Overview

claude-loop was an early automated Claude Code agent inspired by the “Ralph Wiggum” paper on self-directing AI agents. It explored the idea of a Claude agent that could autonomously loop on a task, making decisions and taking actions without step-by-step human guidance. It served as the direct precursor to little-loops.

Key Features

  • Automated agent loop with configurable task targets
  • Early FSM-style task iteration
  • Claude Code integration for code editing and shell execution

Tech Stack

Python, Claude Code

Background

A research prototype built to test the limits of Claude Code as an autonomous agent. The patterns and lessons from claude-loop were refined and formalized into the little-loops plugin.