What problem does it solve? AI coding agents often validate their own work, creating self-reinforcing loops where generated code and generated tests confirm the same assumptions. This Skill introduces independent adversarial review with fresh, context-isolated subagents while preventing unbounded review loops, scope drift, and speculative repairs. ## Core Features & Use Cases - Fresh Adversarial Review: Spawns isolated reviewer subagents that receive only a neutral navigation packet and attack assumptions, failure paths, usability, and implementation completeness. - Deterministic Review Governor: Enforces a two-round automatic budget, evidence authority levels (E0-E4), scope-drift detection, and convergence checks before any repair or extra round. - Auditable Review Trail: Writes a structured Markdown report under /vs_review/ recording reviewer launches, findings, triage decisions, and closure status. - Use Case: After implementing a payment retry flow, run an adversarial review where a fresh subagent challenges concurrency and idempotency assumptions, then let the governor decide whether a single closure round is justified. ## Quick Start Ask the agent to run an adversarial subagent review of the current implementation before treating the task as complete.
