The Agent Skills Directory

[Data Exposure & Exfiltration] (SAFE): The code templates demonstrate best practices for configuration management using a Settings class to handle sensitive data like anthropic_api_key. No hardcoded credentials or unauthorized exfiltration patterns were found.
[Indirect Prompt Injection] (LOW): The skill defines patterns for ingesting external data through API endpoints and WebSockets which could serve as injection surfaces in a live application.
Ingestion points: routers/agents.py (POST/GET requests) and the WebSocket receive_text method.
Boundary markers: None present in the provided templates.
Capability inventory: The templates include service abstractions for LLM interaction (LLMService) and background task execution.
Sanitization: No explicit sanitization or validation logic beyond standard Pydantic schema validation is included in the templates.
[Command Execution] (SAFE): Included bash commands are standard instructions for starting Uvicorn and Gunicorn servers. They do not include privilege escalation or suspicious flags.

fastapi