The Cage

Flock launches agents at full throttle: no permission prompts, no approval dialogs. That's the point of a cockpit, and it's also a real decision to let a process loose on your machine. Secure mode resolves the tension: flip the toggle when creating a workspace and every agent in it (including later splits and spawns) runs inside a Docker container instead of on your Mac. Full permissions inside the jail; nothing outside it.

What the agent sees

Setup

  1. Install and start Docker Desktop, the same dependency the Graph uses.
  2. Create a workspace and enable Secure mode: jail agents in a container. The toggle tells you if Docker isn't ready.
  3. The first secure pane builds the sandbox image right there in the terminal (a few minutes, once). Every spawn after that is instant.

Jailed panes carry a mint jailed chip in their top bar. Closing a pane tears its container down; the sandbox image rebuilds itself automatically when a flock update changes the recipe.

The honest fine print

Secure mode fails closed: if Docker isn't running, secure panes refuse to spawn. They never quietly run unjailed. Network egress stays open, because agents need their APIs, so treat the mounted repo itself as something the agent could send out. And Ctrl+C drops you to a shell inside the container, never to your Mac. The worst a rogue agent can do is trash the repo checkout, and git can undo that.