What is auto-claude-code-research-in-sleep?

ARIS (Auto-Research-In-Sleep) is a collection of workflows for running AI-driven machine learning research automatically while you are away from your computer. The core idea is that you set up a research task before sleeping, and by the time you wake up, an AI agent has reviewed your paper or experiment, identified weaknesses, run follow-up tests, and rewritten parts of the narrative. The project started as a set of skills for Claude Code, an AI coding tool, but has since grown to work with other AI agents and coding environments. The system is built entirely from plain Markdown files called skills. There is no framework to install, no database to configure, and no background process to keep running. Each skill file describes a workflow in plain text that any AI agent can read and follow. This design means you can swap out which AI tool drives the research, whether that is Claude Code, OpenAI Codex, Cursor, or others, without rewriting anything. A distinctive feature is the cross-model review loop. Rather than having one AI model evaluate its own output, ARIS routes the work to two different models: one acts as the main executor that does the research and writing, and a second acts as an independent critic that looks for flaws. The project argues that a single model reviewing its own work creates blind spots, while using two models with different strengths catches more problems. The project includes dozens of bundled skills covering tasks like idea discovery, experiment automation, persistent knowledge storage in a research wiki, and a self-optimization mode where the system analyzes its own past behavior and proposes improvements. A standalone command-line tool called ARIS-Code is also available for users who want a full interactive experience outside of any specific coding environment. The system supports many AI providers including Anthropic, OpenAI, DeepSeek, MiniMax, and others, and can run against local models through LM Studio or Ollama. The full README is longer than what was shown.