Records a browsing task as a committed Python job on a stealth identity, then replays it with no model in the loop — lint-gated, seed-jittered, with a run manifest. Use to make a browser flow repeatable or scheduled, replay or list the saved jobs, cut per-run LLM cost, or diagnose a failed run and repair the job. Not for one-off browsing, creating an identity, or installing the toolchain.
