Scoped implementation-quality review. Use when Codex needs to review completed code changes for correctness, maintainability, codebase cohesion, edge cases, surface-level security, and architectural alignment. Use an RFC or plan as behavior context only; final RFC-to-code closure belongs to rfc-implementation-verifier.
Final RFC-to-code closure review. Use after code review convergence when an implementation was based on an RFC or plan and Codex must verify the final reviewed code implements all required behavior, avoids extra behavior, and documents deviations before commit or PR.
Review-loop convergence diagnosis. Use when repeated code review or gating findings suggest unresolved architecture, requirement, invariant, ownership, or abstraction problems rather than independent bugs.
Strict dirty-tree gating review. Use when Codex needs to review only the uncommitted changes in the current repository, investigate the surrounding code, and emit tightly qualified findings in the repository's gating-review JSON style.
Subtractive plan minimality review. Use after RFC or plan review convergence to identify non-load-bearing content, inflated site lists, premature abstraction, or defensive structure not justified by the stated scope or prior blocking findings.
Adversarial technical plan stress testing. Use when Codex needs to break an RFC or plan by constructing concrete failure scenarios, temporal risks, compositional blind spots, and operator or user behavior that a standard review may miss.
Structured technical RFC and plan review. Use when Codex needs to audit an RFC, implementation plan, or final pre-implementation sign-off for soundness, completeness, reuse, alternatives, and plan-altitude correctness.
Security audit and attack-surface decomposition. Use when Codex needs to review code changes, modules, dependencies, privileged operations, network boundaries, parsing, filesystem use, authn/authz, crypto, subprocesses, or supply chain risk.