Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →CI/CD accessibility agent. Sets up, manages, and troubleshoots accessibility CI pipelines. Supports baseline management, SARIF output, PR annotations, and threshold configuration. Works with GitHub Actions, Azure DevOps, GitLab CI, CircleCI, and Jenkins.
.claude/skills/community-access-ci-accessibility-a5dec5/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-21 | ✓→✗ | ▼ Worse | 6% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | -28% | 0% |
You are a CI/CD accessibility specialist. You help teams set up, maintain, and troubleshoot automated accessibility scanning in their continuous integration pipelines.
axe-baseline.json) so CI only fails on regressionsCheck for existing CI config, accessibility tooling, baseline files, and scan configuration.
Determine CI platform, scanning tool (axe-core CLI, Playwright+axe, Lighthouse CI), gating strategy (strict/standard/baseline), and output format (SARIF, PR comment, artifact).
Create CI config with WCAG 2.2 AA tags, baseline comparison, SARIF output, and pass/fail summary.
Run pipeline in a test PR, document for the team, offer scheduled scans.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→fail | 7,808 | 6,683 | -14% | 1 | 1 | 0% | 1,188 | 1,218 | +3% | 0 | 0 | — |
case-02 | fail→pass | 25,088 | 20,879 | -17% | 1 | 1 | 0% | 4,556 | 4,077 | -11% | 0 | 0 | — |
case-01 | fail→pass | 20,857 | 17,293 | -17% | 1 | 1 | 0% | 4,168 | 3,709 | -11% | 0 | 0 | — |
case-03 | fail→fail | 20,618 | 16,127 | -22% | 1 | 1 | 0% | 3,697 | 3,280 | -11% | 0 | 0 | — |
case-04 | pass→pass | 26,410 | 7,427 | -72% | 1 | 1 | 0% | 2,139 | 1,545 | -28% | 0 | 0 | — |
case-05 | pass→pass | 10,421 | 7,330 | -30% | 1 | 1 | 0% | 1,884 | 1,591 | -16% | 0 | 0 | — |
case-06 | fail→pass | 9,930 | 10,711 | +8% | 1 | 1 | 0% | 1,549 | 1,985 | +28% | 0 | 0 | — |
case-08 | pass→pass | 17,468 | 15,456 | -12% | 1 | 1 | 0% | 2,852 | 2,899 | +2% | 0 | 0 | — |
case-09 | pass→pass | 5,394 | 5,112 | -5% | 1 | 1 | 0% | 811 | 1,141 | +41% | 0 | 0 | — |
case-10 | pass→pass | 14,879 | 9,295 | -38% | 1 | 1 | 0% | 2,243 | 1,679 | -25% | 0 | 0 | — |
case-11 | pass→pass | 16,499 | 9,033 | -45% | 1 | 1 | 0% | 2,860 | 1,800 | -37% | 0 | 0 | — |
case-12 | pass→pass | 10,659 | 5,058 | -53% | 1 | 1 | 0% | 1,588 | 1,006 | -37% | 0 | 0 | — |
case-13 | pass→pass | 9,049 | 7,702 | -15% | 1 | 1 | 0% | 1,360 | 1,445 | +6% | 0 | 0 | — |
case-14 | pass→pass | 17,737 | 14,695 | -17% | 1 | 1 | 0% | 2,630 | 2,664 | +1% | 0 | 0 | — |
case-15 | pass→pass | 10,204 | 9,257 | -9% | 1 | 1 | 0% | 1,599 | 1,734 | +8% | 0 | 0 | — |
case-16 | pass→pass | 15,038 | 13,918 | -7% | 1 | 1 | 0% | 2,485 | 2,498 | +1% | 0 | 0 | — |
case-17 | pass→pass | 11,132 | 14,922 | +34% | 1 | 1 | 0% | 1,699 | 2,629 | +55% | 0 | 0 | — |
case-18 | pass→pass | 15,898 | 15,355 | -3% | 1 | 1 | 0% | 2,324 | 2,791 | +20% | 0 | 0 | — |
case-19 | pass→pass | 16,052 | 13,871 | -14% | 1 | 1 | 0% | 2,483 | 2,593 | +4% | 0 | 0 | — |
case-20 | pass→pass | 9,187 | 8,679 | -6% | 1 | 1 | 0% | 1,714 | 1,856 | +8% | 0 | 0 | — |
case-21 | pass→fail | 16,909 | 16,110 | -5% | 1 | 1 | 0% | 2,832 | 2,989 | +6% | 0 | 0 | — |
case-22 | pass→pass | 19,163 | 14,651 | -24% | 1 | 1 | 0% | 3,170 | 2,541 | -20% | 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. 1 case got worse with the skill loaded, and it is 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.