Coding-Agent

A lightweight coding agent that uses function-calling to safely interact with files and run Python scripts inside a workspace.

What this project does

This repository implements a small CLI agent that:

1. Accepts a natural-language prompt.
2. Lets an LLM decide whether to call a predefined function.
3. Executes only supported local functions (for file listing, file reading/writing, and Python execution).
4. Returns the function result back to the model to generate a final answer.

It is designed as a practical learning project for controlled tool use with an AI assistant.

Repository structure

• main.py: CLI entry point and orchestration loop.
• call_function.py: Routes model-requested function calls to Python implementations.
• config.py: Centralized configuration.
• functions/:
  • get_files_info.py: List files and metadata.
  • get_file_content.py: Read file contents.
  • write_file.py: Write/create files.
  • run_python.py: Run Python files safely in workspace context.
• calculator/: Example target project used by the agent.
• test_run_python_file.py: Tests for Python execution behavior.
• README.md: Project documentation (this file).

How it works

1. You run the CLI with a prompt, for example:
   • “run tests.py”
   • “what files are in the root?”
   • “create a new README.md file with the contents '# calculator'”
2. The model may produce a function call.
3. call_function.py validates and executes the call against local functions.
4. The result is fed back into the conversation for a final response.

Prerequisites

• Python 3.11+ (recommended)
• uv installed
• Required environment variables/API access for your selected LLM provider (if applicable in your local setup)

Setup

# from repo root
uv sync