foundation-build-risk-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 input in the eval scenario is a free-form “Situation brief” authored by the eval harness (outsider), which the agent reads directly as prompt text to generate the Build Risk Review—i.e., outsider-authored prose enters the LLM context via the user-provided scenario payload.