ui-ux-pro-max (nextlevelbuilder/ui-ux-pro-max-skill)

评测

在 Codex CLI（SOTA 档）上跑 3 个任务、对照、一句话提示、装 skill 三组，共 18 次运行，由规则脚本打分（没有 LLM 评委），断网沙箱、每次运行独立临时目录。具体做法见下方英文评测原文。

三组对比

Codex CLI 实测，18 次运行，规则脚本打分，没有 LLM 评委。总分 20.9 / 30，结论 视情况而定。

下面是评测原文，英文。

Verdict: Situational (20.9 / 30). Source: nextlevelbuilder/ui-ux-pro-max-skill, .claude/skills/ui-ux-pro-max, pinned at commit 3a13158. Install: npx skills@latest add nextlevelbuilder/ui-ux-pro-max-skill --skill ui-ux-pro-max. Licence: MIT (repo LICENSE).

What it is

The ui-ux-pro-max skill (nextlevelbuilder/ui-ux-pro-max-skill): a UI/UX design knowledge base with a local Python search tool (scripts/search.py) over bundled CSV data (styles, palettes, type pairings, product-type rules, UX guidelines). Its main flow is to generate a 'design system' recommendation for the product (--design-system -p <name>), then build the UI against it, with a priority-ordered rule set (accessibility, touch targets, responsiveness, motion).

How we tested

Codex CLI 0.166 (gpt-6-astra, low reasoning) on a dedicated Windows test PC, a fresh temporary project and temporary HOME/TMPDIR/caches per run, skill installed (treatment) or absent (control) or absent with the skill's core requirement squeezed into one sentence appended to the prompt (one-liner group), 2 runs per group per task, network off, /tmp writes blocked. The prompts never mention the skill (except the wording noted below).

Three tasks, same prompt for all groups: a landing page for a fictional snack-subscription brand (Tidepool), an admin dashboard for a municipal water-quality network, and a redesign of a plain climbing-gym homepage (old.html) keeping all content. The one-liner group adds one sentence listing the skill's priorities (contrast 4.5:1, 44 px targets, mobile-first, SVG icons, a type scale, reduced motion). Pages were screenshot offline (desktop and mobile) and scored blind from shuffled contact sheets, without condition labels.

1. New landing page from a brief — Create a single-page landing site for Tidepool, a small-batch seaweed snack brand that sells by subscription (three flavours, $14/box, ships monthly). Write one self-contained index.html with inline CSS and minimal JS only if needed. No frameworks, no build st…
2. Data-dense dashboard UI — Build a dashboard page for a municipal water-quality monitoring team: current readings for 6 stations (pH, turbidity, chlorine, temperature), a 7-day trend for each, and an alerts panel. Single self-contained dashboard.html with inline CSS/JS and hard-coded re…
3. Reshape an existing bland page — old.html is the current homepage of Granite Peak Climbing Gym (Boulder, CO). Redesign it into new.html: keep all of its real content (pricing, classes, hours, address) but make the page look like it belongs to this specific gym. Self-contained HTML/CSS, no fra…

Scores come from a blind read by Grok Bot (shuffled ids, no condition shown) against a fixed rubric, 0–5 per run; an AI judge, so treat small gaps as noise.

Results

The one-liner given to the third group: "Follow UI/UX best practice in priority order: accessibility first (4.5:1 contrast, alt text, visible focus rings, labels), touch targets of at least 44 px with 8 px spacing, mobile-first responsive layout with viewport meta and no horizontal scroll, a style matched to the product with SVG icons (no emoji), a consistent type scale (16 px base, 1.5 line height) and semantic colour tokens, and motion that respects prefers-reduced-motion."

Mean score: 4.0 without vs 3.96 with the skill (delta -0.04). Tokens per run: 116k vs 160k (1.37x); time 186s vs 184s. One-liner group: 3.92 (the skill's gain over the one-liner: +0.04; one-liner's own gain over no skill: -0.08); one-liner tokens 123k per run, so the skill costs 1.30x the one-liner.

What happened

• No visible quality difference: blind mean 4.00 control, 3.92 one-liner, 3.96 skill (n=6 each). Per task the groups differ by at most 0.25 points, which is below what two runs per cell can resolve. All 18 pages are polished, responsive and share a similar look (cream or pale-green ground, dark green accent, serif display type).
• The skill was actually used: all 6 treatment runs read SKILL.md and ran search.py --design-system (e.g. for 'snack subscription playful coastal'). The script works on Windows with the system Python.
• Accessibility basics are already standard in plain Codex output: viewport meta, @media queries and focus styles in 18/18 pages and prefers-reduced-motion in 17/18 (the one miss was a control dashboard). So the skill's accessibility-first rules had little left to add.
• Cost: 160k tokens per run for the skill vs 116k control and 123k one-liner (1.4x and 1.3x); wall time was the same (~185 s). Treatment runs also copied the skill folder into the project (.agents), which is where the extra tokens go.
• One check misfired: 'keeps pricing ($22/$79/$720)' failed for 4 of 6 redesign runs in control and skill groups because those pages wrote the dollar sign in a separate <sup> element (<sup>$</sup>79); the screenshots show the prices intact. It does not feed the score.

Dimension scores

Limits of this test

• Blind visual judging by the reviewer from full-page contact sheets (desktop 600 px wide per page, mobile 300 px); differences of 0.25 are within judge noise. Pages are not compared at full resolution or interactively.
• Only n=2 per cell and three tasks; the skill's value on a long, multi-page product is not covered.
• The skill's data-driven recommendations (palette and font pairings) could not load web fonts offline; pages fall back to system fonts, which hides typography choices.
• n = 2 per group per task: differences under about 1 point are noise.
• Blind AI judging: single judge, no inter-rater check; the original control/treatment runs and the later one-liner runs were judged in separate sessions (one-liner on 2026-11-02), so cross-session drift is possible.

Security

See security.md. Scan: 0 CRITICAL / 3 HIGH / 3 MEDIUM. Verdict: safe to run (no-network sandbox). A clean scan does not prove a skill safe.

Bottom line

With a capable base model the pages are already good; the skill adds a design-system recommendation step but no measurable quality gain here, for 1.4x tokens. Use it as an idea generator or checklist, not as an expected quality upgrade.