adversarial-validator

Adversarial audits review artifacts across Architecture, Implementation, and Quality phases to catch issues self-validation misses. This skill classifies findings by severity, generates structured YAML or JSON reports, and optionally applies automated corrections.

An adversarial audit is a post-gate review process that identifies high-severity problems missed during standard validation. It evaluates phase artifacts and classifies issues into CRITICAL, GRAVE, MEDIUM, and LIGHT severity levels to reduce post-release risks.

You can configure fail and warn thresholds to control gate enforcement during artifact reviews. The skill evaluates phase outputs against these limits and integrates with post-gate hooks to halt progression when critical severity limits are exceeded.

Yes, the tool applies to Architecture phase 3, Implementation phase 5, and Quality phase 6 tasks. It reviews artifacts from these specific SDLC stages and leverages LLM-assisted analysis to surface hidden issues.

The skill produces automated reports structured in YAML or JSON formats. These outputs document classified findings and integrate with post-gate hooks to support continuous integration workflows.

LLM-assisted analysis automates issue detection but requires configured fail and warn thresholds to enforce quality gates. It serves as a post-gate adversarial check rather than replacing the initial self-validation process.