The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Google Calendar without proper isolation or sanitization.
Ingestion points: The getCalendarEvents method in handler.ts fetches event summaries and locations from the Google Calendar API.
Boundary markers: The skill returns a plain-text list of events to the agent without utilizing delimiters (like XML tags) or specific instructions to the model to ignore instructions embedded in the event data.
Capability inventory: The agent has access to create_calendar_event, update_calendar_event, and delete_calendar_event tools, which could be abused if the agent follows instructions found within a malicious calendar entry.
Sanitization: No validation, filtering, or escaping is performed on the event summary or location fields before they are formatted and returned to the agent's context.

google-calendar