The Adversary skill is a specialized tool for Claude Code designed to perform high-rigor quality reviews outside of standard development loops. By mapping the criticality of a deliverable (from C1 to C4) to specific adversarial strategies like Red Teaming, Devil's Advocate, and Pre-mortems, it uncovers hidden flaws and strengthens technical arguments. It features an automated LLM-as-Judge scoring system based on a 6-dimension weighted rubric, ensuring that architecture designs, code artifacts, and documentation meet the highest quality standards before deployment.
主要功能
0116 GitHub stars
02Deep integration with project-specific quality-enforcement guardrails
03LLM-as-Judge rubric for objective, multi-dimensional quality scoring
04Standardized templates for Red Teaming, Steelman, and Devil's Advocate strategies
05Criticality-aware strategy selection ranging from routine to tournament-level reviews
06Standalone assessment mode for final validation of major technical deliverables
使用场景
01Running a Pre-mortem analysis to anticipate failure modes in new system designs
02Generating an objective quality score for an ADR before stakeholder approval
03Performing a C4 tournament review on mission-critical architecture proposals