pydantic-ai-skills

pydantic-ai-skills is a standardized framework for creating and managing Agent Skills within Pydantic AI, following the open Agent Skills specification maintained by Anthropic. It enables AI agents to progressively discover and load modular collections of instructions, scripts, and tools for domain-specific tasks, reducing token usage through lazy loading. The package supports both filesystem-based skills defined as Markdown directories and programmatic skills created with Python decorators or dataclasses. It integrates directly with Pydantic AI agents through the SkillsCapability or SkillsToolset APIs, providing type-safe skill definitions with automatic validation and built-in security features like path traversal prevention.

[{"step":1,"action":"Identify the skill to load","details":"Use the skill name (e.g., 'file_processor', 'data_validator') or specify a filesystem path to a custom skill. For Claude agents, ensure the skill is compatible with the agent's environment."},{"step":2,"action":"Define execution parameters","details":"Provide required parameters such as input files, output formats, or chunk sizes. Use clear and specific values to avoid ambiguity."},{"step":3,"action":"Execute the skill","details":"Run the skill using the Pydantic AI manager. Monitor the execution steps for real-time feedback. For filesystem skills, ensure the path is accessible."},{"step":4,"action":"Validate the output","details":"Review the structured output for success status, execution metrics, and any errors. Use the output for further processing or integration with other tools."},{"step":5,"action":"Integrate with agents (optional)","details":"If using a Claude agent, pass the execution context and results back to the agent for further action or reporting. Ensure the agent has access to the output files."}]