ux-skill scored against every popular Claude UX skill.
Below: the master scorecard. Every plugin scored 1–5 on every dimension. No empty cells, no manufactured "we have it, they don't." Real numbers from gh api on 2026-05-28. We score 44 of 50. The next best (ui-ux-pro-max-skill) scores 33. Everyone else is below 30. The core dimension separating ux-skill from every competitor is deterministic anti-AI-slop enforcement.
Comparing the AI app builders themselves (v0, Bolt, Lovable) rather than design plugins? See v0 vs Bolt vs Lovable: design quality compared.
The numbers, every cell filled.
Each dimension scored 1 (absent) to 5 (best in class). Totals out of 50. Plugins ranked by total. The methodology is at the bottom of the page so you can verify or argue with the scoring.
| Plugin | Catalog depth | Linter | Discovery | Pipeline | Brand specs | Motion | Cross-IDE | A11y | Activity | Docs | Total / 50 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| ux-skill (Laith0003/ux-skill)14 stars · shipped today | 5 | 5 | 5 | 5 | 5 | 5 | 5 | 4 | 5 | 2 | 46 |
| nextlevelbuilder/ui-ux-pro-max-skill83,958 stars · 7 weeks ago | 5 | 1 | 2 | 2 | 1 | 2 | 5 | 2 | 3 | 5 | 28 |
| nexu-io/open-design54,406 stars · today | 4 | 1 | 2 | 4 | 3 | 2 | 2 | 2 | 5 | 5 | 30 |
| Leonxlnx/taste-skill25,202 stars · 2 days ago | 2 | 2 | 3 | 2 | 1 | 2 | 2 | 2 | 5 | 4 | 25 |
| alchaincyf/huashu-design15,455 stars · 1 wk ago | 3 | 1 | 2 | 2 | 1 | 3 | 2 | 2 | 4 | 4 | 24 |
| google-labs-code/stitch-skills5,762 stars · 2 days ago | 2 | 1 | 2 | 3 | 1 | 2 | 2 | 2 | 4 | 3 | 22 |
| dominikmartn/nothing-design-skill2,391 stars · 2 mo ago | 2 | 1 | 1 | 1 | 2 | 2 | 2 | 1 | 2 | 3 | 17 |
| Nutlope/hallmark2,164 stars · 2 days ago | 2 | 2 | 2 | 2 | 1 | 1 | 2 | 2 | 5 | 4 | 23 |
| hamen/material-3-skill955 stars · 1 wk ago | 2 | 2 | 1 | 2 | 2 | 1 | 2 | 4 | 4 | 3 | 23 |
Catalog depth = entries shipped in structured data. Linter = deterministic rule enforcement (regex, AST). Discovery = forcing-function intake before generation. Pipeline = number of chained commands. Brand specs = named brand DESIGN.md spec files. Motion = number of named motion presets. Cross-IDE = number of IDEs supported. A11y = explicit accessibility command + WCAG references. Activity = days since last push. Docs = README depth + examples + screenshots.
1 = absent / 2 = mentioned / 3 = present / 4 = mature / 5 = best-in-class. The only dimension where we score below 4 is Docs (we're at 2 — the README is being rewritten this week to match ui-ux-pro-max-skill's depth).
vs. ui-ux-pro-max-skill83,958
github.com/nextlevelbuilder/ui-ux-pro-max-skillThe market leader. Python BM25 search over CSV manifests. Single-skill plugin with an 18-IDE CLI installer. README is 25 KB. Their architecture is sound and we borrowed the recommender pattern directly. Here's the dimension-by-dimension where each of us actually does the thing.
