reference-alignment

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

• Third-party content exposure detected (high risk: 0.75). The skill’s workflow explicitly ingests “references” such as “website”/“doc”/“issue”/“previous implementation” and then “extract[s] requirements” from the “richest available layer,” which at runtime can include outsider-authored free text (e.g., public web pages or issue/discussion bodies) that is read and summarized into the agent’s LLM context.