auto-review-loop-minimax

You can automate external review of ML research drafts by running a multi-round loop that critiques claims, methods, and results via the MiniMax API until a positive assessment is reached. The loop persists state in REVIEW_STATE.json and logs history in AUTO_REVIEW.md.

A round-based AI review loop iteratively critiques research proposals and methods, applying fixes between rounds. It uses the MiniMax API to assess drafts, automatically resuming in-progress runs from saved state until the maximum rounds are reached or a positive assessment is returned.

Yes, you can use the MiniMax API for iterative critique without an MCP tool by utilizing the curl fallback access mode. This allows environments without MCP support to execute the external review loop directly.

To persist state across multiple rounds of AI review, the workflow automatically saves progress in REVIEW_STATE.json. This enables automatic resume of in-progress runs and includes fresh-start safeguards to prevent data conflicts.

Yes, the autonomous review loop supports automatic resume for in-progress runs. It uses a phase-driven workflow with state persisted in REVIEW_STATE.json, allowing the review process to continue exactly where it left off across sessions.

You should avoid using an autonomous external review loop if your environment lacks both MCP tool support and curl access to the MiniMax API. Additionally, it is not suited for tasks requiring fewer than multi-round iterative critiques of research drafts.