Code Execution

Gemini Code Execution Skill

Overview

This skill outlines how to enable Gemini's native, sandboxed Python Code Execution environment. When activated, Gemini can write Python code, execute it in an isolated sandbox during the generation process, inspect the printed outputs, and use the results to build its final answer.

When to Use This Skill

• Complex mathematical equations or calculations.
• Data analysis on provided charts or tables (e.g. calculating standard deviations, regressions).
• Algorithms, code generation testing, and logic verifications.

Quick Start (with runnable code examples)

from google import genai

# Initialize the Gemini GenAI Client
client = genai.Client()

def run_scientific_computation(prompt: str):
    print(f"Sending prompt to Gemini with Native Python Execution: '{prompt}'...")
    
    # Configure the generate_content call with code_execution tool
    response = client.models.generate_content(
        model='gemini-2.5-flash',
        contents=prompt,
        config=dict(
            # Enable native code execution environment
            tools=[{'code_execution': {}}]
        )
    )
    
    print("\n--- Final Text Output ---")
    print(response.text)
    
    # Inspect the code that Gemini wrote and executed in the background
    candidate = response.candidates[0]
    parts = candidate.content.parts
    
    print("\n--- Background Code Executions ---")
    for part in parts:
        if part.function_call and part.function_call.name == "_exec_code":
            print("\n>>> Code Written by Gemini:")
            print(part.function_call.args.get("code"))
        elif part.function_response and part.function_response.name == "_exec_code":
            print("\n<<< Sandbox Output Result:")
            print(part.function_response.response.get("result"))

if __name__ == "__main__":
    query = (
        "Find the sum of all prime numbers between 1 and 1000. "
        "Write a Python script to calculate this, execute it, and give me the result."
    )
    run_scientific_computation(query)

Advanced Usage

Sandbox Limitations

The code execution environment is a locked-down, transient sandbox:

• No Internet Access: The code inside the sandbox cannot query outside network APIs.
• Limited Libraries: Major preloaded packages include standard library modules, numpy, scipy, pandas, and matplotlib.
• Ephemeral Storage: Files written during the sandbox run are discarded immediately after the request finishes.

Key References

• Google GenAI Code Execution reference
• Gemini Capabilities: Python Sandbox Guide

Dependencies

• google-genai>=0.1.1