audience-research

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

• Third-party content exposure detected (high risk: 0.85). The skill’s workflow explicitly includes “Search” and “Extract quotes” from outsider-authored sources (e.g., public web pages, GitHub issues, Reddit/Discord/social posts, review sites like G2/Trustpilot) and then feeds those extracted free-text quotes into the LLM for “Synthesize findings” and the “Research brief” output.