Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Audits web components (custom elements, Shadow DOM) for accessibility. Covers ElementInternals, cross-shadow ARIA, form-associated custom elements, and shadow DOM focus management.
.claude/skills/community-access-web-component-specialist/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 12% | 0% |
| case-03 | ✓→✓ | = Same ✓ | 28% | 0% |
| case-04 | ✓→✓ | = Same ✓ | 21% | 0% |
| case-05 | ✓→✓ | = Same ✓ | 8% | 0% |
Derived from .claude/agents/web-component-specialist.md. Treat platform-specific tool names or delegation instructions as Codex equivalents.
You audit custom elements and Shadow DOM for accessibility. Shadow DOM breaks traditional ARIA references, label associations, and focus management.
attachInternals() for role, ariaLabel, form association instead of attributesaria-labelledby can't cross shadow boundaries; use ElementInternals.ariaLabel or host attributesstatic formAssociated = true, setFormValue(), setValidity(), label associationdelegatesFocus: true, tab order, programmatic focus into shadow DOMcomposed: true, bubbles: true for custom events crossing shadow boundary| Issue | Fix | |-------|-----| | Cross-shadow aria-labelledby | ElementInternals.ariaLabel or host attribute | | Missing delegatesFocus | Add to attachShadow() options | | No role on host | ElementInternals.role | | Not form-associated | static formAssociated = true + ElementInternals |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 23,321 | 15,053 | -35% | 1 | 1 | 0% | 4,455 | 3,230 | -27% | 0 | 0 | — |
case-02 | fail→pass | 26,145 | 19,942 | -24% | 1 | 1 | 0% | 4,693 | 3,833 | -18% | 0 | 0 | — |
case-03 | pass→pass | 5,657 | 5,556 | -2% | 1 | 1 | 0% | 985 | 1,264 | +28% | 0 | 0 | — |
case-04 | pass→pass | 12,778 | 13,287 | +4% | 1 | 1 | 0% | 2,092 | 2,522 | +21% | 0 | 0 | — |
case-05 | pass→pass | 13,369 | 12,341 | -8% | 1 | 1 | 0% | 2,410 | 2,605 | +8% | 0 | 0 | — |
case-06 | pass→pass | 7,853 | 8,928 | +14% | 1 | 1 | 0% | 1,451 | 1,991 | +37% | 0 | 0 | — |
case-07 | fail→fail | 15,861 | 11,504 | -27% | 1 | 1 | 0% | 2,570 | 2,387 | -7% | 0 | 0 | — |
case-08 | pass→pass | 11,392 | 11,042 | -3% | 1 | 1 | 0% | 1,950 | 2,156 | +11% | 0 | 0 | — |
case-09 | pass→pass | 6,898 | 6,399 | -7% | 1 | 1 | 0% | 1,239 | 1,371 | +11% | 0 | 0 | — |
case-10 | fail→pass | 12,619 | 13,549 | +7% | 1 | 1 | 0% | 2,431 | 2,713 | +12% | 0 | 0 | — |
case-11 | pass→pass | 8,843 | 6,577 | -26% | 1 | 1 | 0% | 1,538 | 1,441 | -6% | 0 | 0 | — |
case-12 | pass→pass | 11,623 | 11,193 | -4% | 1 | 1 | 0% | 2,171 | 2,363 | +9% | 0 | 0 | — |
case-13 | pass→pass | 10,394 | 6,560 | -37% | 1 | 1 | 0% | 1,938 | 1,540 | -21% | 0 | 0 | — |
case-14 | pass→pass | 3,698 | 8,019 | +117% | 1 | 1 | 0% | 664 | 1,096 | +65% | 0 | 0 | — |
case-15 | pass→pass | 4,086 | 4,689 | +15% | 1 | 1 | 0% | 673 | 1,099 | +63% | 0 | 0 | — |
case-16 | pass→pass | 10,018 | 11,247 | +12% | 1 | 1 | 0% | 1,720 | 2,181 | +27% | 0 | 0 | — |
case-17 | fail→fail | 15,005 | 14,371 | -4% | 1 | 1 | 0% | 2,428 | 2,669 | +10% | 0 | 0 | — |
case-18 | pass→pass | 13,280 | 11,987 | -10% | 1 | 1 | 0% | 2,409 | 2,616 | +9% | 0 | 0 | — |
case-19 | pass→pass | 8,451 | 10,996 | +30% | 1 | 1 | 0% | 1,425 | 2,410 | +69% | 0 | 0 | — |
case-20 | pass→pass | 5,501 | 4,233 | -23% | 1 | 1 | 0% | 949 | 1,132 | +19% | 0 | 0 | — |
case-21 | pass→pass | 4,882 | 4,859 | -0% | 1 | 1 | 0% | 755 | 1,104 | +46% | 0 | 0 | — |
case-22 | pass→pass | 8,784 | 6,110 | -30% | 1 | 1 | 0% | 1,547 | 1,361 | -12% | 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 +9 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.