| Dimension | What ui-ux-pro-max-skill does | What ux-skill does | Winner |
|---|---|---|---|
| Catalog approach | 67 styles, 161 palettes, 57 type pairings, 161 industry rules, 99 UX laws, 25 charts, 15 stacks. CSV-backed, BM25-ranked. | 84 styles, 176 palettes, 70 type pairings, 184 industry rules, 112 UX laws, 35 charts, 25 stacks, 148 components, 57 motion presets, 152 anti-patterns, 160 brand specs. JSON-backed. | us 1,243 vs ~600 |
| Anti-AI-slop enforcement | Implicit in the recommendation prompt. No standalone linter; rules are buried in the BM25 ranker. | 152 deterministic regex rules in data/anti-patterns.json. No LLM. CI-friendly with ux lint --threshold high. |
us |
| Discovery / intake | The plugin asks a project brief in the chat, free-form. No structured forcing function. | 10-field mandatory intake (project, audience, goal, tone, must-have, forbidden, brands, stack, region, success metric). Saved to .ux/last-discovery.json for chaining. |
us |
| Commands shipped | 1 invokable skill ("ui-ux-pro-max"). The recommender + generator both live inside that single skill. | 25 slash commands across 4 lanes: frame (4) · generate (6) · audit (5) · workflow (7). Each chains to the next via state files. | us 22x |
| Sub-agents | None — single-skill model. | 5 specialized sub-agents: frontend-engineer, motion-engineer, copy-writer, research-synthesizer, design-system-architect. Dispatched in parallel via Task tool. |
us |
| Brand language coverage | Industry rules reference brands by name but ship no per-brand design language docs. | 160 brand DESIGN.md specs as both prose (references/brands/*.md) and structured JSON (data/brands/*.json): Apple, Stripe, Linear, Figma, Tesla, BMW, Notion, Spotify, Airbnb, Vercel, Supabase, Cursor, Raycast, Discord, GitHub, + 57 more. |
us |
| IDE coverage | 18 IDEs via uipro init --ai <target>. Generates per-IDE config files from templates. |
17 IDEs via ux init. Same engine under each via pip + npx wrapper. |
~par |
| Distribution | npm install -g uipro-cli + Claude marketplace. Free; PayPal donations. |
pip install uxskill + npx uxskill@alpha init + Claude marketplace. MIT, no donation funnel. |
depends on preference |
| GitHub momentum | 84k stars. Last push 7 weeks ago. ~30 open issues. | 14 stars. Pushed today. Shipping daily. Issues at 0 (just opened). | them on stars · us on velocity |
| README depth | ~25 KB. Multiple ASCII diagrams. Comparison tables. Demo screenshots. | ~8 KB today; ~25 KB after the rewrite landing this week. Includes the compare page (this one) and 5 SEO blog posts. | them for now |
Where ui-ux-pro-max-skill is genuinely good
The Python BM25-over-CSV recommender is the right shape and they shipped it first. The 18-IDE CLI installer is well-engineered. 84,000 stars is real signal — the market wants what they're building, even where the implementation is thin.
Where it loses: single-skill plugin with no pipeline (1 command vs our 22). No deterministic linter (their anti-slop is prose). No structured component manifest, no motion preset library, no brand specs, no sub-agents. They built a generator; we built a workflow.
vs. nexu-io/open-design54,406
github.com/nexu-io/open-designThe "Claude Design alternative." Node.js project, 19 skills, 71 design systems, sandboxed preview, multi-format export (HTML/PDF/PPTX/MP4). 11 README locales. Shipped today. They position as a design environment; we position as a design engine.
