wenqu-library

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

• Third-party content exposure detected (medium risk: 0.30). 该技能在“搜索→候选合并→下载→整理”过程中会对外部候选 URL 进行网页抓取与通读摘要（如 references/collection-playbook.md 第2步/第3步/第4步），因此外部/第三方作者文本会在运行时进入 LLM 的整理摘要与索引环节。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly downloads and ingests external web pages at runtime (e.g., crwl will fetch pages such as https://mp.weixin.qq.com/s/xxx and uses https://weixin.sogou.com/ as a Referer) and also contains runtime install commands that fetch/execute remote CLI packages, so external content fetched during skill execution can directly influence the agent's context and behavior.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The skill explicitly instructs the agent to perform global CLI/browser installations and system setup on the host after user authorization, which requires modifying machine state (and likely escalating privileges or changing system files).