codex-review-gate

Cross-model code reviews using OpenAI Codex are automated by running a review gate against a project root, which surfaces design trade-offs, concurrency concerns, and risk areas. The gate executes the review and outputs structured findings for downstream workflows.

Codex code review commands support JSON output when integrated with the required Codex companion script. Running the review in JSON mode produces structured results that downstream tooling can consume for automated processing or escalation.

Scoping code review diffs to a specific base commit is supported by providing an optional base SHA. This restricts the cross-model evaluation to changes between the specified commits or worktrees, ensuring the review focuses only on relevant branch modifications.

The Codex review gate does not automatically fix code; it acts as a caller-agnostic primitive that surfaces findings. The caller decides downstream actions, such as triggering an auto-fix loop, escalating to a human, or ignoring the results.

Running cross-model code reviews in repository workflows requires a Codex companion script present in the Claude plugin. This script executes standard review commands in JSON mode to evaluate changes and surface actionable findings across branches and pull requests.

A cross-model review gate is ideal for repository workflows requiring evaluation of changes across models, including risk assessment and design tradeoffs. Use it when checkpoints demand structured, JSON-formatted findings for downstream automation rather than manual review.