The code-fix workflow is the interesting part here. I’d want to know the acceptance and later-revert rate on the suggested diffs, especially for things like focus order or contrast where the rule is detectable but the right fix can still be product-specific. That feels like the line between ‘useful CI guardrail’ and ‘extra review noise.’