All products

Write accessibility tests with AI

Protect a verified repair with a focused test that fails for the original barrier and passes when the interaction works.

Write accessibility tests with AI after a repair has been verified in the original state. A form might now move focus to its error summary, yet an assertion that only checks for a new HTML attribute could pass even when focus stays on the submit button. The accessibility-regression-test skill turns the reported barrier into a user-visible outcome.

The method selects a test layer that can reproduce the behaviour reliably. That might be a lint rule for static misuse, a component test, a browser assertion or a keyboard journey. Judgements that require a person stay in a review record rather than being hidden behind a passing test.

What it does

  • Assert that focus returns to the control that opened a dialog.
  • Test a rendered component when the isolated example reproduces the actual defect.
  • Choose a browser test for a failure that depends on routing or computed styles.
  • Keep test data, viewport and motion preferences stable between runs.
  • Record a human-only judgement without turning it into an arbitrary code assertion.

Install and use

  1. Provide the before and after evidence and your project's test environment. Use a component runner or browser tooling according to the behaviour involved.
  2. State what a user could not do, then choose one observable outcome to protect.
  3. Where practical, run the check against the broken behaviour and the repaired state under the same conditions.

FAQ

Source

Made by a11y-agent-skills contributors (smukh). Original skill: accessibility-regression-test, published on GitHub under the MIT licence.

View on GitHub