meta-ad-library
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted competitor ad content and website data, which presents a surface for indirect prompt injection attacks where malicious instructions could be embedded in ads or landing pages.
- Ingestion points: Content is retrieved from the Meta Ad Library (facebook.com/ads/library) and external competitor URLs provided by the user or researched by the agent.
- Boundary markers: The skill implements explicit constraints in SKILL.md and references/agent-security.md instructing the agent to treat all retrieved content strictly as data and never as instructions.
- Capability inventory: The agent can fetch external web content and write to persistent local storage files in the .agents/ directory.
- Sanitization: The skill mandates that any instructional text found within untrusted data must be reported as a finding and quoted verbatim rather than followed, mitigating the risk of command execution from external sources.
Audit Metadata