leonxlnx/taste-skill
14 skills1M installs531k starsGitHub
Install
npx skills add https://github.com/leonxlnx/taste-skillSkills in this repo
1Design Taste FrontendSenior UI/UX frontend skill that overrides default LLM design bias with tunable visual rules, strict React/Next architecture, and hardware-aware CSS patterns.124kinstalls2High End Visual DesignHigh-End Visual Design is an agent skill that acts as a principal UI/UX and motion choreographer, pushing generated interfaces toward Awwwards-tier polish instead of interchangeable AI aesthetics. Solo builders use it when they need landing pages, dashboards, or marketing sections that feel expensive—clear typography, intentional whitespace, layered shadows, and fluid animation—without manually rejecting every generic font and icon choice. The skill encodes strict anti-patterns (banned system fonts, stock icon sets, harsh drop shadows) and a variance rule so consecutive designs do not clone the same layout while still sharing a coherent premium language. It fits early Build work on frontend markup and styling, and equally supports Validate prototypes or Launch distribution pages where first impressions drive conversion. It does not replace user research, brand strategy, or accessibility audits; it elevates execution quality once you know what you are building.103kinstalls3Redesign Existing Projectsredesign-existing-projects is a design-upgrade agent skill for solo builders who already have a website or app that works but looks like generic AI output. Instead of rebuilding from scratch, the agent reads the project, identifies the styling stack, runs a structured design audit, and applies incremental fixes that raise perceived quality. The audit emphasizes typography (distinct fonts, tighter display type, readable measure, monospaced figures where needed), hierarchy, and missing UI states that make interfaces feel unfinished. The skill explicitly allows any common CSS approach so Cursor or Claude Code users can drop it into mature repos during a polish sprint. It is most valuable when you are preparing a public-facing product for validation demos, launch pages, or growth content where trust and taste matter, but you cannot afford a full design system rewrite. Outcomes are a prioritized problem list and concrete CSS or component tweaks aligned with high-end product aesthetics.101kinstalls4Minimalist Uiminimalist-ui is a design protocol skill for indie builders who want interfaces that feel like a refined workspace document—not another gradient-heavy SaaS template. The agent follows Premium Utilitarian Minimalism rules: warm monochrome bases, strong typographic contrast, generous macro-whitespace, and flat bento grids with sparing muted pastels. A hard ban list rejects ubiquitous typefaces, thin generic icon sets, heavy drop shadows, primary-colored hero slabs, gradients, neon, and flashy glass effects. The intent is to push coding agents toward bespoke editorial layouts comparable to high-end productivity products while staying implementable in modern frontend stacks. Invoke it when you are implementing marketing pages, dashboards, or app shells where taste and restraint matter as much as feature completeness, and you want the agent to self-police against default Tailwind aesthetics.94.8kinstalls5Full Output EnforcementFull-Output Enforcement is an agent skill that overrides default LLM brevity habits so solo builders get paste-ready, production-grade artifacts every time. It sets a production-critical baseline: if you asked for a full file, five components, or an unabridged implementation, the agent must deliver all of it with no placeholder comments, no “rest follows the same pattern” hand-waving, and no offers to continue in a follow-up message. The skill documents explicit banned patterns in both code blocks and prose, plus structural failures like shipping skeletons when a full build was requested. An execution process starts with scoping—count distinct deliverables—then drives complete generation, with guidance for splitting cleanly at real token limits rather than faking completeness. Indie developers using Claude Code, Cursor, or Codex install it when agent replies keep truncating mid-file or summarizing repeated logic instead of writing it. It is procedural meta-knowledge, not an MCP server or external API; you invoke it on any task where incomplete output would block shipping.91.4kinstalls6Industrial Brutalist UiIndustrial Brutalist UI is an agent skill for solo builders who want interfaces that feel like Swiss print systems fused with military or aerospace terminals—not another soft SaaS skin. It documents how to pick one visual archetype per project, enforce high-contrast grids and typography, use purely functional color, and simulate physical media artifacts in the browser. You reach for it when a portfolio, internal dashboard, or editorial product needs authority, density, and anti-generic character without wandering into decorative UI trends. The skill is procedural design knowledge rather than a component library: it tells your coding agent what constraints and motifs to honor while implementing layouts, type scales, and effects. Intermediate-to-advanced teams benefit most because the aesthetic demands disciplined hierarchy and consistent commitment to a single mode. Pair it with your stack’s CSS or design-token workflow; it does not replace user research or accessibility review.88.2kinstalls7Stitch Design TasteStitch Design Taste is a semantic design-system skill for solo builders using Google Stitch to generate UI screens without sliding into template-looking output. It walks your agent through producing a DESIGN.md that states visual atmosphere, hex-calibrated palettes, typographic hierarchy, component behaviors, and motion rules in natural language Stitch can interpret. Use it when you already have access to Stitch and want every new screen prompt anchored to the same premium language—especially if you are iterating a SaaS or mobile flow and tired of resetting taste on each generation. The skill is a generator for documentation artifacts, not a replacement for Stitch itself; optional MCP wiring helps programmatic loops in Cursor-class editors. It pairs naturally with validate-phase prototyping when you are proving look-and-feel before full build, then stays relevant as you add screens during build.87.8kinstalls8Gpt Tastegpt-taste is an elite frontend design-engineering skill for solo builders who want Awwwards-level landing and marketing pages without the statistical biases typical LLM UIs produce. Before writing interface code, the agent must run a deterministic Python-style randomization