vendor-agreement-review

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow ingests the full vendor agreement text provided by the operating user/counterparty (e.g., “Read the whole agreement once…”, then “exact quote from the contract”), which is outsider-authored free text from the vendor’s draft and is fed into the agent’s LLM context for term-by-term comparison.