excavate

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

• Third-party content exposure detected (high risk: 0.75). 该技能在运行时会读取用户提供的 target（可能是 URL）并通过 mcp__web-reader__webReader 抓取网页正文，把“公共网页/任意 URL 的自由文本”作为可读材料进入 LLM 上下文，从而存在外部作者内容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 accepts a user-supplied target URL (fetched at runtime by mcp__web-reader__webReader and Read) — e.g., any external URL provided as target such as https://example.com/remote-source.md — and injects that fetched content into the agent's analysis context, meaning remote content directly controls prompts at runtime.