ARIS
ARIS (Auto-Research-In-Sleep)
A skill set for autonomous machine learning research: literature review, idea discovery, experiments and paper writing with review by a separate model
High risk
We rate an entry high when the tool writes to external systems, handles money, production databases or secrets, or runs arbitrary commands. The CLI installs it only with your consent.
Why this level
- Autonomously runs experiments and executes code with minimal supervision
- Calls external models to review steps
Install
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/aris-auto-research-in-sleepDetects the agents on your machine, checks the risk and pins the version.
Other ways to install
Run one by one in the Claude Code chat
/plugin marketplace add wanshuiyin/auto-claude-code-research-in-sleep
/plugin install aris@arisInstall as a Claude Code plugin: claude plugin marketplace add wanshuiyin/Auto-claude-code-research-in-sleep, then claude plugin install aris@aris, then /aris:setup once and restart. Set up the reviewer model per the README.
Other ways from the author
claude plugin marketplace add wanshuiyin/Auto-claude-code-research-in-sleep
claude plugin install aris@arisAfter install run /aris:setup once and restart the agent, setup registers the bridge to the reviewer model.
This is third-party code. Review the repository files before installing.
What it does
ARIS gives the agent a machine learning research workflow assembled from separate Markdown skills: literature analysis, idea generation and selection, planning and running experiments, paper writing and preparing rebuttals. The core idea is that one executor drives a task while an independent model reviews it, by default through a bridge to Codex, so decisions get a second look. The set is not tied to a platform and runs as skills in Claude Code, Codex CLI, Cursor and other agents, and also installs as a plugin or a standalone CLI. The reviewer model can be swapped for other combinations, including ones hosted on ModelScope, to avoid Claude or OpenAI keys.
Who it is for. For machine learning researchers and engineers who run experiments and write papers together with an agent.
Good fit when
- You need a repeatable process from literature review to a paper draft
- You want research steps checked by an independent model
- You need to run long series of experiments with minimal supervision
Not a fit when
- The task is ordinary development rather than research and experiments
- You cannot connect a second model for review
Example request
Assemble a literature review on the topic and propose several research ideas checked by the second modelLimitations
The review scheme needs a second model: by default Codex, which requires an OpenAI account login, or alternative combinations including ones via ModelScope. Autonomous experiment runs execute code and can occupy a GPU and a lot of time, so results need watching. Some documentation is in Chinese, and not every page has an English version. MIT licensed.
How to disable. If installed as a plugin, remove aris via claude plugin. If installed as skills with the install_aris script, delete the created skill links from the project directory or from ~/.claude/skills/.
Security check
- Autonomously runs experiments and executes code with minimal supervision
- Calls external models to review steps
README in short
The README presents ARIS as a method for autonomous machine learning research built from skills, where an independent model reviews the executor. Installation comes three ways: skills via the install_aris script, a Claude Code plugin with the /aris:setup command, and a standalone ARIS-Code CLI. It documents adaptations for Codex CLI, Cursor, Trae, Antigravity, GitHub Copilot CLI and other agents, plus model combinations and running through ModelScope without major vendor keys. It includes integrations with Feishu, Obsidian, Zotero and a macOS monitor widget. MIT licensed.
FAQ
Do I need Codex and OpenAI?
No. Codex is the default reviewer, but the README describes other model combinations, including ModelScope-hosted ones that need no Claude or OpenAI keys.
Is it a platform or a method?
The author calls it a skill-based method rather than a platform. The same skills run across agents, and the plugin and CLI are just ways to install them.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent