Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use for testing the accessibility of a web project
.claude/skills/kreerc-accessibility-testing/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-18 | ✗→✓ | ▲ Improved | -43% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 59% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 104% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 66% | 0% |
Implementing accessibility correctly is crucial to deliver quality web experiences for everyone, including developers.
To catch accessibility defects quickly and reliably, employ automated and rule-based accessibility checks. For this, set up a development requirement for the pa11y/pa11y CLI tool and the axe-core runner, and run it against fresh development builds. Also cover individual features and components, isolated from the rest of the page.
Include these checker tools into an existing CI pipeline.
For each component, check each component for its full WCAG 2.2 level AAA conformance, and create a table that contains the number of the success criterion, its name, a short description, and the conformance status (pass, fail, not applicable, untested)
If it helps, you may also make this as a template and re-use it accordingly.
You may use browser based tools (like Playwright) to run checks yourself that traditionally required real users for testing. If you can't make a judgement or it might be un-reliable, note a todo with a clear and concise assessment instruction for a developer.
Whenever you have made an accessibility assessment or judgement, produce a short writeup of all documented issues and suggested fixes. Never suggest quick-wins or shorthand fixes if a complete architectural overhaul would be the better and correct choice. This is especially true if the quick fix would involve ARIA patterns.
As required per EU law and local equivalents (e.g. Barrierefreiheitsstärkungsgesetz/BFSG in Germany), projects are required to have a public-facing information page about the accessibility of the page, with the following contents in natural language:
The law also mandates that developers inform the governing bodies of their offerings pro-actively. Draft an email for that and research the relevant email address.
This page must be available on every page, near other regulatory links such as a privacy policy.
Implement this page in any case, even if the service may not be directly covered by the law.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-18 | fail→pass | 14,508 | 4,945 | -66% | 1 | 1 | 0% | 2,749 | 1,574 | -43% | 0 | 0 | — |
case-04 | fail→pass | 15,611 | 20,963 | +34% | 1 | 1 | 0% | 2,889 | 4,591 | +59% | 0 | 0 | — |
case-01 | fail→fail | 30,899 | 32,094 | +4% | 1 | 1 | 0% | 4,488 | 6,661 | +48% | 0 | 0 | — |
case-02 | fail→pass | 15,524 | 25,344 | +63% | 1 | 1 | 0% | 2,684 | 5,474 | +104% | 0 | 0 | — |
case-03 | fail→pass | 24,996 | 30,936 | +24% | 1 | 1 | 0% | 5,292 | 6,777 | +28% | 0 | 0 | — |
case-05 | fail→pass | 12,191 | 15,980 | +31% | 1 | 1 | 0% | 2,118 | 3,524 | +66% | 0 | 0 | — |
case-06 | pass→pass | 8,929 | 8,975 | +1% | 1 | 1 | 0% | 1,916 | 2,081 | +9% | 0 | 0 | — |
case-07 | pass→pass | 13,343 | 22,911 | +72% | 1 | 1 | 0% | 2,266 | 4,930 | +118% | 0 | 0 | — |
case-08 | pass→pass | 15,889 | 28,666 | +80% | 1 | 1 | 0% | 2,613 | 5,881 | +125% | 0 | 0 | — |
case-09 | pass→pass | 11,314 | 9,794 | -13% | 1 | 1 | 0% | 1,805 | 2,158 | +20% | 0 | 0 | — |
case-10 | pass→pass | 16,358 | 22,986 | +41% | 1 | 1 | 0% | 2,797 | 4,234 | +51% | 0 | 0 | — |
case-11 | fail→pass | 8,410 | 7,956 | -5% | 1 | 1 | 0% | 1,273 | 2,043 | +60% | 0 | 0 | — |
case-12 | fail→pass | 10,990 | 8,558 | -22% | 1 | 1 | 0% | 1,608 | 1,833 | +14% | 0 | 0 | — |
case-13 | fail→pass | 15,268 | 21,264 | +39% | 1 | 1 | 0% | 1,689 | 4,227 | +150% | 0 | 0 | — |
case-14 | fail→pass | 19,940 | 24,879 | +25% | 1 | 1 | 0% | 2,780 | 5,256 | +89% | 0 | 0 | — |
case-15 | pass→pass | 15,342 | 10,797 | -30% | 1 | 1 | 0% | 2,731 | 2,368 | -13% | 0 | 0 | — |
case-16 | pass→pass | 15,697 | 10,103 | -36% | 1 | 1 | 0% | 2,212 | 1,994 | -10% | 0 | 0 | — |
case-17 | pass→pass | 16,129 | 20,509 | +27% | 1 | 1 | 0% | 2,220 | 4,204 | +89% | 0 | 0 | — |
case-19 | pass→pass | 14,896 | 12,040 | -19% | 1 | 1 | 0% | 2,416 | 2,532 | +5% | 0 | 0 | — |
case-20 | pass→fail | 19,613 | 28,715 | +46% | 1 | 1 | 0% | 4,316 | 6,228 | +44% | 0 | 0 | — |
case-21 | pass→fail | 20,685 | 31,478 | +52% | 1 | 1 | 0% | 3,582 | 6,605 | +84% | 0 | 0 | — |
case-22 | pass→fail | 28,840 | 39,392 | +37% | 1 | 1 | 0% | 4,663 | 6,778 | +45% | 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 +27 percentage points is the difference between those two pass rates over the 22 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
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.