Browser Use 1.0.5 judges every agent run with your own model: success, failure reason, prompt-injection flags and site gotchas. Python code and tips.