x-mastery-mentor

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

• Third-party content exposure detected (high risk: 0.85). 在场景E“账号诊断与数据采集”中，Skill会通过 computer-use/浏览器工具打开 https://x.com/{username} 并滚动截图、read_page/DOM提取近100条推文的“文本”等可读内容（外部作者的推文正文），随后将其整理成 tweets_{date}.md/json 并用于生成诊断报告，从而把OUTSIDER文本喂入LLM上下文。

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's Scene E explicitly uses the computer-use tool to open and scrape https://x.com/{username} at runtime (screenshot/scroll/DOM extraction) and then injects that fetched tweet data into the agent's analysis/reporting pipeline, so the external URL is fetched during runtime and directly controls the agent's outputs.