author-modeladvisor-checks

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

• Third-party content exposure detected (medium risk: 0.30). At runtime the agent ingests outsider-authored free text from the user’s natural-language modeling constraint (and any provided IDs/path) via the entry “Detect Intent” and then uses it as input to derive the guideline/check/test specifications (Steps 1–3), rather than restricting to only trusted first-party content.