What problem does it solve?
This skill addresses the limitation of standard code reviews where agents often rubber-stamp changes; it provides an adversarial, local-compute-based review process that actively attempts to break code to uncover hidden bugs and security vulnerabilities.
Core Features & Use Cases
- Adversarial Fan-out: Spawns independent subagents tasked specifically with breaking the code across dimensions like correctness, security, and regression risk.
- Verification Loop: Enforces a strict refute-by-default policy where every finding must be verified against the actual code before being accepted.
- Use Case: Before merging a sensitive authentication change or a complex feature, run this skill to ensure that multiple specialized agents have stress-tested the logic locally without requiring external CI services.
Quick Start
Run the review skill on the current unstaged diff to perform an adversarial check for security and correctness.