Aris Autonomous Ml Research

---
name: aris-autonomous-ml-research
description: Skill for using ARIS (Auto-Research-In-Sleep) — a Markdown-only autonomous ML research workflow system that orchestrates cross-model LLM collaboration for paper review, idea generation, experiment automation, and research writing.
triggers:
  - "set up ARIS for autonomous research"
  - "run research pipeline while I sleep"
  - "automate ML paper writing with Claude Code"
  - "cross-model LLM research review loop"
  - "use ARIS to generate research ideas"
  - "run experiment bridge with ARIS"
  - "write rebuttal with ARIS"
  - "autonomous research workflow with Claude Code skills"
---

# ARIS — Autonomous ML Research In Sleep

> Skill by [ara.so](https://ara.so) — Daily 2026 Skills collection.

ARIS is a **zero-dependency, Markdown-only** autonomous ML research workflow system. There is no framework to install, no database, no daemon — every "skill" is a plain `SKILL.md` file that any LLM agent can read and execute. ARIS orchestrates **cross-model collaboration**: one model (e.g. Claude Code) executes research tasks while another (e.g. Codex / GPT-5.4) acts as a rigorous adversarial reviewer, catching blind spots that self-review misses.

Works with: **Claude Code**, Codex CLI, Cursor, Trae, Antigravity, Windsurf, or any LLM agent.

---

## Installation

### 1. Clone the Repository

```bash
git clone https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep.git
cd Auto-claude-code-research-in-sleep

2. Install Skills into Claude Code

Copy the skills directory so Claude Code can discover them:

# Install all core skills globally
cp -r skills/ ~/.claude/skills/

# Or symlink for live updates
ln -s $(pwd)/skills ~/.claude/skills/aris

For Codex CLI, skills live under skills/skills-codex/:

cp -r skills/skills-codex/ ~/.codex/skills/

3. Configure the Cross-Model Reviewer (MCP)

ARIS uses an MCP server to call the reviewer model. Add to your claude_desktop_config.json or ~/.claude/config.json:

{
  "mcpServers": {
    "codex": {
      "command": "npx",
      "args": ["-y", "@openai/codex-mcp"],
      "env": {
        "OPENAI_API_KEY": "$OPENAI_API_KEY"
      }
    }
  }
}

For alternative reviewers (no OpenAI required), use the bundled llm-chat MCP server:

cd mcp-servers/llm-chat
npm install

Then configure your preferred OpenAI-compatible endpoint:

{
  "mcpServers": {
    "llm-chat": {
      "command": "node",
      "args": ["mcp-servers/llm-chat/index.js"],
      "env": {
        "LLM_API_KEY": "$YOUR_API_KEY",
        "LLM_BASE_URL": "https://api.your-provider.com/v1",
        "LLM_MODEL": "your-model-name"
      }
    }
  }
}

Tested reviewer models: GLM-5, MiniMax-M2.7, Kimi-K2.5, LongCat, DeepSeek.

Core Workflows

ARIS has four main workflows, each invoked by a slash command in Claude Code.

Workflow 0 — Full Pipeline (Idea → Paper)