network-social-graph

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime workflow ingests conversation body text from the provided conversation dataset (via extract_reply_pairs(..., body_col='body') and later detect_voice_shifts), which is typically authored by non-operating users (outsider participants) and thus can feed free-form outsider text into the LLM context for analysis/reporting.