Walking the built app page by page via the browser CLI — actually
navigating it, not reviewing code. Here's what a run produces.

WHAT IT DOES
Navigates the app like a user, page by page, capturing bugs +
screenshots as it goes, and records a video of the whole walkthrough for
adversarial review. You get evidence, not a checklist to run yourself.

OUTPUT — bugs bucketed by priority
  CRITICAL (blocks ship)
    ▸ /checkout — "Place order" button does nothing on mobile viewport.
      Screenshot attached. This is a revenue-blocker.
    ▸ /signup — form submits but shows a 500; account not created.
  MEDIUM
    ▸ /dashboard — chart overflows its container below 768px.
    ▸ /settings — save succeeds but no confirmation; users double-click.
  LOW
    ▸ /about — a stale year in the footer.
    ▸ /pricing — hover state missing on one CTA.

Each bug: the page, what broke, a screenshot at the moment it broke.

THE VIDEO IS THE POINT
The full walkthrough is recorded so you (or a reviewer) can watch the
app being used adversarially before it ships — you see the critical
checkout bug happen, not just read that it exists. That's what catches
the "works on my machine" class of issue a code review never would.

WHY DOGFOOD-BEFORE-SHIP BEATS "we'll catch it in QA"
The critical bugs above are the ones that only appear when something
actually clicks through the real built app on a real viewport. This runs
that pass automatically, every build, and hands you a ranked list plus
proof.

Point it at your deployed preview URL and it'll produce the ranked
report + video for your actual app.