Auto-claude-code-research-in-sleep

ARIS (Auto-Research-In-Sleep) automates ML research overnight. You give Claude Code a research question or experiment, walk away, and come back to results, with cross-model review gates designed to catch hallucinations and fabricated citations along the way.

It's a Markdown-only skills bundle, no framework, no lock-in. Claude Code is the driver, but the review loop pulls in GPT, DeepSeek, and other providers to check each other's work. The skill set (67+ bundled) covers literature review, idea discovery, experiment automation, paper writing, and rebuttal drafting. MIT licensed, with hardening that came from real NeurIPS submission cycles, not vibes.

Solo ML researchers and small academic teams running into the same paper-writing grind every conference cycle get the biggest win. Industry research teams with full MLOps platforms probably don't need this. The cross-model review piece is the differentiator. Most overnight-agent setups skip it and ship hallucinated citations.

The catch: autonomous overnight loops only work if your constraints are tight. Vague goals produce vague output and burn through credits. And the cross-model review means every iteration costs multiple LLM calls instead of one. Budget accordingly, especially if you're routing to GPT or Claude Opus for the reviewer role.