What problem does it solve? Research projects need repeated external critique and revision before submission, but manually orchestrating review rounds, implementing fixes, and tracking progress is slow and error-prone. This Skill runs an autonomous loop that reviews work via Codex MCP, implements the suggested fixes, and re-reviews until the work passes or a round limit is reached. ## Core Features & Use Cases - Autonomous Review Cycles: Sends full research context to an external reviewer (Codex MCP with high reasoning effort), parses scores and verdicts, and iterates up to 4 rounds. - State Persistence & Recovery: Writes REVIEW_STATE.json after each round so the loop survives context compaction and can resume in-progress runs. - Structured Review Log: Maintains a cumulative AUTO_REVIEW.md with verbatim reviewer responses, actions taken, and results per round. - Optional Human Checkpoint: Can pause after each review to let the user approve, modify, skip, or stop suggested fixes. - Use Case: A researcher preparing a NeurIPS submission invokes the loop; it gets a 5/10 review, implements the ranked fixes, launches experiments, re-reviews, and stops when the reviewer scores it 7/10 and marks it ready. ## Quick Start Run the auto review loop on my current research draft and keep iterating until the reviewer says it is ready for submission.
