The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The code example for the verify_api_key dependency uses a hardcoded string ("secret-internal-key") for authentication comparison logic, which is an insecure coding practice even in documentation.
[INDIRECT_PROMPT_INJECTION]: The skill describes building web services that ingest untrusted data from external users, creating an attack surface for indirect prompt injection if the processed data is later interpreted by an AI agent.
Ingestion points: API endpoints (e.g., @app.post("/items/"), @app.get("/items/{item_id}")) that ingest data from request bodies, headers, and path parameters.
Boundary markers: The skill uses Pydantic models (e.g., ProductCreate, Item) which provide structural boundaries and type enforcement.
Capability inventory: The skill demonstrates database access capabilities using sqlalchemy and Depends for session management.
Sanitization: Relies on Pydantic v2 for schema validation and payload parsing to filter malformed inputs.

fastapi