Safe Agent Podman

Create a command-line tool, similar to a development environment manager, that uses Podman to create a sandboxed container for running Claude Code. The tool should manage the container lifecycle using commands like `up`, `down`, `shell`, and `exec`. Key features include restricting network access by default to only Anthropic endpoints, while allowing users to whitelist additional hosts using flags like `--allow-host`. It should also support enabling a `bypassPermissions` flag and logging API calls via a host proxy. The CLI should be built using Node.js/npm, and the core logic should manage Docker/Podman container operations.