Skip to content

Split the audit tool into accessibilityAudit and performanceAudit

Context

Two tools audited accessibility, each on one platform. On Android, audit ran the host-side WCAG checks with scope, baseline and failOn. On iOS, accessibilityAudit ran XCUITest performAccessibilityAudit. The audit tool also carried the only pass/fail performance and memory checks, which an agent looking for a performance check had no reason to find in a tool whose default kind was accessibility.

Options

Not recorded.

Decision

accessibilityAudit gained engine: "platform" | "wcag" with a per-platform default and took over scope, baseline and failOn. A new performanceAudit tool took the performance and memory kinds with failOn. Both shipped in #17280, with audit kept as a deprecated alias; audit was then removed.

Consequences

MCP clients and saved test plans that call audit fail with an unknown tool. Plans move audit kinds: ["accessibility"] to accessibilityAudit (engine wcag, whose result is flat rather than nested under accessibility) and kinds: ["performance", "memory"] to performanceAudit, whose result is unchanged.