wechat-cover

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

• Third-party content exposure detected (medium risk: 0.30). 工作流第2步/脚本会把用户提供的“keyword”(外部文本)提交到 doubaoya.com 接口，并在第3步解析其返回的 JSON（含 item.title 等自由文本字段）供主 Agent 总结。

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill makes a runtime POST to https://doubaoya.com/api/apis/gongzhonghao/gongzhonghao-coze-cover/call and the SKILL.md explicitly instructs the agent to relay any top-level "notice" field from that response verbatim, meaning remote content can directly control agent output.