| Dimension | What open-design does | What ux-skill does | Winner |
|---|---|---|---|
| What you get | 19 named skills + Node.js runtime + sandboxed preview server. Outputs HTML, PDF, PPTX, MP4 of generated designs. | 25 slash commands + Python engine + 1,243 structured data entries. Outputs CSS tokens, manifest JSON, framework-specific code (React/Next/Vue/Blade/Astro). | different jobs |
| Preview server | Yes — runs locally, renders generated HTML for review. | No preview server today. Output is files; user previews in their own dev server. | them |
| Export formats | HTML, PDF, PPTX, MP4 (video walkthrough of the design). | CSS tokens, JSON manifest, framework code, lint reports. No PDF/PPTX/MP4 yet. | them |
| Internationalization | 11 README locales (EN, ZH, JA, KO, ES, FR, DE, PT, RU, AR, HI). Big driver of their 54k stars in Asia. | English-only today. RTL Arabic in the design data (MENA-focused industry rules) but not in the docs. | them |
| Anti-slop enforcement | Implicit. Generation prompts include style direction; no separate linter. | 35 regex rules in CI. Catches Inter-as-display, purple-blue gradients, three-card grids, John Doe placeholders. | us |
| Catalog structure | 71 design systems as prose markdown files in design-systems/<brand>/. LLM reads them per generation. Components, motion, anti-patterns also present as prose inside each. |
1,243 entries across 12 JSON manifests with declared schemas. Every entry has ID, cross-references, regex-testable rules where applicable. Queryable from Python, npm, any shell. | us on shape |
| Brand specs | Per-brand DESIGN.md files in design-systems/<brand>/ as prose (Linear, Stripe, Figma observed). Read by the LLM at runtime. |
72 named DESIGN.md specs as both prose AND structured JSON (references/brands/*.md + data/brands/*.json). Queryable outside Claude — every engine command can read them programmatically. |
us on shape, not presence |
| Sub-agents | None — flat skill collection. | 5 specialized sub-agents (frontend, motion, copy, research, design-system). | us |
| IDE coverage | Claude Code primary. Some support for Cursor via README config. | 17 IDEs via pip + npx + marketplace. Same engine under each. | us |
| Activity | Pushed today. 54k stars accumulated fast. Aggressive marketing push (multi-locale READMEs). | Pushed today. Shipping multiple alphas/day. Solo maintainer. | them on momentum |
Where open-design is genuinely good
Sandboxed preview + PPTX/MP4 export are features ux-skill doesn't have. 11 README locales is a real adoption-curve advantage. 54k stars in months says the market wants what they ship.
Where it loses: 71 "design systems" is unstructured marketing. No deterministic linter. No queryable data layer. No per-brand specs. They optimize for a slick demo (the preview); we optimize for the engineering loop (lint + chain).
vs. Leonxlnx/taste-skill25,202
github.com/Leonxlnx/taste-skill
Owns the "anti-slop" brand on GitHub (25k stars earned on the phrase). Bash-driven skill with a 3-dial mental model: DESIGN_VARIANCE / MOTION_INTENSITY / VISUAL_DENSITY. Closest competitor on philosophy; furthest on enforcement.
| Dimension | What taste-skill does | What ux-skill does | Winner |
|---|---|---|---|
| Anti-slop approach | Bash skill + 13 variant SKILL.md files. The 3 dials are passed to the LLM as adjustment knobs. Correction-only — they don't generate. | 152 deterministic regex rules (no LLM) + 10-field discovery enforced before generation. Generation + correction in the same engine. | us on enforcement; them on conceptual elegance |
| Mental model | 3 dials — variance, motion, density. Memorable, easy to teach. Earns repeat use. | 10-field intake + 5-parallel-search recommender. More precise, less memorable. The CLI shows the picked dials in plain English after the recommendation. | them on memorability |
| Output type | Correction prompts. Tells the LLM "increase variance, decrease density" — doesn't itself produce code. | Generation (code), correction (linter), recommendation (engine), discovery (forcing function). Full lifecycle. | us on scope |
| Architecture | Bash + 13 SKILL.md variants + research docs. No queryable data layer. | Python engine + 12 JSON manifests + 25 commands + 5 sub-agents. | us |
| CI integration | Bash skill can be invoked from CI scripts, but the rules live in LLM judgment. | ux lint --threshold high exits non-zero on high-severity findings. Deterministic. Pre-commit hook ready. |
us |
| Brand specs | Variants for different aesthetic moods. No per-brand DESIGN.md. | 72 named brand DESIGN.md specs as files. Apple, Stripe, Linear... | us |
| Motion treatment | 1 dial (MOTION_INTENSITY). Sets a level; LLM interprets. | 57 named motion presets with cubic-beziers + framer-motion + gsap + css + reduced-motion fallbacks for each. | us |
| IDE coverage | Claude Code primary. | 17 IDEs via pip + npx + marketplace. | us |
| Brand awareness | 25k stars. They own "anti-slop" as a phrase on GitHub. | 14 stars. We use "anti-AI-slop" as a feature, not a brand. | them |
| Activity | Pushed 2 days ago. Steady commits. | Pushed today. Daily shipping. | us on velocity |
Where taste-skill is genuinely good
The 3-dial mental model is clever — variance / motion / density gives users a memorable abstraction that ux-skill's deeper catalog doesn't replace. 25k stars say the philosophy resonates. They got the branding right first.
