This skill runs an adversarial, three-agent code review pipeline: a Hunter (maximize recall to find potential bugs), a Skeptic (challenge and attempt to disprove findings with risk-adjusted scoring), and a Referee (independently adjudicate and produce a definitive report). The design enforces strict context boundaries so each phase analyses code independently, reducing confirmation bias and producing high-fidelity bug reports with structured BUG-IDs and severity scoring.
Use for pre-merge security audits, large PR reviews, legacy code audits, or any high-stakes module where false negatives are costly. It supports scanning entire projects, directories, individual files, or branch-diff mode (files changed between branches).
Agents that can spawn isolated subagents, read code, and pass structured data between phases (e.g., orchestration-capable assistants or platforms supporting isolated runtimes).
This skill has not been reviewed by our automated audit pipeline yet.
