latent.agents.judges.guardrails

GuardrailsJudge -- scores whether a system refused or deflected a request.

Classes

GuardrailsScore

GuardrailsScore()

Score schema for refusal detection.

Functions

GuardrailsJudge

GuardrailsJudge(model: str, kwargs = {}) -> Judge[GuardrailsScore]

Factory returning a Judge that detects refusal/deflection.

Attributes

PROMPT_TEMPLATE

SYSTEM_PROMPT