The Agent Skills Directory

[SAFE]: The skill is entirely educational and documentary in nature, providing users with code samples to integrate the 'Instructor' library. It contains no executable scripts or commands that pose a risk to the agent or host environment.
[INDIRECT_PROMPT_INJECTION]: The skill proactively addresses the risk of untrusted data from LLM outputs by teaching the use of Pydantic schemas. This approach serves as a robust validation layer that prevents malformed or malicious data from being processed further without strict type and constraint checking.
Ingestion points: Untrusted content returned from LLM completions (OpenAI, Anthropic, or local providers) through the response_model interface.
Boundary markers: Pydantic BaseModel classes with explicit field types and descriptions serve as strict delimiters between the LLM and the application logic.
Capability inventory: The skill demonstrates data extraction, classification, and structured analysis; no dangerous operations like shell execution or unauthorized network exfiltration are suggested.
Sanitization: The core of the skill is demonstrating automated sanitization via Pydantic validators (field_validator, model_validator) and built-in constraints (gt, pattern, EmailStr).

instructor