Accessibility audit agent
Design an agent that audits accessibility on every UI change before merge. Be honest about the technology: which classes of accessibility problems can it genuinely catch, which require a human or a real assistive-technology user, and how should the agent communicate that boundary so a clean report isn't mistaken for 'fully accessible'? How would you measure whether it's actually improving outcomes, and how do you pitch it without overpromising?
route_accessibility_findings(findings: list[str], verified_fixed_rules: list[str]) → list[str][["contrast=btn-primary","focus-order=modal","alt-text=hero-img"],["hero-img"]]out["advise","human","block","detected issues cover only mechanically checkable categories"][[],[]]out["clean scan is not a claim of accessibility"][["cognitive-load=checkout"],["checkout"]]out["human","clean scan is not a claim of accessibility"]Turn the fuzzy goal into a bounded agent task. Say what the agent reads, what it may do on its own versus draft for a human, and what it must never touch. Then earn the trust: how you’d verify it before rollout, and the pitch (the problem, what you built, why it matters).
Agent build: design the agent in plain language (or narrate it). The coach grades the decomposition, the guardrails, the verification plan, and the pitch.