codex-review-loop

Adversarial code review uses an AI model to analyze uncommitted changes or pull requests, identifying bugs, security flaws, and convention violations. It automatically applies approved fixes, re-verifying quality iteratively until the codebase reaches a clean state.

Integrating automated code fixes into CI/CD pipelines catches common code quality issues before production. The adversarial review loop analyzes findings, facilitates human-in-the-loop approval for fixes, and iteratively applies them to maintain consistent code standards.

Codex CLI is required to execute the adversarial code review and fix loop. The skill depends on this command-line interface to perform AI analysis, apply atomic fixes, and verify code quality against safety limits.

The automated fix loop facilitates human-in-the-loop approval by parsing AI output into actionable, categorized findings. Users review these structured findings before the system enters an atomic fix-verify-commit loop to apply the approved changes.

AI code review loops are bounded by safety limits that halt the process if a clean state is not reached. The adversarial review depends on AI model accuracy and requires adherence to existing project conventions for accurate severity assessment.