The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data retrieved from emails and threads, creating a surface for indirect prompt injection.
Ingestion points: The skill ingests untrusted data via spark emails, spark search, and spark thread commands as described in SKILL.md.
Boundary markers: There are no explicit boundary markers or instructions to the agent to ignore or delimit potentially malicious instructions embedded within the email content.
Capability inventory: The skill possesses capabilities to write data through spark draft, and modify state through spark action changeReminder and spark action pin.
Sanitization: The instructions do not specify any sanitization, filtering, or escaping mechanisms for the content retrieved from external email threads.

spark-persona-sales-rep