Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Set up or run design QA gates — token + hardcoded-value lint, automated a11y (axe), contrast, visual regression across variants/states/themes/RTL, and the manual a11y checklist. Use when the user wants CI quality gates, to prevent design regressions, or to QA a component/screen before shipping.
.claude/skills/plugin87-design-qa/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 29% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -63% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -28% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -74% | 0% |
Stand up the automated + manual gates that stop quality from regressing.
workflows/design-qa.md (the QA pyramid: token/lint gates → automated a11y → visual regression → manual a11y).python3 scripts/validate_tokens.py, python3 scripts/validate_contrast.py (batch WCAG over the token pairs), and python3 scripts/lint_hardcodes.py <src> (no raw hex/px/timing in component code). The repo's .github/workflows/ci.yml runs these.node scripts/measure_render.mjs <file.html> (and --dark) — it opens the page in headless Chromium, disables transitions, and measures the true computed-style + alpha-composited contrast of every text element (catches what static token checks miss).accessibility/*).| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 26,738 | 28,884 | +8% | 1 | 1 | 0% | 4,753 | 6,126 | +29% | 0 | 0 | — |
case-02 | fail→fail | 27,168 | 29,958 | +10% | 1 | 1 | 0% | 5,080 | 6,622 | +30% | 0 | 0 | — |
case-03 | fail→fail | 31,697 | 26,734 | -16% | 1 | 1 | 0% | 5,567 | 5,731 | +3% | 0 | 0 | — |
case-04 | fail→pass | 16,198 | 3,613 | -78% | 1 | 1 | 0% | 2,944 | 1,078 | -63% | 0 | 0 | — |
case-05 | fail→pass | 11,637 | 5,793 | -50% | 1 | 1 | 0% | 1,945 | 1,407 | -28% | 0 | 0 | — |
case-06 | fail→pass | 10,810 | 2,261 | -79% | 1 | 1 | 0% | 1,831 | 792 | -57% | 0 | 0 | — |
case-07 | fail→pass | 20,423 | 3,265 | -84% | 1 | 1 | 0% | 3,480 | 913 | -74% | 0 | 0 | — |
case-08 | fail→pass | 9,915 | 1,976 | -80% | 1 | 1 | 0% | 1,619 | 693 | -57% | 0 | 0 | — |
case-09 | pass→pass | 14,341 | 10,396 | -28% | 1 | 1 | 0% | 2,292 | 2,002 | -13% | 0 | 0 | — |
case-10 | pass→pass | 12,989 | 2,260 | -83% | 1 | 1 | 0% | 1,919 | 761 | -60% | 0 | 0 | — |
case-11 | pass→pass | 17,541 | 12,602 | -28% | 1 | 1 | 0% | 2,692 | 2,307 | -14% | 0 | 0 | — |
case-12 | pass→pass | 5,211 | 5,735 | +10% | 1 | 1 | 0% | 723 | 1,236 | +71% | 0 | 0 | — |
case-13 | pass→pass | 16,970 | 14,640 | -14% | 1 | 1 | 0% | 2,896 | 2,733 | -6% | 0 | 0 | — |
case-14 | pass→pass | 5,864 | 4,354 | -26% | 1 | 1 | 0% | 897 | 1,009 | +12% | 0 | 0 | — |
case-15 | pass→pass | 4,700 | 3,826 | -19% | 1 | 1 | 0% | 732 | 1,079 | +47% | 0 | 0 | — |
case-16 | pass→pass | 3,867 | 1,986 | -49% | 1 | 1 | 0% | 538 | 729 | +36% | 0 | 0 | — |
case-17 | fail→pass | 10,450 | 8,098 | -23% | 1 | 1 | 0% | 1,520 | 1,755 | +15% | 0 | 0 | — |
case-18 | pass→pass | 15,294 | 14,224 | -7% | 1 | 1 | 0% | 2,470 | 2,777 | +12% | 0 | 0 | — |
case-19 | pass→pass | 7,090 | 3,225 | -55% | 1 | 1 | 0% | 1,118 | 971 | -13% | 0 | 0 | — |
case-20 | pass→pass | 15,877 | 18,050 | +14% | 1 | 1 | 0% | 2,500 | 3,307 | +32% | 0 | 0 | — |
case-21 | pass→pass | 11,006 | 10,935 | -1% | 1 | 1 | 0% | 2,109 | 2,399 | +14% | 0 | 0 | — |
case-22 | pass→pass | 10,989 | 8,129 | -26% | 1 | 1 | 0% | 2,208 | 1,835 | -17% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +32 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.