relevance-coarse-filter

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

• Third-party content exposure detected (high risk: 0.85). The skill ingests per-signal fields like title/excerpt/evidence and profile_matches that can originate from outsider-authored sources (e.g., public web pages or third-party messages) and are fed as free-text into the LLM context at runtime for classification.