Auto-claude-code-research-in-sleep
ARIS (Auto-Research-In-Sleep) is a set of custom Claude Code skills for autonomous machine-learning research. The idea is cross-model collaboration: Claude Code drives the research while an external LLM, by default GPT-5.4 at extra-high reasoning reached through Codex MCP, acts as the reviewer. The README argues that a single model reviewing its own work tends to fall into local minima and that two complementary models, fast execution plus rigorous critique, do better.
It offers 18 composable skills chained into workflows: idea discovery (literature survey, brainstorming 8 to 12 ideas, novelty checks, GPU pilot experiments, a ranked report), an auto review loop of four autonomous review rounds, a paper writing pipeline from outline to figures, LaTeX and PDF, peer review of other papers, and GPU deployment with rsync, screen sessions and multi-GPU runs. Human-in-the-loop checkpoints are configurable (AUTO_PROCEED), model combinations beyond Claude and GPT are supported, and optional Zotero, Obsidian and Feishu/Lark integrations exist. Install by copying the skills into ~/.claude/skills and adding the Codex MCP server.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-11-06
- Commits, last 90 days
- 100+
- Releases, last 90 days
- 7
- Latest release
- v0.4.28 (2026-09-28)
- Open issues
- 75
- Recent issues with at least one comment
- 7/11
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Most discussed open issues
- #115 Code版本Windows平台25 comments · 2026-04-07
- #240 README写的太繁杂了,可读性差,希望精简流程,增加标题层级编号6 comments · 2026-05-22
- #430 严重bug,停不下来4 comments · 2026-09-03
- #111 请问独立安装版相较于普通版有什么优势?4 comments · 2026-04-05
- #433 [Codex] Mark workflow-only skills as explicit-only to reduce routing/context overhead3 comments · 2026-09-07
Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.
Who it fits
- You use Claude Code and want an autonomous skill loop for machine learning research that ideates, executes experiments, and reviews results overnight.
- You want a dual-model peer-review mechanism where a secondary LLM audits research ideas and code before running.
- You need an autonomous web-search deep research agent for general factual reports; GPT Researcher is tailored for text reports.
- You do not use Claude Code CLI or prefer a graphical node-based workflow builder.
How it differs
Auto-claude-code-research-in-sleep automates iterative ML coding, training scripts, and experiment loops via Claude Code. GPT Researcher focuses on web research, source aggregation, and structured factual reporting.
Compatibility
Quick start
Use cases
Alternatives
Comments
No comments yet. Be the first!