meta-audience-targeting

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content, establishing a vulnerability surface that is managed through specific defensive instructions and reference protocols.
  • Ingestion points: Fetched website content, Meta documentation, and platform exports read by the agent (SKILL.md).
  • Boundary markers: The skill includes explicit directives to ignore instructions found within data and to treat retrieved content as information to be reported, not followed (SKILL.md, references/agent-security.md).
  • Capability inventory: The skill utilizes network fetching for research and local file system access for persisting product context and decisions (.agents/product-context.md).
  • Sanitization: The referenced security protocol mandates HTML-escaping and URL validation for all third-party content interpolated into deliverables (references/agent-security.md).
  • [EXTERNAL_DOWNLOADS]: The skill retrieves platform-specific data and documentation from recognized services to ensure accuracy.
  • Evidence: The agent fetches real-time targeting documentation and audience size estimates from Meta and Intempt services (SKILL.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 02:23 PM