This project provides a minimal, state-machine-driven autonomous agent for software development tasks, powered by OpenAI's GPT-4. It's designed for researchers and developers experimenting with AI agent capabilities, offering a foundational framework for LLM-driven code generation and modification.

How It Works

The agent operates as a finite state machine, where each state executes a specific action and determines the next state. The core MAIN state leverages an LLM to select the subsequent action, such as modifying files, adding files, or running tests. This approach allows the agent to iteratively build and test software based on a defined "purpose."

Quick Start & Requirements

• Install via python run.py.
• Requires an OPENAI_API_KEY environment variable.
• Python 3.x.

Highlighted Details

• Demonstrates autonomous code generation and modification for a Python terminal-based guessing game.
• Includes an example of the agent creating a file, modifying it, and then writing and running tests.
• Agent can update its task but not its core purpose.
• Agent struggles with writing accurate tests, particularly those involving numeric comparisons.

Maintenance & Community

No specific community channels, contributors, or roadmap are detailed in the README.

Licensing & Compatibility

The README does not specify a license.

Limitations & Caveats

The agent has significant limitations, including difficulty in writing comprehensive tests for complex applications (web, terminal, REST) and a tendency to err on numeric comparisons in tests. Future work includes adding file-peeking, self-reflection, and documentation loading states.