auto-review-loop-llm

Automate LLM research paper review by running an autonomous loop that scores drafts, extracts ranked weaknesses, and applies minimum fixes until acceptance thresholds are met or maximum rounds are reached.

The autonomous review loop re-assesses research by applying evidence-aligned minimum fixes and then prompting the LLM reviewer to re-evaluate the updated draft, generating a new score and verdict for each round.

Yes, automated critique and experiment iteration require an OpenAI-compatible LLM integration via an llm-chat MCP server, with a curl fallback available for environments lacking direct MCP access.

Track reviewer scoring and actions through persistent JSON recovery state files and cumulative markdown documentation that log each round's raw reviewer output, numeric scores, and applied fixes.

Limitations of using a self-healing LLM loop for paper writing include a hard cap on maximum review rounds, meaning the process stops when iterations are exhausted even if acceptance thresholds are not fully met.

Citation verification in LLM-assisted research review workflows ensures the evidence-aligned re-assessment process maintains factual accuracy, directly supporting the minimum required fixes extracted by the reviewer.