ai-performance

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

• Third-party content exposure detected (medium risk: 0.65). 该技能在运行时会把用户提供的 materialText/materialUrl/materialFile 作为“contextComposer”输入到 LLM；其中 materialUrl 指向的公开网页内容属于外部来源，且其正文可能被读取并进入 LLM 上下文（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 runner script performs runtime POSTs to the external AI Skills API (https://ai-skills.ai/api/execute, via AISKILLS_BASE_URL), so responses from that endpoint directly control the skill's prompts/outputs and are a required dependency.