jai - AI 에이전트를 위한 손쉬운 격리 도구

jai addresses a critical vulnerability in AI tool usage: recent incidents of data loss from tools like Claude Code and Cursor accessing the home directory have exposed security gaps. This Stanford-developed open-source sandboxing tool isolates AI agents on Linux with minimal overhead, offering three isolation modes (Casual, Strict, Bare) to balance security and functionality. By protecting the home directory with copy-on-write overlays without complex container setup, jai provides a practical defense against unintended file system damage from AI tools.