Автоматический цикл рецензирования исследований через LLM

★ 7.1 · research

auto-review-loop-llm is a Claude Code skill that autonomously runs an iterative research review cycle using any OpenAI-compatible LLM API. It executes up to 4 rounds of review → fix → re-review, stopping when the external reviewer returns a score of ≥ 6/10 and a verdict of "ready" or "almost". Supported providers include OpenAI, DeepSeek, MiniMax, Kimi (Moonshot), ZhiPu (GLM), SiliconFlow, Qwen, and others — configured via the `llm-chat` MCP server or environment variables. Round state is persisted to `review-stage/REVIEW_STATE.json`, and the cumulative review log is written to `review-stage/AUTO_REVIEW.md`. Researchers who need automated iterative feedback on paper drafts without manual intervention between rounds will find this skill particularly useful.