OptionalcustomPrompt: stringOptional override for the default judge prompt
ProtectedbuildBuild the judge prompt request from context. Protected so Layer 2 (MJLLMJudge) can override to use MJ prompt entity templates.
ProtectedCreateCreate a verdict with standard fields populated. Utility for subclasses to avoid boilerplate.
Evaluate the current state and determine whether the goal is met.
Full context including goal, screenshots, step history
Verdict with Done, Confidence, Feedback, and Reason
ProtectedparseParse the LLM response into a JudgeVerdict. Protected so Layer 2 can override if the prompt format changes.
Expects JSON with: done (boolean), confidence (number), reason (string), feedback (string). Handles both raw JSON and JSON wrapped in markdown code blocks.
Callback to execute the judge prompt (provided by the engine)