judge-architecture-reviewer

What problem does it solve? Agent systems often lack a sound judgment layer: actions go out without validation, failure modes like specification gaming or escalation drift go undetected, and memory has no provenance. This Skill audits a running or planned agent system for judge-layer gaps, failure-mode risks, and architectural weaknesses, then delivers a concrete remediation plan. ## Core Features & Use Cases - Judge Placement Audit: Maps where judgment exists (before action, after action, at handoffs, at delivery) against every boundary where work can go wrong. - Failure Mode Assessment: Rates risks such as correlated judgment, specification gaming, escalation drift, latency/cost, and prompt drift as low, medium, or high. - Memory, Provenance & Human Review Assessment: Evaluates whether memories distinguish observed facts from agent inferences and whether human review coverage matches risk. - Use Case: You have a multi-agent system that sends emails and writes to a database. Describe it to the Skill, and receive a structured architecture review with a remediation roadmap prioritized by consequence and effort. ## Quick Start Use the judge-architecture-reviewer skill to review my agent system's judge architecture and produce a prioritized remediation plan.