Don't let Claude Code read your secrets: why you need to set up sandboxing today

While this is definitely a good first step, nothing stops Claude from following an injected prompt and writing malicious code in your writeable development directory, waiting for you to execute it manually with your full local permissions.

The point is that anything produced by Claude should only ever run in a sandboxed environment if you are really dead set on protecting yourself.

	▲	aozisik an hour ago \| parent [-]
		That’s right. I guess in the end it’s just a question of balancing security and convenience. If you lock everything down, it hurts productivity (or you start hitting “allow” or execute blindly). But if you don’t lock anything down, your system is just a sitting duck. I got too comfortable too fast with this skip permissions mode. All these recent global incidents made me slow down and think what the hell we have been doing on our personal computers. Not to mention all the OpenClaw people out there who have given AI full access to their gmails, etc. Now I feel even lucky nothing happened so far…