The Complete Production SOP: How to Deploy and Run HuggingFace Smolagents Minimalist Code Agents in Autonomous AI Agents

---
name: "smolagents-lightweight-python-agents"
description: "Run code-writing agents that execute Python logic directly instead of serial JSON actions."
metadata:
  author: "HuggingFace"
  version: "1.0.0"
  runtime_compatibility:
    - "codex"
    - "claude-code"
    - "cursor"
    - "windsurf"
triggers:
  - "create smolagents codeagent"
  - "run python action agent"
  - "use duckduckgo tool smolagents"
permissions:
  fs_read: true
  fs_write: "workspace"
  shell_exec: false
  net_access: false
---

# HuggingFace Smolagents Minimalist Code Agents

Implements HuggingFace smolagents (CodeAgent, ToolCallingAgent): expressing agent actions directly as executable Python code snippets, simplifying complex loops and data transformations.

## When to Invoke
Invoke this skill whenever you need to:
- create smolagents codeagent
- run python action agent
- use duckduckgo tool smolagents

## Execution SOP & Rules
1. Initialize agents using agent = CodeAgent(tools=[...], model=HfApiModel()).
2. Write custom tools as standard Python functions decorated with @tool.
3. Review generated Python code in sandboxed execution environments to maintain security.
4. Inspect execution logs and tool call outputs using agent.run(prompt).

## Verification Checklist
- Run local linting, syntax validation, and type-checks before reporting completion.
- Ensure all created or modified files adhere to workspace boundaries and avoid uncommitted secrets.
- Verify tests or build commands exit cleanly with return code 0.

import os
from dotenv import load_dotenv
from smolagents import CodeAgent, HfApiModel, tool

load_dotenv('.env.local')

@tool
def calculate_churn_rate(total_lost: int, total_start: int) -> float:
    """Calculates customer churn percentage based on cohort numbers.

    Args:
        total_lost: Number of customers lost in interval.
        total_start: Number of customers at start of interval.
    """
    if total_start <= 0:
        return 0.0
    return round((total_lost / total_start) * 100, 2)

model = HfApiModel(model_id="Qwen/Qwen2.5-Coder-32B-Instruct", token=os.getenv("HF_TOKEN"))
agent = CodeAgent(tools=[calculate_churn_rate], model=model, add_base_tools=True)

if __name__ == "__main__":
    result = agent.run("Given 45 lost clients out of 1200 initial users, calculate churn.")
    print("Execution Result:", result)

import os
from smolagents import CodeAgent, DuckDuckGoSearchTool, HfApiModel
from dotenv import load_dotenv

load_dotenv('.env.local')

search_tool = DuckDuckGoSearchTool()
model = HfApiModel(model_id="Qwen/Qwen2.5-Coder-32B-Instruct", token=os.getenv("HF_TOKEN"))

# Configure agent with explicit search tool and standard library execution capability
agent = CodeAgent(
    tools=[search_tool],
    model=model,
    additional_authorized_imports=["math", "json", "datetime"]
)

response = agent.run("Search for top 3 open-source vector databases in 2026 and format as markdown.")
with open("output.md", "w", encoding="utf-8") as f:
    f.write(str(response))

import unittest
from agent_runner import calculate_churn_rate

class TestAgentTools(unittest.TestCase):
    def test_calculate_churn_standard(self):
        self.assertEqual(calculate_churn_rate(45, 1200), 3.75)

    def test_calculate_churn_zero_division(self):
        self.assertEqual(calculate_churn_rate(10, 0), 0.0)

if __name__ == '__main__':
    unittest.main()