volcengine-knowledge-search

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

• Third-party content exposure detected (high risk: 0.75). 运行时会把“用户提供的文档链接/URL”作为 fetch 的输入传给上游文档服务并读取其返回的 Result.Content，再经 clean_html() 变成可读 markdown 文本进入脚本输出/LLM上下文；该 URL 属于外部未由操作用户选择的网页内容来源（public web content fetched at runtime）。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill performs runtime POST requests to https://docs-api.cn-beijing.volces.com/api/v1/doc (used for both "search" and "fetch") and directly parses and injects the returned document Content into its output/context, so externally fetched content at runtime controls the agent's responses and is a required dependency.