cmo
Pass
Audited by Gen Agent Trust Hub on Mar 7, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Risk of indirect prompt injection from external web content.
- Ingestion points: The skill uses
WebSearchandWebFetchto ingest content from external sites like Reddit and competitor pages (referenced inSKILL.md). - Boundary markers: Absent; there are no instructions to the agent to treat external data as untrusted or to isolate it from the main instruction flow.
- Capability inventory: The agent has permission to use the
Writetool for thedocs/marketing/directory and theEdittool for theHUMAN_AGENDA.mdfile. - Sanitization: Absent; no explicit filtering or validation of the fetched data is required before it is used to generate or modify local files.
Audit Metadata