ip-profile

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

• Third-party content exposure detected (medium risk: 0.30). 在“第一次建档/重新蒸馏文风 DNA”的运行路径中，agent 会把用户直接提交的范文 content（通过 POST /api/ip-profile/<id>/samples 写入后再用于蒸馏）作为输入喂给 LLM 进行分析并生成 writingDnaJson，因此存在可由外部作者控制的自由文本间接注入到模型的运行时读取。

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

• Potentially malicious external URL detected (high risk: 1.00). The skill issues runtime requests to doubaoya endpoints (e.g., https://doubaoya.com/api/ip-profile) and explicitly uses the fetched profile.writingDnaJson.voiceSystemPrompt to preface the agent's system prompt, so remote content from doubaoya.com can directly control the agent's prompts.