audience-belief-mapper

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

• Third-party content exposure detected (high risk: 0.75). The skill ingests “interview transcripts, win-loss notes, sales-call summaries, and support tickets (all User-provided)” and also may ingest “review-site voice … as User-provided pasted excerpts,” so at runtime any pasted outsider-authored free text (e.g., customer/support ticket text or pasted review excerpts) can be read and then fed into the agent’s LLM context for extraction and verbatim preservation.