afa-explore

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

• Third-party content exposure detected (high risk: 0.90). The skill's core workflow (模式 A: VOC 挖掘 and 模式 E: 趋势与信号监控) explicitly requires fetching and mining public, user-generated sources such as Amazon reviews, Reddit, social media comments, Google Trends and TikTok (see "从 7 大数据源逐一挖掘（Amazon评论 / Reddit / 社媒评论 ...）" and "扫描 Google Trends、... Reddit/TikTok"), which the agent is expected to read and use to drive angle generation and decisioning—thus exposing it to untrusted third‑party content that could carry indirect prompt-injection.