For future enhancements, the Agent should be able to run shell. Example use case: 1. Write `main.py` 2. **Run** `python main.py` to see if it crashes 3. If it crashes, read the error message and fix the code automatically
