The prompt service over AIPromptRunner. Every template comes from the prompt rows in the database,
so a prompt edited or swapped in metadata takes effect on the next evaluation.
Run renders the evaluator prompt with the judge composed into its judgePrompt slot (the
runner's child-prompt composition, as a loop agent's system prompt and agent prompt), sends the
subject as its own user message, and returns the reply and the prompt run it wrote. ModelID
pins the model; ModelSelection Judge lets the judge's bindings choose it.
RenderCriteria renders the criterion prompt once per criterion, with no model call.
Preview returns the composed system prompt Run would send, with no model call.
The prompt service over AIPromptRunner. Every template comes from the prompt rows in the database, so a prompt edited or swapped in metadata takes effect on the next evaluation.
judgePromptslot (the runner's child-prompt composition, as a loop agent's system prompt and agent prompt), sends the subject as its own user message, and returns the reply and the prompt run it wrote. ModelID pins the model; ModelSelectionJudgelets the judge's bindings choose it.