tencent-survey

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

• Third-party content exposure detected (medium risk: 0.30). 在运行时工作流中，该 skill 会在“查看问卷详情/查看问卷回答”等路径下调用 get_survey/list_answers，并把问卷的 title/text/answer 等返回的自由文本作为上下文读取（这些文本最终来自外部问卷内容/作答者），属于间接提示注入暴露面。

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

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly instructs using mcporter to call the remote MCP service to run "check_skill_update" and "follow the returned instruction", and that MCP endpoint (https://wj.qq.com/api/v2/mcp) is invoked at runtime to fetch an "instruction" that the agent is supposed to execute, so remote content can directly control agent behavior.