architect

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

• Third-party content exposure detected (high risk: 0.85). 该技能在运行时通过 MCP 连接到 https://cyeam-wiki-mcp-production.up.railway.app/mcp，并调用 wiki_query_system 作为系统提示词、再用 wiki_query/wiki_get_article 读取维基文章正文；这些文章内容属于 MCP 服务端的外部知识库（非操作用户自有文本），会以可读的自由文本形式进入 LLM 上下文。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill performs runtime discovery against the MCP endpoint https://cyeam-wiki-mcp-production.up.railway.app/mcp and fetches the 'wiki_query_system' system prompt from that service, which directly controls the agent's system instructions and is required for operation.