auto-review-loop-minimax

Automating iterative peer review for ML research ideas requires an autonomous workflow that sends ideas to an external reviewer, parses feedback, and applies fixes across multiple rounds. This skill manages that loop up to a defined maximum limit.

State persistence in a multi-round review loop works by reading and writing progress to REVIEW_STATE.json and AUTO_REVIEW.md files. This ensures the workflow tracks rounds, retains feedback, and maintains guardrails across continuous evaluation cycles.

Yes, you can use MiniMax for automated external research review instead of Codex MCP. This skill sources prompts from the MiniMax API, utilizing the MCP pathway when available and falling back to curl requests.

The best way to run a multi-round assessment workflow with automated fixes is to use a stateful loop that requests reviews, integrates feedback, and re-evaluates changes. This skill automates the entire process up to the defined MAX_ROUNDS limit.

No specific dependencies are required to run this autonomous external review workflow. The skill operates independently, managing the multi-round assessment loop and state persistence without relying on external component installations.

Limitations of autonomous multi-round review loops include the MAX_ROUNDS guardrail, which caps the number of iterative assessments, fixes, and re-evaluations. The workflow also depends on MiniMax API availability for sourcing review prompts.