auto-review-loop

An autonomous review loop orchestrates iterative evaluation cycles where a secondary Codex agent scores, critiques, and improves research artifacts until achieving a positive assessment or hitting a maximum round limit.

You can automate iterative code reviews by providing a draft artifact to the loop, which then generates actionable fixes, executes them, and logs each round until the artifact passes assessment.

Yes, the autonomous review workflow persists round state, scores, and pending actions using REVIEW_STATE.json, allowing the evaluation cycles to resume smoothly after unexpected interruptions.

Yes, the iterative review loop supports reproducibility and auditability by integrating with AUTO_REVIEW.md and maintaining round-based logs that document every evaluation, critique, and action-item execution.

A key limitation of an autonomous review loop is the maximum round limit; if a research paper requires extensive revisions beyond this threshold, the artifact may not achieve a positive assessment without manual intervention.