churn-prevention

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

• Third-party content exposure detected (high risk: 0.75). 该技能在运行时会把用户提供的 materialText/materialFile/materialUrl（尤其是 materialUrl 指向的公开网页内容）作为“留存与流失内容”进行上下文拼接（contextComposer），从而可能引入第三方作者的自由文本到 LLM 上下文中。

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

• Potentially malicious external URL detected (high risk: 0.90). The included runner contacts the external AI Skills API (default base URL https://ai-skills.ai and POSTs to /api/execute) at runtime to execute the skill and consume responses that directly control the agent's behavior, so this is a required remote execution dependency.