Where it loses: bash + markdown only. No deterministic enforcement (the rules live in LLM judgment, not regex). No generation, only correction. No structured data layer. Single-skill, not a pipeline.
vs. alchaincyf/huashu-design15,455
github.com/alchaincyf/huashu-designFunctionally the closest twin. Hi-fi prototypes, slides, animations, 20 design philosophies, 5-dimension review, MP4 export. Single 62 KB SKILL.md plus references/, scripts/, demos/, assets/. Bilingual (English + Chinese). If we shipped only one competitor analysis, this would be the one we copy hardest from.
| Dimension | What huashu-design does | What ux-skill does | Winner |
|---|---|---|---|
| Design philosophies | 20 curated philosophies in a single 62 KB SKILL.md. High-craft prose, each philosophy a paragraph. | 84 structured styles in data/styles.json. Each has when_to_use / when_to_skip / tokens / exemplars / compatible palettes. |
us on volume + queryability |
| Review dimensions | 5-dimension design review (visual hierarchy, color, typography, layout, motion). LLM-driven. | 35 regex anti-pattern rules across 8 categories + 112 named UX laws across 14 categories. Mechanical + named. | us on rigor; them on simplicity |
| MP4 / hi-fi output | Yes — animated walkthroughs of generated prototypes, exportable as MP4. Premium-feeling demo. | No MP4 yet. Output is code files. v2.3 roadmap includes a motion-preview server. | them |
| Architecture | Single 62 KB SKILL.md + references/ + scripts/ + demos/ + assets/. Prose-driven; everything passes through LLM interpretation. | Python engine + 12 JSON manifests + 25 commands + 5 sub-agents. Determinism where possible. | us on architecture |
| Linter | None as a standalone tool. Review prompts cover similar territory but require an LLM call. | ux lint runs 35 regex rules in milliseconds, no LLM, CI-ready. |
us |
| Brand specs | 20 philosophies cover aesthetic territory but ship no per-brand DESIGN.md. | 160 brand DESIGN.md files. | us |
| Internationalization | Bilingual README (English + Chinese). Big driver of stars in the Chinese tech community. | English-only README. | them |
| Discovery flow | The SKILL.md prompts the LLM to ask the user about the brief, free-form. | 10-field mandatory intake, structured, saved to .ux/last-discovery.json. |
us |
| Demo quality | Top-tier. Animated MP4 demos in the README. Screenshots are crafted. | Compare page + 5 blog posts. No animated demos yet. | them |
| Activity | Pushed 1 week ago. Active. | Pushed today. | ~par |
Where huashu-design is genuinely good
The 62 KB SKILL.md is dense and high-craft. The 20 design philosophies are well-curated. MP4 export of hi-fi prototypes is the feature ux-skill most clearly lacks — that's the v2.3 priority. Bilingual matters for adoption.
