Leaderboard/Research/autoresearch-claude-code
Last commit on September 7, 2026·Created on March 12, 2026

drivelineresearch/autoresearch-claude-code

An autonomous loop for testing scoped changes against fixed benchmarks
Combined rank
#595
across all skills
In Research
#30
category rank
Stars
344
+0.3% in last 7d
Forks
34
0.0% in last 7d
Watchers
2
0.0% in last 7d
Traction scoreGitHub stars can be faked, so popularity alone can be misleading. Traction Score looks for broader signs of recent attention, adoption, and active maintenance.
TL;DR

Autoresearch enables AI research engineers to run measured experiment loops within Claude Code or Codex. It automates the cycle of applying a scoped change, measuring it against a benchmark, and recording the outcome to iteratively optimize performance.

The system operates within defined budgets and user authorizations, allowing researchers to pause, resume, or inspect the progress of an optimization loop while maintaining manual control over scope changes.

WHO IT'S FOR
AI research engineers
running autonomous experiment loops for optimization
ML performance engineers
measuring scoped changes against fixed benchmarks
Claude Code power users
automating iterative code optimization tasks
Automation engineers
configuring unattended optimization loops via Python
Repository contents

1 skill file

Compatible AgentsThe repository documents support for these agents. The skills may also work with other agents that can load SKILL.md files, but they may need some setup or small changes.

Use this skill

Follow the documented setup, then try a first task.

I’m using

Install the Codex skill

Installs: Installs the autoresearch skill into ~/.agents/skills/autoresearch using a symlink; keep the repository checkout in place. · Codex

Before you start
  • Python 3.10+
  • Bash
  • Git
  • GitHub CLI (`gh`)
  • Codex client
  • Linux or macOS; use WSL on Windows

Platform: Linux and macOS; WSL on Windows

Run from: The parent directory for the checkout initially, then the autoresearch-claude-code repository root.

  1. Clone the repository from the directory where its checkout should be created.

    gh repo clone drivelineresearch/autoresearch-claude-code
  2. Enter the repository checkout.

    cd autoresearch-claude-code
  3. From the repository root, run the Codex installer. The trailing text is the author's shell comment describing the destination.

    ./install.sh --codex      # Codex skill in ~/.agents/skills/autoresearch
README.md · Checked Sep 18, 2026skills/autoresearch/SKILL.md · Checked Sep 18, 2026

Give it something to do.

Author’s example

Optimize test-suite runtime with a bounded experiment loop

Uses autoresearch

  • A target project with a committed starting point
  • The test or benchmark command whose runtime should be optimized
  • The primary runtime metric and its direction
  • The exact files the skill may modify
  • Any scorer, data, splits, or correctness checks that must remain fixed
$autoresearch optimize test suite runtime with at most 20 runs
README.md · Checked Sep 18, 2026skills/autoresearch/SKILL.md · Checked Sep 18, 2026