Deep Dive: ui-ux-pro-max-skill — The Repo That Outranks taste-skill

The author originally assumed taste-skill was #1 in the anti-slop frontend skill race. It isn't: nextlevelbuilder/ui-ux-pro-max-skill (120k stars, created 2025-11-30) leads taste-skill (79k stars, created 2026-02-19) by more than 50%. More importantly, they are fundamentally different kinds of things — taste-skill is an 87KB aesthetic-discipline prompt loaded fully into context; ui-ux-pro-max is a running product: a Python retrieval engine, 37 CSV datasets, 54 real font binaries, 11 unit-test files, and an npm CLI installer.

Key points

• Product, not prompt file: 662 files / 20.5MB across 7 skill directories. The flagship ui-ux-pro-max (3.4MB) holds 37 CSVs and 15 Python files, of which 11 are tests — including data-contract and relevance-evaluator tests, rare in this space.
• Deterministic retrieval: core.py implements BM25 scoring plus regex matching over CSVs, returning sorted rows with explicit fallback on zero hits. Styling choices don't depend on the model's improvisation.
• Broad stack coverage: 22 framework stacks — Web (React, Next, Vue, Svelte, Astro, Angular, shadcn, etc.), desktop (JavaFX, Avalonia, UWP, WinUI, WPF, Uno), and native mobile (SwiftUI, Jetpack Compose, Flutter, React Native). taste-skill hard-binds to React/Next (criticized in its issue #86).
• Claims verified, 10/10 accurate: styles.csv (79 searchable styles), 192 product palettes, 192 products, 192 reasoning profiles, 74 font pairings, 119 UX guidelines, 105 icons, 17 GSAP presets, 25 chart types, 22 stack CSVs. By contrast, taste-skill's five academic citations in research/laziness fall apart under tracing (misattributed EmotionPrompt data, over-anthropomorphized LazyBench, one unverifiable).
• Data freshness as a failing test: test_web_stack_freshness.py hard-codes current versions (react 19.2.x, nextjs 16.2, svelte 5, astro 7.1.6, angular 22.x, …), and validate_data.py's STACK_OFFICIAL_HOSTS enforces official-domain citations. Knowledge-base rot becomes a CI failure. A sibling test covers native/desktop stacks.
• Context strategy is the real watershed: the flagship SKILL.md is only 15,969 bytes (~16KB), acting as a dispatcher — full rules live in quick-reference.md (24.5KB) and pro-rules.md (10.9KB), loaded on demand. taste-skill keeps 87KB (~21.8k tokens) permanently in context, where compression can silently drop rules. taste-skill's issue #67 proposes progressive disclosure (~12KB core + on-demand references); ui-ux-pro-max was architected that way from the start.
• Real knobs: --variance, --motion, --density (1–10) — concept borrowed from taste-skill, but wired into Python retrieval, so changing values actually changes (reproducibly) which styles are recommended.
• Design-system generation: --design-system aggregates product type, style, palette, typography, and motion; --persist writes a MASTER.md plus per-page overrides for later reuse. taste-skill has no equivalent.

The commercial structure

The README openly documents Basic vs Premium tiers, with pricing at uupm.cc and PayPal donations. Basic (this repo) includes the retrieval engine, palettes, font pairings, 22 stacks, and the CLI design-system generator. Premium covers branding/logo/CIP, banners, slides, custom icons, AI-image integration, and enterprise design tokens. The brand (87KB), slides (19KB), and banner-design (13KB) directories are thin shells with no CSV data — storefronts for the paid tier. Legitimate open-core, but evaluators should know they hold a deliberately segmented free layer.

Head-to-head

| Dimension | ui-ux-pro-max (120k stars) | taste-skill (79k stars) | |---|---|---| | Form | Python engine + 37 CSVs + 54 fonts + CLI, 20.5MB | 13 markdown files, 87KB flagship | | Context cost | 16KB SKILL.md, knowledge on demand | 87KB (~21.8k tokens) always resident | | Mechanism | BM25 + regex over datasets | Pure prompt, rules in model judgment | | Determinism | High, sorted rows with fallback | None | | Claim accuracy | 10/10 verified | 4.5/5 citations check out | | Data freshness | Freshness tests fail CI on staleness | None | | Tests | 11 files incl. data contracts | None | | Design system | Generates and persists MASTER.md | Single-page only | | Frameworks | 22 stacks incl. desktop/mobile | React/Next bound | | Anti-homogenization | None | Core capability | | Business model | Basic open + Premium paid | Free, sponsor wall | | Dependencies | Python runtime, 5.2MB fonts | Zero, pure text |

Which one should you use?

• Single-page marketing sites, landing pages, portfolios: taste-skill — the negative-ban approach directly targets "looks AI-made," with zero dependencies.
• Multi-page products, dashboards, data tables: ui-ux-pro-max — taste-skill's §13 explicitly excludes dashboards/data tables; this repo has a density knob and chart/app-interface datasets.
• Cross-platform projects: only ui-ux-pro-max covers desktop and native mobile.
• A project-wide design system: ui-ux-pro-max's --design-system --persist is unique here.
• CI quality gate: neither works. ui-ux-pro-max's determinism lives at retrieval, not output validation. A regex linter in PR checks is the missing piece — taste-skill's §14 62-item checklist is effectively a ready-made linter spec.
• Ideal combo: distill taste-skill's §9 negative list into ~30 lines of project rules, and let ui-ux-pro-max handle retrieval. Running both heavy skills resident in context is expensive.

Chinese-language gap

Both need patching for Chinese projects: google-fonts.csv's 1,934 entries contain almost no Chinese fonts; recommend explicit stacks (Source Han Serif SC / Songti SC for display contrast, PingFang SC for body). CJK typography rules (line-start punctuation bans, 0.15em CJK-Latin spacing, text-spacing-trim), adjusted metrics (body line-height ≥1.8, display sizes ~10% smaller than Latin), and translated bans (full-width quotes, no ellipsis/tilde-as-range) must be supplied yourself.

Conclusion

ui-ux-pro-max-skill is the most engineering-mature project in this niche: a genuinely running retrieval engine, fully verified claims, freshness enforced by failing tests, a 16KB context-footprint SKILL.md with on-demand loading, 22 stacks, and persistent design systems. But it does not cure homogenization — it answers "what style and colors should I use," not "why does my page look AI-made." taste-skill remains the tool for the latter. One is a design-intelligence engine; the other is an aesthetic-discipline code. They complement rather than replace each other. The true gap is a third position neither occupies: a deterministic validator that can exit 1 in CI — between two projects totaling 200k stars, none can block a commit.

--- *References: github.com/nextlevelbuilder/ui-ux-pro-max-skill | uupm.cc | comparison repo: github.com/Leonxlnx/taste-skill | data snapshot 2026-08-24*