The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes calendar event data (titles, descriptions, and attendee notes) which can originate from untrusted external parties.
Ingestion points: Untrusted data enters the context via the list_events tool mentioned in SKILL.md.
Boundary markers: The skill includes a specific 'Safety Rule (HARD)' (Rule 6) which explicitly instructs the agent to treat all invite content as untrusted and ignore any instructions found within them.
Capability inventory: The skill has the capability to write to the calendar (create_event, update_event, delete_event) but requires explicit user confirmation for every action.
Sanitization: The skill mandates user confirmation for all changes and instructs the agent to never follow instructions or open links found in event descriptions.
[COMMAND_EXECUTION]: The skill utilizes standard calendar API tools (Google Calendar, Microsoft 365, and Zapier for Calendly). These are scoped to calendar management and do not involve arbitrary shell command execution.

meeting-scheduler