The Agent Skills Directory

[SAFE]: The skill is a documentation resource for the PydanticAI library and does not contain malicious code or instructions.
[EXTERNAL_DOWNLOADS]: The skill includes instructions to install pydantic-ai and its associated provider extras (OpenAI, Anthropic, Gemini, etc.) using pip. These packages are maintained by a well-known organization and are considered safe.
[INDIRECT_PROMPT_INJECTION]: The skill describes an architecture for agents that ingest and process external data. Ingestion points: Data enters through agent.run() and agent.run_sync() calls as shown in the examples. Boundary markers: While simple examples do not show explicit delimiters, the framework's use of structured outputs and system prompts provides natural boundaries. Capability inventory: Examples demonstrate network requests via httpx and potential database interactions via dependency injection. Sanitization: The skill provides proactive security guidance in the 'Security & Safety Notes' section, recommending input validation and human-in-the-loop confirmation for sensitive operations.

pydantic-ai