facebook-ads-audit
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Meta ad account exports and external websites, which serves as a potential attack surface.
- Ingestion points: Ad campaign names, CSV/text exports, and content fetched from external URLs (SKILL.md).
- Boundary markers: The skill explicitly directs the agent to treat untrusted content as data rather than instructions and to quote any detected injection attempts as findings (references/agent-security.md).
- Capability inventory: File writes to .agents/product-context.md, network access for research, and visual artifact rendering.
- Sanitization: Enforces HTML-escaping for interpolated values and HTTPS validation for URLs to mitigate XSS risks (references/agent-security.md).
- [PERSISTENCE]: The skill records research and business context to .agents/product-context.md to maintain state across sessions. This behavior is documented for the user and follows a policy of minimal data retention to reduce liability.
- [PROMPT_INJECTION]: Analysis of stylistic constraints (e.g., banning em-dashes) and mandatory output lines (e.g., 'No changes were made') indicates these are intended for readability and audit logging rather than malicious concealment or behavior overriding.
Audit Metadata