Skip to content

Add measured-design-audit and design-regression-guard (Development & Code Tools) - #1556

Open
Yuxiang-Ma wants to merge 2 commits into
ComposioHQ:masterfrom
Yuxiang-Ma:add-measured-design-skills
Open

Add measured-design-audit and design-regression-guard (Development & Code Tools)#1556
Yuxiang-Ma wants to merge 2 commits into
ComposioHQ:masterfrom
Yuxiang-Ma:add-measured-design-skills

Conversation

@Yuxiang-Ma

Copy link
Copy Markdown

Adds two skills to Development & Code Tools, both distilled from real design-engineering sessions (not hypothetical):

  • measured-design-audit — design review by measurement instead of eyeballing: computed WCAG contrast against composited backgrounds, 44 px touch targets, font-size/tracking censuses, multi-viewport overflow. In the source sessions it found 133 contrast failures and seventeen distinct font sizes that three visual review rounds had missed; color fixes are solved numerically for the minimum passing shift rather than nudged.
  • design-regression-guard — the counterpart: every fixed defect class becomes a Playwright assertion in the same commit (including semantic contracts like "drawn ink equals the reported percentage"). Its credential: the guard caught its own author twice within hours of being written.

Checklist per CONTRIBUTING:

  • Real use case: yes — both extracted from multi-day sessions on a live research site.
  • No duplicates found: existing entries cover running Playwright (webapp-testing, Playwright Browser Automation) and screenshot capture; neither covers measurement-first auditing nor freezing fixes as regressions.
  • Documented with runnable starter blocks in each SKILL.md; tested with Claude Code.
  • Entries inserted alphabetically with attribution.

Two skills distilled from real design-engineering sessions on a static
research site: an audit that replaces visual review with measurement
(computed WCAG contrast on composited backgrounds, 44px touch targets,
type-scale census, multi-viewport overflow; found 133 contrast failures
and seventeen font sizes that three visual reviews missed), and its
counterpart that freezes each fixed defect class as a Playwright
assertion in the same commit -- in the source sessions the guard caught
its own author twice within hours.

Both live in Yuxiang-Ma/agent-skills (research-kit plugin), tested with
Claude Code, with runnable starter blocks in each SKILL.md.
The previous commit's tooling wrote the README back with CRLF endings,
turning a two-line addition into a 505-line rewrite and tripping the
validator's 'changes before the Skills section' rule. The file is
restored from upstream byte-for-byte with only the two entries added.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant