Replays a recorded browser session trajectory against the original URL or a modified variant, using embedding similarity on selectors to recover when the DOM has drifted. You'd reach for this when regression testing a UI flow after a deploy, reproducing a captured bug, or preparing a visual diff between two runs. The architecture bets heavily on selector recovery since Browserbase doesn't offer native replay. It defaults to an 85% step match threshold before declaring failure, which is aggressive but honest about the nondeterminism you'll hit with timing and content variation. If you need more than one retry per step, the site probably needs a fresh recording rather than another replay attempt.
npx skills add https://github.com/ruvnet/ruflo --skill browser-replay