weekly-review
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from emails and calendar invites, which serves as a surface for malicious instructions.\n
- Ingestion points: Accesses data from Gmail and Google/Microsoft 365 calendars (SKILL.md).\n
- Boundary markers: The skill contains explicit safety rules (Rule 2) instructing the agent to treat external content as hostile and ignore any instructions found within (SKILL.md).\n
- Capability inventory: Reading private emails and calendar events, with the ability to write to Notion after user confirmation (SKILL.md).\n
- Sanitization: Explicitly forbids fetching external links or images to prevent data exfiltration and requires summarization rather than direct interaction with embedded content (SKILL.md).
Audit Metadata