ai-native-knowledge-rag

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

• Third-party content exposure detected (medium risk: 0.65). 该 Skill 的运行时流程是“资料来源分析/清洗/索引/检索”，其中“资料来源”在实际部署中通常会接入 Confluence/Wiki/PDF/群聊等外部（非操作用户自有）文档内容并将其清洗后转成可读文本喂给 LLM 做生成与评估，因此存在把他人/第三方撰写的自由文本进入 LLM 上下文的间接提示注入风险。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly describes runtime fetching from external services — e.g., Confluence REST API, GitHub Wiki (via git clone), 飞书 API exports and a regulatory-website crawler — and injects those fetched documents into the model context for RAG-driven generation, so these runtime external dependencies can directly control prompts.