# ARIS

> Research skills that take your agent from the literature to ideas, GPU experiments, a paper and a rebuttal, with a second model reviewing.

ARIS (Auto-Research-In-Sleep) is an open-source research workflow by Yang Ruofeng, packaged as a Claude Code plugin. Its skills are plain Markdown and take an agent through the life of a machine-learning paper: reading the literature, finding and testing ideas, running experiments, writing the paper and replying to reviewers. Claude does the work and a second model reviews it, OpenAI's Codex by default, so the agent isn't grading its own work.

What you can do with it
- Survey a research direction, brainstorm ideas, check them for novelty and rank them with short pilot experiments (/aris:idea-discovery)
- Turn an experiment plan into code, deploy it to your GPU and collect the results (/aris:experiment-bridge)
- Leave a review loop running overnight: the reviewer scores the paper, Claude fixes what it flags and runs the experiments it asks for, and it repeats until the score clears a threshold or the rounds run out (/aris:auto-review-loop)
- Go from a narrative report to a compiled LaTeX paper, with real BibTeX from DBLP and CrossRef instead of made-up citations (/aris:paper-writing)
- Draft a rebuttal within a venue's character limit that won't invent results or promise anything you haven't approved (/aris:rebuttal)
- Turn an accepted paper into slides, a poster or a talk, and keep a research wiki that remembers papers, ideas and failed experiments across sessions

How to install it
In Claude Code, run /plugin marketplace add wanshuiyin/Auto-claude-code-research-in-sleep, then /plugin install aris@aris. Run /aris:setup once and restart Claude Code: it registers the Codex reviewer bridge that ships inside the plugin. Skills are typed with the aris: prefix, such as /aris:research-pipeline "your research direction".

What it needs
- OpenAI's Codex CLI, installed and signed in (npm install -g @openai/codex, then codex login), for the review skills, plus Python 3.9 or later for the bridge
- A LaTeX setup with latexmk and pdfinfo for paper writing; idea discovery and the review loop don't need it
- A GPU for experiments: your own machine, a server over SSH, or one rented from Vast.ai. Without one, the review and writing skills still work and experiments are flagged for you to run
- Optional: a Gemini API key for AI illustrations (Mermaid diagrams need none), Exa or Firecrawl keys for extra web search, and Zotero, Obsidian or Feishu if you use them

Good to know
ARIS is a community project, not an official Anthropic or OpenAI tool. Its author says it speeds research up but doesn't replace your own judgement, so check its ideas and claims yourself. Other models, such as GLM, MiniMax, Kimi or DeepSeek, can stand in for the Codex reviewer.

- Type: Plugin
- Made by: wanshuiyin
- Category: Research
- Formats: Claude plugins
- Version: latest
- Website: https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep
- Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep
- Page: https://pluginboard.ai/plugins/aris

## Works with

- Claude
- GitHub Copilot
- Claude Code
- OpenClaw

## Install

### Claude

In Claude (web or desktop): Customize → Plugins → Add → Add marketplace, enter wanshuiyin/Auto-claude-code-research-in-sleep, then add aris

### Claude Code

In Claude Code:

```text
/plugin marketplace add wanshuiyin/Auto-claude-code-research-in-sleep
/plugin install aris@aris
```

---
From Pluginboard (https://pluginboard.ai).