Opinionated devcontainer launcher to run agents in yolo mode with persistent config across containers. Installs a ready-to-go template into a repo, brings the container up, and drops you into tmux. Based on the banteg/devc.

• Container runtime: Docker Desktop or OrbStack (recommended).
• devcontainer CLI: npm install -g @devcontainers/cli.

Homebrew:

brew tap egornomic/cage
brew install cage

Fallback (from source):

./install.sh

This installs cage to ~/.local/bin and the template to ~/.local/share/cage/template.

• cage <repo> or cage .: install template, start devcontainer, open tmux.
• cage install <repo>: install template only.
• cage rebuild <repo>: replace template, rebuild container, open tmux.
• cage exec <repo> -- <cmd>: run a command inside the devcontainer.
• cage mount <repo> <host_path> <container_path> [--readonly]: add a bind mount and recreate the container.
• cage upgrade <repo>: update claude/codex/pi inside the devcontainer.

Notes:

• After first install, log in once (or paste auth tokens from host). Auth persists across containers via mounted config volumes.
• Installs/overwrites .devcontainer in the target repo.
• Containers are named after the repo and, with OrbStack, are reachable from the host at <name>.orb.local.
• CAGE_TEMPLATE_DIR overrides the template source.
• CAGE_SCREENSHOTS_DIR overrides the screenshots mount (defaults to ~/Desktop).
• Screenshots are mounted because you can't paste images into a devcontainer; dropping files into the host screenshots folder is the simplest bridge.
• macOS default screenshot shortcuts: ⌘⇧3 saves the full screen to a file, ⌘⇧4 saves a selection.