OptionalcontextOptional additional context for the judge, such as filesystem state,
command outputs, or structured observations collected during custom
assertion steps. Included in the evaluation prompt under
<additional_context>. Useful when conversation messages contain raw
tool-call JSON that is difficult for the judge to interpret directly.
OptionalcriteriaOptional criteria to evaluate, overriding the judge agent's configured criteria.
Encapsulates a request for the judge agent to evaluate the conversation.
When present on AgentInput, signals the judge to produce a verdict. Optionally carries inline criteria that override the judge's own criteria.