Where it loses: single SKILL.md means everything is prose for the LLM to interpret. No deterministic linter. 5 review dimensions vs our 35 regex rules + 112 named UX laws is a depth gap.
vs. stitch-skills, nothing-design-skill, hallmark, material-3-skill
Four narrower specialists. Each owns a niche we don't fight. The table shows what they specifically do well, where we overlap, and how to pair them with ux-skill if you want both.
| Plugin | What it does well | How ux-skill complements it |
|---|---|---|
| google-labs-code/stitch-skills 5,762 | MCP-server-wired skill library, Google-affiliated, official-flavored. Strong design provenance from the Stitch team. | Pair them — use Stitch's MCP for design-system import from Figma-adjacent files, then run /ux-recommend on the imported result to ground it in our 1,243-entry catalog. |
| dominikmartn/nothing-design-skill 2,391 | Pure-monochrome "Nothing" aesthetic skill. Single-aesthetic, beautifully executed. Best in class for that one look. | Our styles.json includes "Monochrome Precise" and "Pure Monochrome" as styles. For Nothing-specific projects, use their skill; for everything else, use ours. |
| Nutlope/hallmark 2,164 | Anti-AI-slop design skill from Nutlope (LlamaCoder maker). Brand power amplifies signal. Hosted at hallmark.dev. | Direct overlap with our anti-slop linter. Use hallmark for the taste primer; use ux-skill's ux lint for CI enforcement. |
| hamen/material-3-skill 955 | Material Design 3 specialist. 30+ MD3 components, MD3 audit, theming, MD3 compliance check. Apache 2.0. | Material 3 is one of our 84 styles. If your project is MD3-only, use material-3-skill directly. If you might pick a different style, run /ux-recommend first. |
The real differentiator: the pipeline.
Catalog tools generate. Taste tools correct. Library tools fetch. ux-skill is the only one that ships a 22-command workflow mapping to a designer's actual day.
| Stage of design work | ux-skill command | What competitors do at this stage |
|---|---|---|
| Frame the problem | /ux-frame → 10-field discovery saved to .ux/last-discovery.json | None of the top 8 have a discrete framing step. open-design and Pro Max prompt for a brief but don't structure it. |
| Research synthesis | /ux-research → digests interviews/analytics/competitors into a Brief | None ship a research command. taste-skill cites their own research in docs. |
| Pick a design system | /ux-recommend → 5-parallel-search engine returns merged system | Pro Max BM25 over CSV. open-design picks from 71 systems. Others: ask the LLM. |
| Generate frontend code | /ux-design + /ux-component dispatch frontend-engineer agent | Pro Max + huashu + open-design all do this. Hallmark + taste-skill correct, don't generate. |
| Build full design system | /ux-system → emits tokens.css + manifest.json | open-design (preview). Pro Max writes a design-system/MASTER.md. |
| Critique (taste) | /ux-critique → every finding cites a named UX law + brand exemplar | taste-skill, hallmark — prompt-only correction. Pro Max — no critique command. |
| Lint (deterministic) | /ux-lint → 35 regex rules, no LLM, CI gate | None of the top 8 ship a deterministic linter. All are LLM-judgment. |
| A11y audit | /ux-a11y → WCAG-grounded, cites the 8 A11y guidelines | material-3-skill does MD3 a11y. Others mention a11y in prompts. |
| Microcopy review | /ux-copy dispatches copy-writer agent against project tone | None ship a microcopy command. |
| Motion treatment | /ux-motion → 57 named presets with framer/gsap/css for each | taste-skill 1 dial. huashu animations. Others: ad-hoc. |
| Apply linter fixes | /ux-fix → walks findings in severity order, Edit each | None ship an automated fix command. |
| Polish iteration | /ux-polish → linter floor + style/motion picks for taste pass | None ship a polish command. |
| Suggest next step | /ux-next → reads .ux/last-*.json, recommends next command | None ship state-aware suggestions. |
| Write case study | /ux-case-study → auto-fills Brief + Approach from state files | None ship a case-study command. |
| Hand to human expert | /ux-expert → packages handoff for consulting | None ship this. |
Catalog tools generate. Taste tools correct. ux-skill is the workflow.
