This matters to me a lot because accessibility tooling gets dishonest here very quickly. A product runs a scan, says something grand about screen readers, and suddenly teams act as if assistive technology has been tested. It has not.
A11Y Cat keeps that boundary visible. The repo includes a manual accessibility checklist, explicit AT sign-off and validation material, and product language that keeps manual review separate from automated checking. The tool can prepare, support, and organise that work, but it does not get to impersonate it.
That choice probably makes the product sound less magical, and good. I would rather sound less magical than mislead people into releasing something that has not actually been checked with NVDA, JAWS, VoiceOver, keyboard flows, focus changes, live regions, and all the other behaviour that matters.
This is one of the areas where I think the project is strongest. Not because it automates everything, but because it refuses to lie about what automation can and cannot do.
Documentation trail
MANUAL_ACCESSIBILITY_CHECKLIST.mdEXTENSION_RELEASE_CHECKLIST.mdREADME.md
