agent-browser

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

• Third-party content exposure detected (high risk: 0.75). 该技能的运行时会通过 materialUrl/materialText 触发 agent-browser 的“网页研究/浏览”并把抓取到的公开网页正文（外部作者的自由文本）拼入 LLM 上下文；因此存在来自 OUTSIDER 的免费文本注入路径。

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

• Potentially malicious external URL detected (high risk: 0.80). The runner (scripts/run.py) makes runtime requests to the external API at https://ai-skills.ai (POST to /api/execute and polling endpoints), meaning this remote service is required for execution and its responses directly control the skill's output/instructions.