ARIS (Auto-Research-In-Sleep)

Tracked

ARIS is a lightweight Markdown-only skills collection for autonomous ML research, enabling cross-model review loops, idea discovery, and experiment automation that works with any LLM agent.

Author wanshuiyin Open Sourced 2026-03-10 Last Commit Unknown

Overview

ARIS (Auto-Research-In-Sleep) is a set of plain-Markdown skills that turn an LLM coding agent into an autonomous research assistant. Instead of a framework or runtime, it ships opinionated workflows for cross-model paper review, research idea generation, and experiment automation. The skills are model-agnostic and drop into Claude Code, Codex, OpenClaw, or any agent that reads Markdown instructions.

Key Features

  • Cross-model review loops where multiple LLMs critique and refine each other's outputs
  • Research idea discovery and paper-review workflows for ML literature
  • Experiment automation that runs and iterates on ML jobs hands-off
  • Pure Markdown skills with no framework lock-in or extra dependencies
  • Compatible with Claude Code, Codex, OpenClaw, and general LLM agents

Use Cases

  • Running overnight autonomous literature reviews and paper critiques
  • Generating and stress-testing novel research ideas across models
  • Automating repetitive ML experiment and ablation pipelines
  • Building a self-improving research loop inside an existing coding agent

Technical Details

  • Markdown-only skill definitions — no daemon, runtime, or vendor SDK required
  • Designed around cross-model orchestration and iterative agent loops
  • Integrates via standard skill/tool-loading mechanisms in agents like Claude Code