coop is a Rust CLI tool that spins up isolated, ephemeral virtual machines for running Claude Code and Codex with full tool access—Docker, git, compilers, package managers—without exposing your host system to risk. It uses Firecracker on Linux and Lima on macOS to create cheap, reproducible VMs you can destroy after each session. Installation is straightforward: curl the installer or build from source, then run `coop setup` to build the VM template. Day-to-day use looks like `coop up` to launch an instance in your project directory, then `coop claude` or `coop codex` to invoke the agent. The source material is documentation boilerplate—it doesn't detail threat model, isolation boundaries, or failure modes, so you'd need the linked docs to understand what risks remain or what actually happens when an agent escapes. One sharp question: does coop isolate network access, or can Claude/Codex reach arbitrary hosts from inside the VM?