What problem does it solve?
It reduces the risk of publishing weak research by running an autonomous multi-round loop that sends your work for external review, applies the reviewer’s fixes, and re-submits for re-evaluation until quality criteria are met.
Core Features & Use Cases
- Autonomous multi-round review loop: review → implement fixes → re-review until a positive assessment or MAX_ROUNDS is reached.
- MiniMax-based external reviewer integration: uses an MCP tool when available or a curl-based API fallback when it is not.
- State persistence for recovery: writes and overwrites review-stage/REVIEW_STATE.json to resume safely after context compaction or interrupted runs.
Use case: You have a research project that needs ICML/NeurIPS/ICLR-level critique; run the loop to collect a ranked list of weaknesses, implement the minimum fixes, and document each round in review-stage/AUTO_REVIEW.md.
Quick Start
Trigger an autonomous review loop for your current research topic by asking: auto review loop minimax <topic-or-scope>.