Show HN: I built a sandboxed VM for letting AI agents go wild without risks

3 points

4 months ago

I wanted to run Claude Code in full autonomous mode (--dangerously-skip-permissions) but was terrified of what it might do to my system. One hallucinated rm -rf and my weekend is gone.

So I built YOLO Kingdom: a pre-configured Ubuntu VM (UTM/Apple Silicon) with Claude Code + Chrome DevTools MCP for autonomous browser testing. Share your project folders via virtiofs, SSH in, run `yolo`, and let it cook.

The VM is your sacrificial sandbox. It can delete everything inside and you just restore from snapshot.

V1 includes: Ubuntu 24.04, Claude Code, headless Chromium with DevTools protocol, smart welcome screen that guides setup.

Future plans: autonomous agents that can architect and build entire projects while you sleep.

https://github.com/pancakeInDev/yolo-kingdom