The Agent Skills Directory

[DATA_EXFILTRATION]: OAuth authentication tokens are routed through an external third-party service (google-workspace-extension.geminicli.com).
The perform_oauth_flow function in scripts/auth.py configures the Google OAuth redirect_uri to point to this external domain rather than a local loopback address during the initial code exchange.
Authentication credentials, including access and refresh tokens, are processed by this third-party infrastructure during the login and refresh (/refreshToken) flows.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data retrieved from the Google Calendar API, which could be manipulated by an external party to influence agent behavior.
Ingestion points: Calendar event summaries, descriptions, locations, and attendee data are fetched via the list_events and get_event functions in scripts/gcal.py and returned to the agent context.
Boundary markers: The instructions in SKILL.md do not implement delimiters or explicit instructions for the agent to ignore potential commands or formatting embedded within the retrieved calendar data.
Capability inventory: The skill possesses the capability to make authenticated network requests to the Google API and perform state-changing operations including creating, updating, and deleting calendar events across all calendars the user can access.
Sanitization: There is no evidence of validation, escaping, or content filtering performed on the strings retrieved from the Google Calendar API before they are processed by the agent.

google-calendar