Ask HN: What's your current AI agentic setup?

Re:Ask HN: What's your current AI agentic setup? by tikkabhuna on Thursday November 05, 2026 @09:03AM (#50379187)

Claude code uses bubblewrap[1] to offer some isolation. I'd love to hear from others whether its a reasonable solution or not.

[1] https://github.com/containers/bubblewrap

[ Parent | on HN ]

Re:Ask HN: What's your current AI agentic setup? by Matl on Thursday November 05, 2026 @09:15AM (#50379315)

Also used by ZeroStack[1], which I quite like, but whether it provides 'meaningful' protection am less sure.

1 - https://github.com/gi-dellav/zerostack

[ Parent | on HN ]

Re:Ask HN: What's your current AI agentic setup? by rdg1991 on Thursday November 05, 2026 @10:17AM (#50380164)

Claude frequently has decided to disable its own sandbox when I've had it running so I absolutely do not trust their implementation. Wrapping with `nono` or `sbx` has mitigated that on my side.

[ Parent | on HN ]

Re:Ask HN: What's your current AI agentic setup? by f53d on Thursday November 05, 2026 @02:36PM (#50383403)

I have also experienced this. Now I run Claude Code (or preferably Pi) inside a bubblewrap instance spawned and controlled by me that the agentic harness cannot disable by it's own choice.

[ Parent | on HN ]