The Agent Skills Directory

[SAFE]: The skill provides documentation and examples for PydanticAI, a reputable Python library for building AI agents. It follows secure coding practices by advising against hardcoding credentials.
[DATA_EXFILTRATION]: No data exfiltration risks were identified. Examples demonstrate fetching data from the well-known weather service 'wttr.in' without exposing sensitive information.
[PROMPT_INJECTION]: The skill facilitates the development of agents that handle external inputs, which inherently presents a surface for indirect prompt injection.
Ingestion points: External data is processed via agent.run(), agent.run_sync(), and tool-specific logic in SKILL.md.
Boundary markers: The provided examples use standard string interpolation; however, the framework's use of Pydantic models for structured output helps enforce data schemas.
Capability inventory: Across SKILL.md, the agents demonstrate capabilities such as making network requests (httpx) and interacting with a database through dependency injection.
Sanitization: The skill includes a dedicated security section recommending input validation and manual confirmation for high-risk actions, which mitigates injection risks.

pydantic-ai