saving-codeacts

This skill enables AI agents to transition from ephemeral code execution to building a persistent, modular library of reusable tools. By standardizing the workflow for saving successful code actions into the gentools package, it allows users to capture agent experience as executable knowledge. This process involves establishing a strict directory structure consisting of an empty initialization file, a public api.py defining Pydantic models, and an impl.py module containing the specific implementation details. The approach ensures that agent-generated tools remain maintainable, type-safe, and easily discoverable for future tasks within the Freeact environment.

• Automated tool structure generation using specific directory hierarchies for categorized tool management.

• Enforces the use of Pydantic models for predictable and validated data output, ensuring reliable data integration between tools.

• Implements a clean separation of concerns by utilizing a run interface in api.py that leverages lazy imports from implementation modules.

• Facilitates the composition of existing MCP (Model Context Protocol) tools and Python functions into higher-level, reusable actions.

• Enhances context window efficiency by allowing agents to invoke these saved tools on demand rather than re-implementing complex logic.

• Ensure all tool components define clear, comprehensive docstrings to assist the agent in semantic discovery and future tool selection.

• Maintain strict adherence to the defined package structure to ensure compatibility with Freeact's tool discovery mechanisms.

• Utilize Pydantic Field types to provide detailed metadata and validation constraints for tool parameters and output models.

• Best practice: Always encapsulate complex logic within impl.py, keeping the API layer as a thin, descriptive interface for the agent to interact with.

• The skill is intended for scenarios where repetitive agent tasks, such as data processing, API interactions, or multi-step logic workflows, can be distilled into modularized and versionable code.