auto-review-loop-llm

You can automate iterative research review by repeatedly scoring, critiquing, and prescribing minimum fixes for a draft until it meets a positive readiness threshold. This loop applies to paper-writing workflows with configurable maximum rounds and persistent state recovery.

An LLM-based review loop for experiment planning is an automated cycle that scores and critiques weaknesses in a research draft, prescribes minimum-change fixes, and enforces re-validation across multiple rounds until an external reviewer's verdict is positive.

You need an OpenAI-compatible LLM accessible via an MCP llm-chat server to run the iterative research review loop, though a curl fallback is supported if the MCP server is unavailable. No additional dependencies are required.

You can recover progress after an interruption because the automated review loop saves raw reviewer responses, cumulative review logs, and a compact recovery state to review-stage logs, ensuring progress survives across multiple rounds.

Limitations of the automated review loop include a configurable MAX_ROUNDS termination that stops the process if the positive readiness threshold is never met, and reliance on an OpenAI-compatible LLM to accurately rank remaining weaknesses and prescribe fixes.

The best way to iterate research reviews until venue-ready is to automate the review loop, letting an LLM score weaknesses and prescribe minimum-change fixes, then re-validate changes across multiple rounds with persistent logging until readiness is achieved.