deep-research

该技能的研究编排、文件写入和本地脚本使用与其“深度研究框架”目的大体一致，未见明确凭据窃取或恶意外传。但它依赖另一项来自个人 GitHub 仓库的技能，形成可变的 transitive trust 链；同时其核心流程会大规模摄取不受信任网页内容，并结合文件写入、脚本执行和高自主子代理调度。综合看更适合判定为 SUSPICIOUS 而非恶意：主要风险来自依赖供应链与间接提示注入面，而非明显恶意行为。

The code appears to be a research-dashboard HTTP server rather than malware. It performs only local file reads and serves the resulting data. However, it is unsafe to expose directly to an untrusted network because it binds to all interfaces without authentication, exposes reports and potentially static files, leaks filesystem information in 404 responses, and constructs HTML without proper escaping. Restrict it to localhost or add authentication, constrain file access to an intended directory, and HTML-escape all interpolated content.