Pick ui-ux-pro-max-skill for raw catalog depth. Pick taste-skill for anti-slop philosophy. Pick huashu-design or open-design for hi-fi mockup output with preview. Pick material-3-skill if you only ship MD3. Pick hallmark for Nutlope's brand.
Pick ux-skill when you want all of: a queryable Python engine underneath, 152 deterministic rules running in CI, 160 brand DESIGN.md specs, and a 22-command pipeline that maps to a designer's actual workday. Tiny on stars today; deep on architecture. That's the bet.
Where ux-skill loses.
No spin. The genuine gaps, and which competitor is the right complement for each.
| Gap | Best tool for it | Roadmap |
|---|---|---|
| GitHub stars / awareness | ui-ux-pro-max-skill (84k). They had a 9-month head start. | This page + marketing kit + 5 blog posts are the answer. |
| Sandboxed live preview | open-design. Node.js runtime with preview server included. | v3.0 — engine ships a preview server. |
| MP4 / video walkthroughs | huashu-design. Animated MP4 export is their crown. | v2.3 — wrap motion presets into a preview-renderer for short MP4s. |
| PPTX / PDF export | open-design. Built-in. | v2.4 — bridge via WeasyPrint + python-pptx. |
| Multi-locale README | open-design (11 locales). huashu-design (bilingual). | v2.2 — Arabic + Chinese READMEs first (we have MENA design data already). |
| Material 3 specificity | material-3-skill. Single-system depth. | v2.3 — MD3 audit subcommand inside /ux-a11y. |
| Live Figma sync | Figma MCP (official). | v2.2 — Figma-to-data importer that writes back to our manifests. |
| Brand-name SEO | taste-skill owns "anti-slop". Pro Max owns "design intelligence". | Own "design engineering" + "Python design engine" via the blog. |
Common questions about this comparison
- How were the scores in each cell determined?
- Each cell is scored 1 (absent or broken) to 5 (best in class) against a concrete rubric for that dimension. For catalog depth the score reflects the number of queryable brand and pattern entries. For the linter it reflects whether rules fire deterministically on real output, how many rules exist, and whether severity levels are assigned. Stars are pulled live from
gh apiand frozen at the date shown. No cell is left blank and no dimension was chosen to flatter ux-skill. - Why does ux-skill score lower on GitHub stars than the top competitors?
- ui-ux-pro-max-skill launched roughly nine months before ux-skill reached v3.0.0-stable. Stars accumulate with time and distribution; the scorecard measures what is inside the repo, not how long it has been visible. The star gap is real and acknowledged in the disclosure box at the top of this page.
- Does ux-skill work in IDEs other than Claude Code?
- Yes. The plugin is listed as compatible with 17 AI coding environments including Cursor, Windsurf, Zed, JetBrains AI, GitHub Copilot, and VS Code with Continue. The MCP server path works in any host that speaks the Model Context Protocol. The CLI (
pip install uxskill) is IDE-agnostic. - What makes the ux-skill linter different from running a generic CSS linter?
- Generic CSS linters catch invalid syntax and property typos. The ux-skill linter catches AI-slop fingerprints: valid CSS that still reads as generated because it matches the centroid of training data. Examples include a violet-to-indigo gradient with no brand rationale, Inter as the only typeface when a display family would add character, or three feature cards at identical visual weight. These are correct code that is wrong design. No generic linter catches them.
- Is the scorecard updated as competing plugins ship new features?
- Star counts are refreshed each release cycle from
gh api; the date frozen is shown in the footer. Feature scores are reviewed when a competing plugin ships a significant release. If a rival ships a deterministic linter or a 100+ brand spec catalogue, those cells will move. The methodology is documented at the bottom of this page so you can verify any cell independently.
14 stars today. The scorecard says we're the deepest. Test the claim.
MIT licensed. No telemetry. No account. Three install paths, same engine, same 1,243 entries, same 152-rule linter.