codex-review

Adversarial plan review uses multi-round dialogue between Codex and CC to identify vulnerabilities, ambiguities, and infeasible points in a P-stage plan. It applies template-driven validation to rigorously test the plan before execution.

Yes, template selection is automated by matching plan.md keywords to the appropriate template, such as api-review, frontend-review, infra-review, or general-review. This ensures the adversarial review uses the correct validation criteria.

Multi-round validation works by iterating dialogue between Codex and CC until the plan reaches an approved or needs-modification state. This loop rigorously exposes flaws and feasibility issues before generating a final verdict.

Review results are persisted as a structured document at reviews/review-{REVIEW_ID}.md. This file contains the final verdict, round-by-round critique, and actionable remediation suggestions for the plan.

No external dependencies are required to run automated plan review. The process relies entirely on built-in template matching and multi-round dialogue loops between Codex and CC to validate P-stage plans.

The best way to validate feasibility issues is using template-driven adversarial review with Codex. It performs multi-round critique to identify vulnerabilities and generates actionable remediation suggestions for the plan.