uditgoenka

uditgoenka / autoresearch

Public

Claude Autoresearch Skill — Autonomous goal-directed iteration for Claude Code. Inspired by Karpathy's autoresearch. Modify → Verify → Keep/Discard → Repeat forever.

ai autonomous-agent autoresearch claude claude-code
120
9
100% credibility
Found Mar 14, 2026 at 117 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
AI Summary

A skill for an AI coding tool that enables autonomous iteration to improve any measurable task, such as code quality, content, or processes, by making targeted changes, verifying outcomes, and retaining enhancements.

How It Works

1
🔍 Discover the Magic

You learn about a clever skill that lets your AI coding helper improve projects automatically, like boosting test coverage or speeding up websites.

2
📥 Add It Easily

You simply place the skill into your AI coding workspace, and it's ready to go.

3
🎯 Pick Your Goal

You describe what you want better, such as higher test scores or smaller file sizes, and how to check if it's improving.

4
🚀 AI Takes Over

Your AI starts smartly tweaking things one small change at a time, testing each to keep only the winners and undo the losers – all by itself!

5
⏰ Step Away

You relax, sleep, or focus elsewhere while the AI keeps working non-stop, stacking up improvements.

🎉 See the Results

You return to find your project much better, with a clear log of every successful step and overall progress.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 117 to 120 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is autoresearch?

Autoresearch is a Claude Code skill that turns Anthropic's Claude into an autonomous agent for goal-directed iteration on code, content, or docs—modify files, verify with a metric via shell command, commit improvements, revert failures, and repeat forever. Inspired by Karpathy's autoresearch on GitHub, it generalizes the loop to any task with mechanical verification, like boosting test coverage or shrinking bundle sizes, using git for versioning and TSV logs for tracking. You invoke it via slash command with a goal, file scope, metric direction, and verify script, then step away while it runs.

Why is it gaining traction?

It stands out by constraining Claude to atomic changes with fast, deterministic metrics, enabling hundreds of experiments overnight without babysitting—unlike manual tweaks or vague prompts. Developers dig the Claude GitHub integration via skills and MCP servers for real tools like databases or APIs in verification steps, plus seamless git commits for reviewable progress. The hook: set a quantifiable goal like "cut API latency under 100ms" and get compounding gains autonomously.

Who should use this?

Software engineers chasing test coverage, perf opts, or refactors in TypeScript/Node projects; marketers iterating SEO copy, ad variants, or email templates with readability scores; ops teams hardening runbooks or IaC via compliance checks. Pairs best with Claude Code users already in git repos needing hands-off iteration on metric-driven tasks.

Verdict

Worth forking for Claude Code setups with solid verify scripts—72 stars and detailed docs show promise, but 1.0% credibility flags early maturity and unproven scale. Test on toy goals first; scales if your metrics stay under 10s.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.