15-social-listening

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

• Third-party content exposure detected (high risk: 1.00). The SKILL.md explicitly requires collecting and reading untrusted public content from third‑party sources (see "Kenh giam sat" and "Quy trinh research" which list Facebook Groups/Pages, TikTok, Reddit VN, Google News, forums like Voz, Google Reviews, Shopee, etc.), and uses those inputs to drive sentiment analysis, crisis detection and prescribed response actions—so external user-generated content can directly influence the agent's decisions.