[moderation]

There is a lot of chatter on Twitter and here about sandboxes for AI, however there appears to be a lack of awareness of the native built in sandboxing capabilities of Claude Code, Codex and Gemini CLI. Claude Code, Codex and Gemini CLI all use seatbelt on MacOS. Claude Code uses bubblewrap on Linux. Codex uses seccomp + landlock on Linux. Codex has experimental native sandboxing on Windows with AppContainer.

[RandomPoes]

Interesting, but do these native sandboxes limit access only to specific files? And I'm not sure, but when these agents invoke a system command, is that also sandboxed, or is it only the agent process itself that's sandboxed (assuming that is even useful)?

[moderation]

This is Claude Code specific but there are similar capabilities for Codex.

"These OS-level restrictions ensure that all child processes spawned by Claude Code’s commands inherit the same security boundaries." [0]

There is a rich deny and allow system for file access that can be used in conjunction with the sandbox [1]

0. https://code.claude.com/docs/en/sandboxing#os-level-enforcem...

1. https://code.claude.com/docs/en/settings#excluding-sensitive...