Fallback Python Execution

Fallback Python Execution Pattern

When to Use

Use this pattern when:

• execute_code_sandbox returns unknown errors or fails repeatedly
• shell_agent cannot successfully execute Python code
• You need to create files (spreadsheets, documents, data files) via Python
• Direct delegated approaches prove unreliable in the current environment

Core Technique

Instead of delegating Python execution to agents, use this two-step inline approach:

1. Write Python code to a .py file using write_file
2. Execute the file using run_shell with python

Step-by-Step Instructions

Step 1: Write Python Code to File

Use write_file to create a Python script with all necessary code inline:

write_file
path: /path/to/script.py
content: |
    import pandas as pd
    # Your complete Python code here
    df = pd.DataFrame({...})
    df.to_excel('output.xlsx', index=False)

Step 2: Execute via run_shell

Run the script directly:

run_shell
command: python /path/to/script.py

Step 3: Verify and Clean Up

• Check the output for success/errors
• Verify the expected files were created
• Optionally remove the temporary script if no longer needed

Why This Works

This approach is more reliable because:

• Avoids agent interpretation layers that can introduce errors
• Provides direct control over execution environment
• Gives clear error output for debugging
• Bypasses sandbox delegation issues

Example: Excel File Creation

# Step 1: Write the script
write_file:
  path: create_report.py
  content: |
    import pandas as pd
    from openpyxl import Workbook
    
    # Create data
    data = {'Column1': [1, 2, 3], 'Column2': ['A', 'B', 'C']}
    df = pd.DataFrame(data)
    
    # Save to Excel
    df.to_excel('report.xlsx', index=False)
    print('Excel file created successfully')

# Step 2: Execute
run_shell:
  command: python create_report.py

Tips

• Include error handling in your Python code for better debugging
• Use absolute paths when possible to avoid working directory issues
• Add print statements to track execution progress
• Keep scripts self-contained with all imports at the top
• For complex tasks, break into multiple scripts if needed

Troubleshooting

| Issue | Solution | |-------|----------| | Module not found | Add pip install commands before python command | | Permission errors | Check file paths are writable | | Script not found | Use absolute path or cd to directory first | | Output not created | Check for Python errors in run_shell output |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: HKUDS
• Source: HKUDS/OpenSpace
• License: MIT

Install and usage instructions live in the source repository linked above.