ARIS (Auto-claude-code-research-in-sleep)

ARIS is an open-source collection of research workflow Skills by GitHub user wanshuiyin; the full name means "let Claude Code do research while you sleep." One AI agent does the work and a model from another company reviews it independently, with Codex as the default reviewer. The repository holds 84 main Skills arranged into workflows: finding ideas and refining methods, bridging to experiments and automatic improvement loops, writing the paper (planning, figures, LaTeX, compiling, revising), answering reviewers, resubmitting to a new venue and preparing conference talks. You can run a whole workflow or just one part. Besides Claude Code, it supports Codex, Cursor, Copilot CLI and other tools. What sets it apart is that the split between the one doing the work and the one reviewing it runs through every workflow.

Who it's for:researchers who know research methods, use the command line and can manage their own models and compute. If you don't write code, this isn't for you.

Before you start: install it through the plugin or install script and pick groups selectively rather than installing everything; the paper-writing workflow needs a local LaTeX setup. We recommend only the text-only parts: literature research, paper writing and rebuttals. We don't recommend the automatic experiment and improvement loops; remote servers or rented cloud GPUs (Vast.ai, Modal and others; the author's docs say budget settings are only warnings, not hard caps, and idle instances can keep billing); Overleaf sync, which uploads your manuscript, sets git credentials and installs a commit hook; or outbound features such as Feishu notifications. Both the working model and the reviewing model are billed under their own subscriptions or APIs. Our review didn't read every Skill one by one. MIT license.

Based on the project's public documentation. Not tested by AI Inspiration World.