What problem does it solve? Iterating on ML research without external critique leads to blind spots and weak submissions. This Skill runs an autonomous review-fix-re-review loop against the MiniMax API, giving your research repeated senior-reviewer feedback until it reaches a submission-ready score or hits the round limit. ## Core Features & Use Cases - Autonomous Review Loop: Runs up to 4 rounds of review, fix implementation, and re-review, stopping when the score reaches 6/10 with a "ready" or "almost" verdict. - Dual API Access: Uses the MiniMax MCP tool when available, with a curl fallback against the MiniMax chat completions endpoint. - State Persistence: Writes REVIEW_STATE.json after each round so the loop survives context compaction and can resume mid-run. - Use Case: You have a NeurIPS draft with experimental results. Trigger the loop, and it sends your claims and results to MiniMax-M3, implements the ranked fixes (code changes, experiments via SSH/screen, analysis), and re-reviews until the work is assessed as ready. ## Quick Start Ask the agent to run an auto review loop with MiniMax on the current research project and iterate until the reviewer scores it ready for submission.
