pm-need

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

• Third-party content exposure detected (high risk: 0.85). 该 skill 在运行时会在 /pm-collect 阶段对 $ARGUMENTS 中的 URL 进行“URL 抓取”并把抓取到的网页/文档正文作为材料进入后续 /pm-refine 的 LLM 上下文；这属于“公共 web 内容在运行时被抓取并喂给模型”的 outsider 来源。

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly performs runtime URL fetching of any URLs provided in $ARGUMENTS (e.g., https://docs.example.com/dashboard-spec or https://example.com/spec) and injects the fetched page/document content into the agent's collect/refine context, so remote content can directly influence and effectively control the model's prompts/outputs.