The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection.
Ingestion points: Tools such as list_events and get_event retrieve event summaries and descriptions which can be authored by external parties (e.g., meeting organizers).
Boundary markers: There are no delimiters or instructions provided to the agent to treat the ingested calendar data as untrusted or to ignore embedded commands.
Capability inventory: The skill provides write and delete capabilities via create_event, update_event, delete_event, and quick_add_event, which could be triggered by malicious instructions embedded in a calendar invite.
Sanitization: There is no evidence of input sanitization or validation of the retrieved event content before it is processed by the agent.
[DATA_EXFILTRATION]: The skill provides access to potentially sensitive information.
Tools can retrieve attendee email addresses, meeting locations, and private notes, which could be exposed if the agent's context is manipulated by an attacker.
[NO_CODE]: The skill consists solely of tool definitions in markdown.
It lacks the underlying implementation code (e.g., Python or JavaScript scripts) and dependency manifests required to actually perform the described operations.

google-calendar