Audit page accessibility with AI
Audit a particular screen state and leave a record another tester can reproduce, including the checks that need human judgement.
Audit page accessibility with AI by identifying the state that matters first. An account form before submission and the same form after a failed submission present different interfaces. The accessibility-audit skill records the actions, viewport and sign-in mode needed to reach the chosen state.
The resulting report separates detected rule failures from observations that still need someone to judge them. A tool might find a missing name or a contrast issue. Whether alternative text describes the image meaningfully depends on its purpose. That distinction stays visible when the report reaches the team that owns the page.
What it does
- Capture the route and steps that expose a hidden error panel.
- Record the viewport so a mobile-only control can be tested again.
- Keep a selector and a small piece of evidence beside each detected issue.
- Identify questions about labels and alternative text that need the content owner.
- Report a blocked screen state explicitly when the audit could not reach it.
Install and use
- Install the published a11y-agent CLI and Playwright Chromium with a supported Node runtime. Use an authorised test page and representative sample data.
- Reproduce the chosen state and collect the rule findings without recording tokens or browser session secrets.
- Review the evidence and assign specific manual checks before treating the audit as complete.
FAQ
Source
Made by a11y-agent-skills contributors (smukh). Original skill: accessibility-audit, published on GitHub under the MIT licence.
View on GitHub