eliteforge-competitive-analysis

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

• Third-party content exposure detected (high risk: 0.85). 该 Skill 的运行时会在 Step 2/Phase 1-3 通过 search_engine 抓取并 scrape_as_html/scrape_as_markdown 读取官网与扩展搜索页面、以及下载并转换第三方文档为 Markdown（这些内容通常来自非用户选择的外部作者/网站），从而把“网页/文档正文的自由文本”喂入 LLM 上下文进行摘要与报告撰写。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill's Phase 2 workflow explicitly instructs the agent to download arbitrary external files at runtime using commands like curl -L -o "{分析对象名}-资料/raw/{序号}-{描述}.{ext}" "{文件URL}", which will be fetched, converted and injected into the agent's context as required input—thus external URLs can directly control prompts/content.