by uditgoenka · Claude Skill · ★ 2.2k
Claude Autoresearch Skill Turn Claude Code into a relentless improvement engine. Based on Karpathy's autoresearch — the principle that constraint + mechanical metric + autonomous iteration = compounding gains. What Is This? A Claude Code skill that makes Claude iterate autonomously on ANY task with a measurable outcome — like Karpathy's autoresearch, but generalized beyond ML. The loop: LOOP (FOREVER or N times): Review current state + git history + results log Pick the next change (based on what worked, what failed, what's untried) Make ONE focused change Git commit (before verification) Run mechanical verification (tests, benchmarks, scores) If improved → keep. If worse → git revert. If crashed → fix or skip. Log the r
| Stars | 2,156 |
| Forks | 159 |
| Language | Shell |
| Category | Claude Skill |
| License | MIT |
| Quality Score | 58.44/100 |
| Open Issues | 1 |
| Last Updated | 2026-03-24 |
| Created | 2026-03-13 |
| Platforms | claude-code, cli |
| Est. Tokens | ~38k |
Explore other popular claude skill tools:
autoresearch is Claude Autoresearch Skill — Autonomous goal-directed iteration for Claude Code. Inspired by Karpathy's autoresearch. Modify → Verify → Keep/Discard → Repeat forever.. It is categorized as a Claude Skill with 2.2k GitHub stars.
autoresearch is primarily written in Shell. It covers topics such as ai, autonomous-agent, autoresearch.
You can find installation instructions and usage details in the autoresearch GitHub repository at github.com/uditgoenka/autoresearch. The project has 2.2k stars and 159 forks, indicating an active community.
autoresearch is released under the MIT license, making it free to use and modify according to the license terms.