Running Claude Code dangerously (safely)

18 days ago

Copy Link

Using Claude Code with the --dangerously-skip-permissions flag allows it to execute actions without asking for permission, improving workflow efficiency.
Docker was considered for isolation but rejected due to the need for Docker-in-Docker, which compromises security by requiring privileged mode.
Vagrant was chosen as a solution for its VM isolation, reproducibility, and shared folder capabilities, avoiding Docker-in-Docker issues.
A Vagrantfile configuration is provided, setting up an Ubuntu VM with Docker, Node.js, and Claude Code installed, ready for use.
Claude Code in this setup can perform tasks like starting webapps, installing browsers for testing, setting up databases, and building Docker images safely.
Performance is adequate with the VM, and shared folder sync works well, though initial setup may take time.
Safety measures protect against accidental filesystem damage and configuration changes but not against deliberate VM escapes or data exfiltration.
The setup is recommended for users of Claude Code with dangerous flags, offering a balance between functionality and safety.

Hasty Briefsbeta