The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from meeting transcripts and emails which presents a potential surface for indirect prompt injection attacks. \n- Ingestion points: The agent reads external content via the spark search, spark emails, and spark meeting <id> --transcript commands. \n- Boundary markers: The skill includes a specific safety instruction: "Always confirm the draft with the user before creating it," ensuring human-in-the-loop oversight. \n- Capability inventory: The persona has access to read communications and draft messages through the spark CLI. \n- Sanitization: There is no automated sanitization of ingested data; however, the requirement for human verification of drafts mitigates the risk of the agent performing unintended actions based on injected instructions.

spark-persona-meeting-manager