step in a design plan to pick hero architecture, typography stack, and layout variants so pages do not collapse into the same narrow-container, six-line headline pattern. The skill encodes hard rules for AIDA structure, editorial type scale, gapless bento grids, inline micro-imagery, and large vertical rhythm, then drives motion through GSAP ScrollTriggers with pinning, stacking, and scrubbed timelines. It is aimed at builders shipping branded SaaS or content sites in Claude Code, Cursor, or Codex who already have copy or a brief and need production-grade layout and motion—not a generic component library dump.79kinstalls9Image To CodeImage-to-code is a taste-forward agent skill aimed at Codex users who refuse generic landing-page output. It reframes the job as art direction plus implementation: for visually important web work, the agent must create design images itself, study them carefully, and only then build HTML and CSS that hews closely to those references. The readme calls out failure modes of typical AI sites—single oversized compressed mockups, illegible type in screenshots, dark centered hero tropes, card nesting, and weak spacing—and replaces them with section-scoped, high-resolution references and clean hero regions that stay readable on a small laptop. Scope includes hero blocks, full landing pages, startup marketing sites, editorial brand pages, product pages, portfolios, and multi-section redesigns. It is advanced in practice because it couples generative visual iteration with strict layout discipline. Solo builders use it when the product’s credibility depends on looking designed, not assembled. It assumes you will accept multiple image-generation passes for fidelity rather than one-shot codegen from a vague prompt.67.3kinstalls10BrandkitBrandkit is a Prism agent skill for solo founders and indie builders who need brand identity visuals that read like a studio deliverable, not a stock AI collage. Invoke it when you are shaping how a product should feel—minimalist, cinematic, editorial, dark-tech, luxury, or developer-tool—and you want a single high-end image that communicates logo logic, typography restraint, UI application mockups, and atmospheric art direction together. The skill positions the agent as art director: it prioritizes intentional logo concepting, strong symbolic meaning, flexible grid layouts, browser or terminal framing where appropriate, and presentation-ready density. It is optimized for brand-guidelines boards with charcoal outer fields, clean gutters, large negative space, and image-led storytelling rather than wordy slides. Use during validation when pairing positioning with visual proof, or during build when aligning UI chrome with an identity system. It does not replace a full trademark search or vector handoff workflow, but it compresses weeks of moodboarding into one strategic composite for agents that generate images.66kinstalls11Imagegen Frontend WebImagegen Frontend Web is an elite image-direction agent skill for solo builders who need website design references, not vague mood boards. It forces strict output discipline: every section gets its own horizontal image, announced sequentially when the agent can only render one at a time, defaulting to six sections for a generic landing page or eight for a full website template. The skill pushes conversion-aware layout variety—mixed compositions, flexible heroes, second-read moments, and one palette locked across the set—so coding models and developers can translate visuals into HTML/CSS without fighting composite mockups. Use while validating positioning on a landing page, scoping a marketing site redesign, or briefing frontend build with section-accurate references. It does not replace component libraries or accessibility review; it produces directional art comps that accelerate Validate and Build frontend work.65.1kinstalls12Imagegen Frontend MobileImagegen Frontend Mobile is an elite mobile product art-direction skill for solo founders and designers who need believable, non-generic app screen images before writing Swift, Kotlin, or React Native. The agent acts as a creative director: strong hierarchy, comfortably readable text, multi-screen consistency, image-led composition, and tasteful mockup framing so stakeholders see app-native flows rather than flat wireframes. It spans common product surfaces from onboarding and auth through dashboards, chat, ecommerce, and fintech, prioritizing premium texture and intentional color over stock UI clichés. The skill generates images only—it does not implement frontends or convert mocks to code—so it best supports pitch decks, App Store preview exploration, and visual validation while scope is still flexible.63.5kinstalls13Design Taste Frontend V1design-taste-frontend-v1 is the preserved first-generation taste-skill for solo builders shipping web frontends who depend on predictable, high-agency UI output. It treats design as parameterized craft: variance, motion, and density baselines steer how symmetric, cinematic, or data-dense the interface feels, while a mandatory dependency gate stops the agent from hallucinating packages that are not in your `package.json`. Unless you specify another stack, the skill follows its default architecture conventions so generations stay consistent across sessions. Prism lists this install name for teams pinned to v1 semantics; new projects should prefer `design-taste-frontend` (v2) unless a regression or style lock requires the exact v1 checklist and section flow documented in the skill body.30.8kinstalls14Image Taste FrontendImage Taste Frontend is an elite image-to-code agent skill for solo builders who care how their marketing site looks on a real laptop, not just in a chat preview. Documented for Codex, it enforces an image-first pipeline: for visually important web work, the agent must create implementation-friendly reference images, analyze typography, spacing, and hierarchy deeply, then build frontend that tracks those references closely. It fights default AI UI collapse—single compressed mockups, unreadable text, repetitive left-text/right-image blocks, and nested card nesting. Heroes stay clean, spacious, and legible at small laptop widths. Use it when rebuilding a startup landing page, portfolio, or product marketing surface where brand taste beats generic components. Intermediate to advanced: you need a frontend stack and willingness to iterate on generated assets. The outcome is section-level design fidelity and a site that reads as art-directed rather than templated, with explicit guidance to regenerate images instead of under-generating or reusing cropped scraps.3.6kinstalls