Skip to content

Worktrees

slicer worktree lets you push a git repository or worktree into a Slicer microVM, work on it in isolation in the VM or run an agent there, and pull the commits back to your host branch. This is useful when you want to give an agent a fully isolated environment, without losing the ability to commit, review, and push changes from your host under your own identity.

Use slicer worktree to:

  • Push a whole repo — you want to work on or run an agent against your current branch in an isolated VM, then bring the commits back.
  • Push a worktree — you want to work on a feature branch in a VM while keeping your main working directory untouched on the host, without stashing or committing unfinished changes.
  • Use an agent sandbox — you want to provision a VM with a coding agent installed and then sync a worktree to have the agent work on the code in isolation.

Push a repo into a VM

Push the current directory into a new VM, open a shell to work in it, then pull commits back:

cd ~/src/myrepo

# Launch a VM and push the repo in. Take note of the VM name
slicer worktree push --launch .

# Open a shell in the VM and work there
slicer worktree shell vm-1

# Pull commits back and fast-forward your host branch
slicer worktree pull vm-1 .

# Push to remote from your host
git push

To push into an existing VM instead:

slicer worktree push vm-1 .

Push a worktree into a VM

If you want to work on a feature branch in a VM while keeping your main working directory untouched on the host, create a git worktree first. This lets you continue work on the host without stashing or committing unfinished changes, and gives each branch its own isolated VM.

Create a worktree for a branch, push it into a dedicated VM, and sync commits back:

cd ~/src/myrepo
# Create a git worktree on a new feature branch
git worktree add ../myrepo-feature -b feature
cd ../myrepo-feature

# Launch a VM and push the worktree in. Take note the VM name
slicer worktree push --launch .

# Open a shell in the VM and work there
slicer worktree shell vm-1

# Pull commits back and fast-forward your host branch
slicer worktree pull vm-1 .

# Push to remote from your host
git push

To push into an existing VM instead:

slicer worktree push vm-1 .

Don't edit the host worktree while a VM holds it

worktree pull overwrites host files with the VM's copy. Anything changed on the host since the push will be lost. Treat the host worktree as read-only until you have pulled.

Use a worktree with agent sandbox commands

Agent commands (slicer opencode, slicer claude, slicer codex, and others) boot a VM and install a coding agent, see Sandboxing Coding Agents. When run without a path argument they skip copying any code into the VM, giving you a clean environment ready to receive a worktree.

Use slicer worktree push to sync the worktree into the VM afterwards:

# Provision VM only — coding agent installed, no code copied in
# Take note of the VM name
slicer opencode

# Push the worktree into the vm
slicer worktree push vm-1 .

# Attach and let the agent work
slicer opencode vm-1

# Pull commits back and fast-forward your host branch
slicer worktree pull vm-1 .

# Push to remote from your host
git push

Passing a path directly to an agent command (e.g. slicer opencode .) copies the working directory into the VM but does not have an easy command to re-sync commits back to your host. Use slicer worktree push whenever the checkout is a git worktree.