What problem does it solve? Designing a validator LLM that reliably approves, blocks, revises, or escalates agent actions is hard: vague criteria produce judges that rubber-stamp risky actions or block everything. This Skill guides you through writing a rigorous judge system prompt with explicit, testable decision rules. ## Core Features & Use Cases - Structured Judge Prompt Generation: Produces a complete judge system prompt with role definition, input expectations, a criteria checklist (authorization, evidence, exposure/risk, policy), and four decision outcomes (ALLOW, BLOCK, REVISE, ESCALATE). - Anti-Gaming Protections: Builds in instructions so the judge evaluates structured claims against evidence rather than the persuasiveness of the actor's prose, and escalates under uncertainty instead of defaulting to allow. - Implementation Guidance: Delivers runtime integration notes covering outcome handling, logging, and known limitations with recommendations for complementary checks. - Use Case: After defining judge criteria and an action proposal format for an agent that sends outbound emails, use this Skill to produce the validator prompt that inspects each proposed email before execution. ## Quick Start Use the judge-prompt-writer skill to write a judge system prompt for my agent that reviews outbound email action proposals.
