skills/pollow/c-suite-skills/cmo/Gen Agent Trust Hub

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 WebSearch and WebFetch to ingest content from external sites like Reddit and competitor pages (referenced in SKILL.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 Write tool for the docs/marketing/ directory and the Edit tool for the HUMAN_AGENDA.md file.
  • 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
Risk Level
SAFE
Analyzed
Mar 7, 2026, 11:50 PM