3 ms·
They're only as dangerous as the capabilities you give them. I just created a `codex` and `claude` user on my Linux box and practically always run in yolo mode.
by athrowaway3z 1y ago
They're only as dangerous as the capabilities you give them. I just created a `codex` and `claude` user on my Linux box and practically always run in yolo mode. I've not had a problem so far.
Also, I think shellagent sounds cooler.
- simonw 1y agoThat's a great way to run this stuff. I expect the portion of Claude Code users who have a dedicated user setup like this is pretty tiny!
- steveklabnik 1y agoOne nice thing is that Anthropic provides a sample DevContainer: https://github.com/anthropics/claude-code/tree/main/.devcontainer https://github.com/anthropics/claude-code/tree/main/.devcont... Not the exact setup, but also pretty solid.
- globular-toast 1y agoI tried this but it's incredibly annoying as you'll get a mixture of file ownerships and permissions. Instead I run it in bubblewrap sandbox: https://blog.gpkb.org/posts/ai-agent-sandbox/ https://blog.gpkb.org/posts/ai-agent-sandbox/
- jama211 1y agoThis is a really clever solution
- athrowaway3z 1y agoSet umask from 022 to 002 to give group members the same permissions as a user.
- raphman 1y agoThanks - seems to work quite well.
- tuyiown 1y ago> They're only as dangerous as the capabilities you give them. As long as the supply chain is safe and the data it accesses does not generate some kind of jail break. It does read instructions from files on the file system, I pretty sure it's not complex to have it poison its prompt and make it suggest to build a program infected with malicious intent. It's just one copy pasta away from a prompt suggestion found on the internet.
- data-ottawa 1y agoI have run it in a podman container and I mount the project directory.