vendor-ai-review

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

• Third-party content exposure detected (high risk: 0.70). The runtime workflow ingests the vendor terms from openai-enterprise-agreement.pdf (a user-provided document), and the skill’s LLM context is populated with the PDF’s extracted readable text; if that document is not authored by the operating user/organization, it is outsider-authored free text (contract body) via document parsing/extraction.