aiming-lab/AutoResearchClaw ? reverse-engineered prompt

I want to build an autonomous research agent in Python. The core idea is that I can give it a simple topic or research idea as a chat message, and it will generate a complete, conference-ready academic paper from scratch.

The system should handle the entire research pipeline. It needs to find and use real literature from sources like arXiv and Semantic Scholar to build a literature review, making sure not to hallucinate citations. It should then be able to design and run its own experiments in a secure sandbox, analyze the results, and generate figures.

The final output should be a full paper formatted in LaTeX. I want it to have two modes: a fully autonomous "fire-and-forget" mode, and a "co-pilot" mode where it prompts me for guidance at key decision points, like approving an experiment design or co-writing the abstract. Make it a multi-agent system where different AIs collaborate on tasks like coding, peer review, and writing.