platform app screens, prioritizing clear hierarchy, readable text, and consistency across multiple screens in one flow. brandkit works a level above either: guideline boards, logo systems, identity decks, and full visual-world presentations, trained across styles from minimalist and cinematic to dark-tech, luxury, security, gaming, and developer-tool branding.

What does output-skill do, and why is it not a design skill?

output-skill is not about how a page looks; it is a fix for a model behavior problem instead. It overrides the truncation an LLM tends to fall into on a long generation, bans placeholder code in the output, and handles a token-limit split cleanly instead of leaving a file half-written. Pair it with any of the design skills above when the page or component you are generating is long enough that truncation is a real risk.

How do I install this, and does Flowy maintain it?

Flowy indexes this plugin from its GitHub source; it does not own, bundle, or host a separate copy of it. Install it from the listing page linked at the top of this guide, which resolves the current install step for this repository, and expect every skill inside to move on Leonxlnx's own release schedule rather than Flowy's.