auto-review-loop

Automated iterative peer review repeatedly sends research context to a reviewer to obtain scores, apply minimum fixes, and re-review until acceptance or a maximum round limit. It supports NeurIPS/ICML-style critical scrutiny and experimental-result cross-checking across rounds.

To run an automated fix loop, execute the skill with your topic to obtain reviewer scores, apply requested minimum fixes, rerun experiments, and continue until positive assessment or maximum rounds. State persists in REVIEW_STATE.json for recovery.

Iterative research review requires Codex MCP routing or Codex exec for adversarial review. The loop depends on this routing to obtain reviewer scores, verdicts, and prioritized weaknesses.

Yes, you can cross-check experiment reruns across multiple review rounds. The loop implements reviewer fixes, reruns experiments, and persists cumulative review logs and state in REVIEW_STATE.json for compact recovery.

When iterative review hits the maximum round limit, the loop terminates. It stops applying minimum fixes and re-reviewing if the external reviewer has not reached a positive assessment.