Agent Skills: Saving Code Actions as Reusable Tools

Saving Code Actions as Reusable Tools

Save executed Python code as a tool for later reuse.

Package Structure

{generated_rel_dir}/gentools/<category>/<tool>/
├── __init__.py          # Empty file
├── api.py              # Public interface with structured models
└── impl.py             # Implementation details

Procedure

1. Create Package Directory

mkdir -p {generated_rel_dir}/gentools/<category>/<tool>

Create empty __init__.py files in both <category> and <tool> directories.

2. Define Tool API (api.py)

from __future__ import annotations

from pydantic import BaseModel, Field


class OutputModel(BaseModel):
    """Description of output."""
    field: type = Field(..., title="Description")


def run(param1: type, param2: type = default) -> OutputModel:
    """Tool description.

    Args:
        param1: Description
        param2: Description (default: value)

    Returns:
        OutputModel with structured data
    """
    from .impl import implementation_function
    return implementation_function(param1, param2)

Requirements:

• Define Pydantic models for structured output
• Create run() function with typed parameters
• Use lazy import from impl.py inside run()
• Include comprehensive docstring
• Export OutputModel and run in {generated_rel_dir}/gentools/<category>/<tool>/__init__.py:

from .api import OutputModel, run

__all__ = ["OutputModel", "run"]

3. Implement Details (impl.py)

from __future__ import annotations

from mcptools.<category>.<tool> import Params, run_parsed
from .api import OutputModel


def implementation_function(param1: type, param2: type) -> OutputModel:
    """Implementation description."""
    # Use tools from mcptools or gentools packages
    result = run_parsed(Params(...))

    # Transform and return structured output
    return OutputModel(field=result.data)

Requirements:

• Import tools from mcptools or gentools packages
• Import models from api.py
• Return structured models defined in api.py

4. Test the Tool

from gentools.<category>.<tool>.api import run

result = run(param1=value1, param2=value2)
print(result)

Best Practices

• Separation: Keep API clean; hide complexity in implementation
• Type Safety: Use Pydantic models for all outputs
• Modularity: Break complex logic into smaller functions
• Defaults: Provide sensible defaults for optional parameters