Auto-claude-code-research-in-sleep
ARIS ⚔️ (Auto-Research-In-Sleep) — Lightweight Markdown-only skills for autonomous ML research: cross-model review loops, idea discovery, and experiment automation. No framework, no lock-in — works with Claude Code, Codex, OpenClaw, or any LLM agent.
ARIS (Auto-Research-In-Sleep) is a collection of 83 Markdown-based agent skills that automate end-to-end machine learning research workflows—from literature review and idea discovery through experiments, peer review, paper writing, and rebuttal. Skills are plain SKILL.md files that work across Claude Code, Codex CLI, Cursor, and other LLM agents, with cross-model review loops (Claude executes, GPT reviews via Codex MCP). It includes persistent research memory (Research Wiki), auto-experiment deployment to GPU servers, and optional integrations with Zotero, Obsidian, and Feishu for literature and notifications.
Key Features
Use Cases
- 01End-to-end ML research: run /idea-discovery to survey literature, brainstorm 8-12 ideas, pilot top ideas on GPU, then chain into /experiment-bridge and /paper-writing for a submission-ready PDF
- 02Overnight autonomous review loops: /auto-review-loop runs GPT-6-Astra reviews → Claude fixes/runs new experiments → repeats until score threshold, all while you sleep
- 03Cross-model paper writing: /paper-writing turns a narrative report into LaTeX, auto-generates figures, fetches real BibTeX, compiles PDF, and runs 2 rounds of GPT improvement
- 04Rebuttal drafting: /rebuttal parses peer reviews, builds strategy, optionally auto-runs supplementary experiments, drafts grounded responses, and runs GPT stress-tests before finalization
- 05Proof campaigns: /proof-orchestrator manages multi-session proof runs with local attempts, GPT Pro handoff packages, and cross-run continuation for hard theorems
- 06Research memory: /research-wiki init creates a persistent wiki that accumulates papers, ideas, and experiment results, preventing re-derivation of failed ideas
Related Skills
View moresuperpowers
An agentic skills framework & software development methodology that works.
skills
Skills for Real Engineers. Straight from my .agents directory.
skills
Public repository for Agent Skills
ponytail
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
Auto-claude-code-research-in-sleep — FAQ
What is ARIS?+
ARIS is a methodology and skill collection for autonomous ML research. It provides 83 Markdown-based agent skills that compose into full research workflows (idea discovery, experiments, review loops, paper writing, rebuttal). Skills are plain SKILL.md files with no framework dependency.
How do I install ARIS skills?+
Clone the repo, then run bash tools/install_aris.sh ~/your-project to symlink skills into .claude/skills/. For Codex projects, use install_aris_codex.sh. For Claude Code plugin, run /plugin install aris@aris. Update with git pull in the repo; new skills auto-reconcile on next installer run.
Which AI clients does ARIS work with?+
ARIS skills work with Claude Code, Codex CLI, Cursor, GitHub Copilot CLI, OpenClaw, Trae, and Antigravity. Cross-model review requires Codex MCP (ChatGPT subscription) or Oracle MCP (GPT-5.5 Pro). The executor can be any agent; the reviewer runs via MCP.
Do I need API keys or subscriptions?+
You need Claude Code (Claude subscription) or Codex CLI for execution. Cross-model review needs Codex MCP (ChatGPT Plus/Pro, no OpenAI API key) or Oracle MCP (optional GPT-5.5 Pro). Optional integrations: GEMINI_API_KEY for AI illustrations, EXA_API_KEY for Exa search, DeepXiv SDK for literature.
Is ARIS free and open source?+
Yes, ARIS is MIT-licensed and free. The skills are plain Markdown. You pay only for the underlying LLM subscriptions (Claude for execution, ChatGPT for Codex MCP review). Optional GPU experiments require your own hardware or Vast.ai rental.
How does cross-model review work?+
Claude Code (executor) writes code/papers, then calls GPT-6-Astra or GPT-5.5 Pro (reviewer) via Codex MCP or Oracle MCP. The reviewer sees raw files and provides independent critique, preventing single-model blind spots. Codex MCP needs ChatGPT subscription; Oracle MCP needs API key or browser mode.
How do I install Auto-claude-code-research-in-sleep?+
Open the source repository on GitHub and follow its README. Auto-claude-code-research-in-sleep is a skill — MCP Agents Market links you directly to the official repo.
Is Auto-claude-code-research-in-sleep free?+
Auto-claude-code-research-in-sleep is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.