auto-review-loop-minimax

Automating iterative code review for ML research uses an autonomous external review loop with MiniMax, running multiple rounds until a positive assessment or maximum rounds are reached. It persists state to REVIEW_STATE.json for traceability.

An autonomous external review loop orchestrates repeated evaluation rounds for ML research drafts and experiments using MiniMax as the reviewer. It applies structured workflow thresholds like MAX_ROUNDS and POSITIVE_THRESHOLD for automated decision making.

To start an automated multi-round MiniMax review loop, trigger the workflow by saying "start auto-review minimax". The system then executes review rounds, logging stages to AUTO_REVIEW.md after each iteration.

The MiniMax autonomous review loop does not strictly require an MCP tool. It uses MCP-based MiniMax chat when available, but includes a curl fallback for direct API access to ensure the external review process runs.

Use an automated external review loop for ML conference submissions when rigorous iterative critique is needed before finalizing drafts. It applies round-based decision making to improve experiments and submissions until a positive assessment threshold is met.

Limitations of a MiniMax-based autonomous review loop include a hard stop at MAX_ROUNDS if the POSITIVE_THRESHOLD is not met. It relies on external API integration and may require manual intervention if iterative improvements fail to produce a positive assessment.