AG2 Bug Fixer Agent

Overview

This example demonstrates how to build an AI-powered bug fixer using two AG2 agents and Daytona sandboxes. A bug_fixer LLM agent analyzes broken code and proposes fixes, while a code_executor agent runs each fix attempt inside an isolated Daytona sandbox. If execution fails, the bug fixer sees the full error output and retries — looping until the code passes or the turn limit is reached.

The executor supports Python, JavaScript, TypeScript, and Bash — inferring the language automatically from the code block returned by the LLM.

Features

• Execution-verified fixes: Every proposed fix is actually run in a sandbox — the agent only terminates when the code passes, not just when it looks correct
• Secure execution: Fix attempts run in isolated Daytona sandboxes, not on your machine
• Multi-language support: Python, JavaScript, TypeScript, and Bash — language is inferred automatically from the LLM's fenced code block
• Iterative refinement: If a fix fails, the agent sees the full error output and retries automatically
• Automatic cleanup: The sandbox is deleted as soon as fix_bug returns, regardless of outcome

Requirements

• Python: Version 3.10 or higher

Environment Variables

• DAYTONA_API_KEY: Required for access to Daytona sandboxes. Get it from Daytona Dashboard
• OPENAI_API_KEY: Required for GPT-4o-mini access. Get it from OpenAI Platform

Getting Started

1. Create and activate a virtual environment:

python3.10 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

2. Install dependencies:

pip install "ag2[daytona,openai]" python-dotenv

3. Set your API keys in .env:

DAYTONA_API_KEY=your_daytona_api_key
OPENAI_API_KEY=your_openai_api_key

4. Run the example:

python main.py

How It Works

1. code_executor sends the broken code to bug_fixer as the opening message
2. bug_fixer (GPT-4o-mini) analyzes it and replies with a fix wrapped in a fenced code block
3. AG2 detects the code block, runs it in a Daytona sandbox via DaytonaCodeExecutor, and sends the result back
4. If execution fails, bug_fixer sees the error output and proposes a new fix
5. The loop continues until the code passes or max_turns is reached
6. The sandbox is automatically deleted when fix_bug returns

Examples

The script runs three examples — one per supported language:

• Python — postfix expression evaluator with swapped operands for - and /
• JavaScript — run-length encoder with wrong concatenation order in two places
• TypeScript — Kadane's max subarray algorithm using Math.min instead of Math.max

Example Output

============================================================
Example 1: Python — Postfix Expression Evaluator Bug
============================================================
code_executor (to bug_fixer):

Fix this broken code:
...

>>>>>>>> USING AUTO REPLY...
bug_fixer (to code_executor):

```python
def eval_postfix(expression):
    ...
    elif token == '-':
        stack.append(a - b)  # Fixed order of operands for subtraction
    elif token == '/':
        stack.append(a // b)  # Fixed order of operands for division
    ...
```

>>>>>>>> USING AUTO REPLY...
>>>>>>>> EXECUTING CODE BLOCK (inferred language is python)...
code_executor (to bug_fixer):

exitcode: 0 (execution succeeded)
Code output: All postfix tests passed!

>>>>>>>> USING AUTO REPLY...
bug_fixer (to code_executor):

TERMINATE

API Reference

For the complete API reference, see the DaytonaCodeExecutor documentation.

License

See the main project LICENSE file for details.

References

• AG2
• AG2 DaytonaCodeExecutor
• Daytona