AIToday
Zenn AI/MLPublished: Nov 20, 2026, 04:00 JST

web-ui-quality Skill lifts AI screen quality to 79 of 81

Send on LINE
web-ui-quality Skill lifts AI screen quality to 79 of 81

3 Key Points

  1. What happened

    In four fictional scenarios tested in November 2026, AI-built screens met 52 of 81 quality criteria without the web-ui-quality Skill and 79 of 81 with it; in the form scenario, cost per run rose from $0.04 and 28–29 seconds to $0.10–0.12 and 53–66 seconds.

  2. Why it matters

    The Skill forces agents to open pages in a browser at two widths and measure before reporting, so teams get screens checked against fixed baselines instead of handed over unseen — though the extra checking raises time and cost per request.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The Skill grew out of a specific failure pattern: AI agents produce code that looks tidy but leaves small text, low-contrast buttons and unlabeled errors in place, then ask a human to check it in a browser. The author tested this by giving the same requests to an AI without the Skill three times each across four fictional scenarios, and had a separate measurement script open the results in Chromium at 390px and 1440px.

The first version of the Skill overcorrected. It changed button wording the user had not asked to change, following its own rule that buttons should use verbs, and it shrank table text to 14px on phones. The second version, whose numbers appear in the results, keeps existing wording and only proposes changes, and lets tables scroll horizontally instead of shrinking text.

The author notes several limits on the evidence: the scenarios and criteria were written by the same person who wrote the Skill, the numbers were measured on the scenarios used to fix it, and only Chromium was tested — not Safari on iPhone or Edge on Windows. Visual taste was not measured at all, and some unskilled screens looked better organized. The Skill sets a floor, not a style.

FAQ
How do I install web-ui-quality?
Run npx skills add itoken-lab-jp/skills -g --skill web-ui-quality. Node.js is required, and the -g flag installs it for the whole user rather than one project.
What tools can use this Skill?
It targets agents that read the common Agent Skills format, such as Claude Code or Codex. The tests, however, were run only with Claude Code and only with Claude Sonnet as the working model.
Does the Skill measure accessibility?
Yes. It requires a text-to-background contrast ratio of 4.5 or higher, visible keyboard focus, and touch targets around 44px, drawing on WCAG 2.2 and the Digital Agency Design System.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Next articleAI人事: AI for hiring, evaluation and labor tasks