Claude Skill MartBrowse skillsWhat is a Skill?
 
← Back to list

Autoresearch (Autonomous Research Orchestrator)

Runs an entire AI research project autonomously with a two-loop experiment/synthesis engine, from literature survey to paper draft.

AutomationAdvanced★ 11,544⑂ 842AI score 7/10Last updated: Jun 16, 2026

What it does

Turns Claude into a research project manager that orchestrates while domain skills execute.

  • Inner loop: pick hypothesis → lock protocol in git → run experiment → sanity check → record metric → learn
  • Outer loop: cluster results, explain mechanisms, update findings.md, and decide DEEPEN / BROADEN / PIVOT / CONCLUDE
  • Bootstraps a standard workspace (literature/, experiments/, data/, to_human/, paper/) plus persistent state files
  • Routes work to domain skills (data processing, fine-tuning, distributed training, evaluation, interpretability, MLOps)
  • Keeps running via Claude Code /loop 20m or an OpenClaw 20-minute cron, shipping HTML/PDF progress reports
  • Calls ml-paper-writing at the end for a venue-formatted draft with citation verification

Who it's for

  • ML/AI researchers, PhD students and research engineers running many experiment iterations
  • People who want an agent to keep making progress overnight
  • Teams that struggle to keep hypotheses, negative results and lessons organized
  • Overkill if you only need a single one-off task

Examples

  1. "Investigate what accelerates grokking in small transformers" — survey papers into literature/, form H1–H3 with predictions, commit protocols before runs, reflect every 5–10 experiments.
  2. Benchmark optimization — track run_014 → metric 0.847 vs baseline 0.812 trajectory JSON, then auto-generate the progress curve plot in a to_human/ report.
  3. Concluding a project — once findings.md reads like a paper backbone, hand off to the paper-writing skill for a NeurIPS/ICLR LaTeX draft.

⚠️ The skill instructs the agent to act without asking permission and to never idle. Set compute/API budget limits before enabling it.

· · · Install guide · · ·

Install in the Claude app (no terminal)
  1. Download the ZIP with the button below.
  2. In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
  3. Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
↓ Download ZIP
Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on Claude Skill Mart.
Copy the 0-autoresearch-skill folder from the GitHub repo Orchestra-Research/AI-Research-SKILLs into my ~/.claude/skills/autoresearch/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/Orchestra-Research/AI-Research-SKILLs.git /tmp/ai-research-skills && mkdir -p ~/.claude/skills && cp -r /tmp/ai-research-skills/0-autoresearch-skill ~/.claude/skills/autoresearch

⚠ This is a third-party skill. Check the source repository before installing.

  1. Open a terminal.
  2. Clone the repo: git clone https://github.com/Orchestra-Research/AI-Research-SKILLs.git
  3. Create the skills directory: mkdir -p ~/.claude/skills
  4. Copy the skill: cp -r AI-Research-SKILLs/0-autoresearch-skill ~/.claude/skills/autoresearch
  5. Recommended: also copy companion skills such as 20-ml-paper-writing and 21-research-ideation.
  6. Restart Claude Code and say: "Use the autoresearch skill to start this research question…".
  7. Safety step: the skill sets up a repeating /loop 20m or cron job. If you don't want that, explicitly tell it "run a single cycle, do not create a continuous loop".
  8. Optional search tooling: pip install semanticscholar arxiv (Exa MCP if you have it).
View source on GitHub ↗License: MIT