coding agent

Single Prompt - AI Coding Agent

This project implements an AI coding agent capable of autonomously building software projects based on a single user prompt. It leverages the Anthropic API to execute the entire development process, creating a fully functional project with minimal human intervention.

Features ?

• Development plan creation
• Step-by-step project building using AI
• File reading, writing, and modification capabilities
• Terminal command execution for package installation and other system-level operations
• Logging of LLM history

Project Structure

• main.py: The entry point of the application
• idea_gathering.py: Handles the initial project idea interpretation phase
• build.py: Executes the development plan
• llm_thread.py: Manages conversations with the Anthropic Claude model
• llm_utils.py: Utility functions for file operations and terminal commands

Usage

1. Install Python
2. Create a secret_key.txt file in the project root and add your Anthropic API key
3. Run the main script: main.py
4. Provide a project idea when prompted

The script will then:

1. Set up the project directory and logging
2. Connect to the Anthropic API
3. Interpret your project idea and create a development plan
4. Build the project step by step based on the AI-generated plan

Important Notes ?

• Ensure you have the necessary permissions to create directories and files in the specified path
• User confirmation is required before executing terminal commands
• Only files inside the project directory can be accessed by the agent
• This tool is designed to work autonomously, with minimal user intervention after the initial prompt

Logging

All activities are logged in agent_logs/agent_log.txt. Check this file for detailed information about the build process.

Caution ?

This script executes terminal commands and modifies files on your system. Always review the actions it's about to take and use it in a controlled environment. It's recommended to use this tool in a sandboxed or isolated development environment.

Limitations ?

• The AI Agent's capabilities are limited by the underlying language model
• Limited to locally installable tools and packages; no web browsing capability
• Complex projects may require human intervention
• May not always produce optimal or bug-free code
• Optimized for macOS

Support ?

For any questions or issues, please open an issue in the GitHub repository. For personal inquiries, feel free to reach out to me on LinkedIn.

Disclaimer

This tool is for experimental and educational purposes. Always review and test the generated code before using it in production environments.