The Agent Skills Directory

[SAFE]: No malicious patterns or behaviors were detected. The skill's functionality is focused on enhancing agent observability and implementing runtime safety guards.
[INDIRECT_PROMPT_INJECTION]: The skill implements observers that ingest and process model-generated tool arguments and responses, which constitutes an attack surface for indirect prompt injection.
Ingestion points: event.arguments in the PathGuardian observer (assets/safety_guard.py) and event.content in the log_response observer (SKILL.md).
Boundary markers: None identified in the provided code snippets.
Capability inventory: The skill interacts with the agent's tool execution flow, specifically demonstrating a mock write_file tool in the safety example.
Sanitization: The PathGuardian example uses basic string containment checks (e.g., "/etc/" in event.arguments) to identify and block potentially dangerous operations.

ag2-observers-and-alerts