smart-search

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

• Third-party content exposure detected (high risk: 1.00). The SKILL.md explicitly instructs the agent to construct and navigate to public social media and forum search URLs (e.g., Reddit, Twitter/X, Zhihu, Weibo, YouTube) and then "extract the relevant information from the visible results," meaning the agent ingests untrusted, user-generated third‑party content that can influence its subsequent actions.