Browser Use 1.0.5 adds a built-in judge for agent runs

The update adds a judging step that uses the developer's own model to compare each run with the task and record a verdict in result.judgement, covering success, failure reason, description, prompt-injection detection, bot-block or CAPTCHA detection, and gotchas. Developers can pass judge_context to set criteria or judge=False to turn it off; the earlier use_judge and ground_truth options work again. The judge reviews the run trace, final screenshots and final page unless vision is disabled, and reports site pushback even when a run succeeds. The release is available now through pip for Python developers.