问HN:为什么这么多人推出自己的AI/LLM代理沙盒解决方案?
看到很多人使用自定义沙箱(如Docker/VMs、firejail/bubblewrap)来运行编码代理(如Claude Code等),以及一些限制文件或网络访问的脚本。我很好奇,是什么导致人们需要自己动手做这些?那么,什么样的标准才算是“足够好”?
查看原文
Seeing a lot of people running coding agents (Claude Code, etc.) in custom sandboxes Docker/VMs, firejail/bubblewrap, scripts that gate file or network access.<p>Curious to know what's missing that makes people DIY this? And what would a "good enough" standard look like?