The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing untrusted external data from inbound Gmail replies and shared Google Sheets.
Ingestion points: In SKILL.md, the 'Inbound reply operating procedure' uses gog gmail messages search to ingest thread content. The 'Tracker update guidance' uses gog sheets get to read data from {{GOOGLE_SHEET_ID}}.
Boundary markers: There are no defined boundary markers or instructions to the agent to disregard instructions potentially embedded within the inbound email bodies or sheet cell values.
Capability inventory: The skill has permissions to send emails (outreach workflow) and write to Google Sheets (tracker updates) using the gog toolset.
Sanitization: The instructions lack requirements for sanitizing, escaping, or validating the content of inbound messages before the agent acts upon them or logs them into the tracker.

business-development