UX Audit (v2)

Walk through a live web app AS a real user. The audit is interaction-first — typing, clicking, sending, watching, screenshotting. A static DOM sweep cannot produce a verdict.

Why this skill exists in v2

Most catastrophic UX bugs only surface under interaction:

• The Send button that doesn't clear the input
• The @-mention picker that double-inserts the handle
• The thread pane that breaks layout when the sidebar is also open at 1280px
• The form submit that disables the button but silently 5xx's
• The console warning that's been logging for weeks but no audit ever flagged

A DOM probe sees none of these. They require: type → wait → screenshot → diff → console-read → repeat. The previous audit method drifted toward sweep mode (faster, feels productive, blind to behaviour). v2 makes that drift impossible — the agent cannot output "Pass" without an Interaction Manifest that proves real clicks happened.

Verdict states

The audit ends in exactly one of: