AI Skill MartBrowse skillsPromptsProgramsLearnWhat is a Skill?
 
← Back to list

Autoresearch Agent (Autonomous Experiment Loop)

An autonomous edit-measure-keep/revert loop that optimizes any file against a measurable metric using git as the safety net.

AutomationAdvanced★ 24,151⑂ 3,405AI score 8/10Last updated: Aug 9, 2026

What it does

An autonomous experiment loop inspired by Karpathy's autoresearch. The agent edits one target file, runs a fixed evaluation command, keeps improvements via git commit, discards regressions via git reset --hard, and repeats indefinitely.

  • Slash commands: /ar:setup, /ar:run, /ar:loop, /ar:status, /ar:resume
  • Logs every run to results.tsv (commit, metric, keep/discard/crash, description) with TSV/CSV/Markdown dashboards
  • Built-in evaluators: speed (p50_ms), file/bundle size, test pass rate, build time, peak memory, plus LLM judges for headlines, copy, and prompts
  • Guardrails: one change per experiment, never modify the evaluator, timeout at 2.5× budget, pause after 5 consecutive crashes, escalate strategy over time

Who it's for

  • Engineers optimizing latency, bundle size, build time, or test pass rate
  • Marketers and prompt engineers who can score copy or prompts numerically
  • Anyone wanting an agent to run measured experiments overnight

Prerequisites: a git repo, one target file, an eval command that prints a metric to stdout, and Python.

Examples

  1. Cut p50 latency in src/api/search.py using a pytest benchmark as the evaluator
  2. Improve headlines in content/titles.md scored by an LLM judge (ctr_score 0-10)
  3. Run a nightly loop shrinking a frontend bundle measured by size_bytes

· · · Install guide · · ·

Try it now, no install

Paste this into Claude to use the skill without installing anything.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/alirezarezvani/claude-skills/HEAD/.gemini/skills/autoresearch-agent/SKILL.md

What I want: (describe your task here)

If Claude can't open the link, open it yourself and paste the contents instead.

↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.

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 AI Skill Mart.
Copy the .gemini/skills/autoresearch-agent folder from the GitHub repo alirezarezvani/claude-skills into my ~/.claude/skills/autoresearch-agent/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/alirezarezvani/claude-skills.git && mkdir -p ~/.claude/skills && cp -r claude-skills/.gemini/skills/autoresearch-agent ~/.claude/skills/

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

  1. Open a terminal and clone the repo: git clone https://github.com/alirezarezvani/claude-skills.git
  2. Locate the skill folder: find claude-skills -name autoresearch-agent -type d
  3. Create your skills directory: mkdir -p ~/.claude/skills
  4. Copy the whole folder: cp -r claude-skills/.gemini/skills/autoresearch-agent ~/.claude/skills/
  5. Confirm the scripts/ Python helpers came along and that Python 3 is installed.
  6. Make sure the project you want to optimize is a git repo (git init and commit if not).
  7. Restart Claude Code, then run /ar:setup or simply ask "optimize this file for speed".
  8. Before starting a long loop, run your evaluation command once manually and verify it prints metric_name: value.
View source on GitHub ↗License: MIT