audience-builder

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The skill’s runtime flow takes an outsider-provided prompt (“Audience research for my offer: …”) and then uses the Meta Ads MCP in read-only mode to research interests/behaviors and estimate audience sizes, meaning the LLM ingests arbitrary user/free text at least in the initial instruction and likely uses it to drive Meta queries.