add claude code support

Author: marcoieni

README.md

@@ -3,8 +3,9 @@
 > [!NOTE]
 > This project is meant for people working on the Rust Project infrastructure.
 
-Run codex in a Linux VM using [lima](https://lima-vm.io/).
-Control it from [ChatGPT desktop](https://chatgpt.com/download/).
+Run Codex and Claude Code in a Linux VM using [lima](https://lima-vm.io/).
+Use either CLI over SSH, or control Codex from
+[ChatGPT desktop](https://chatgpt.com/download/).
 
 The VM has read-only access to Cloud services like DataDog and Fastly.
 Credentials are read automatically from the Rust Foundation 1Password.
@@ -50,15 +51,22 @@ Credentials are read automatically from the Rust Foundation 1Password.
   you are experiencing issues, check the
   [authentication docs](./docs/authentication.md).
 
-- Enable "Device code authorization for Codex" in
+- If using Codex, enable "Device code authorization for Codex" in
   [ChatGPT security settings](https://chatgpt.com/#settings/Security).
 
-- Login to Codex, Datadog and Fastly from the guest:
+- Login to Codex, Claude Code, Datadog and Fastly from the guest:
 
   ```sh
   just login
   ```
 
+  For Claude Code, open the URL printed in the terminal in your host's browser
+  and paste the returned login code into the terminal when prompted. See the
+  [Claude Code SSH login instructions](https://code.claude.com/docs/en/troubleshoot-install#oauth-login-fails-in-wsl2-ssh-or-containers).
+
+  To use only one coding agent, run `just login-codex` or `just login-claude`,
+  followed by `just login-datadog login-fastly`.
+
 ### 3. Set up projects
 
 The host directory `~/buddy` is mounted inside the VM at `$HOME/work`.
@@ -86,7 +94,12 @@ Place your projects in the host directory:
   ssh lima-buddy
   ```
 
-- (Optional) Launch the `codex` command inside the VM to start the Codex TUI.
+- Launch `codex` or `claude` inside the VM from your project directory:
+
+  ```sh
+  cd ~/work/<project>
+  claude
+  ```
 
 See the [Lima SSH documentation](https://lima-vm.io/docs/usage/ssh/) for more details.
 
@@ -112,6 +125,10 @@ for more details.
 
 - Check the [justfile](./justfile) for other available commands.
 
+- Run `just upgrade` to update the guest and both coding agents. This also
+  installs Claude Code in an existing VM. New VMs install it automatically
+  using the [native installer](https://code.claude.com/docs/en/installation).
+
 - Run commands in the VM:
 
   ```sh
@@ -134,14 +151,14 @@ for more details.
 
 The template explicitly sets `user.passwordlessSudo: true`, which is
 [Lima's default for Linux guests](https://lima-vm.io/docs/config/sudo/).
-This gives Codex unrestricted root access inside the guest, but does not grant
-root access on the host. The VM boundary and the resources exposed to the VM,
-such as the writable `$HOME/work` mount, remain the main security boundary.
+This gives Codex and Claude Code unrestricted root access inside the guest, but
+does not grant root access on the host. The VM boundary and the resources exposed
+to the VM, such as the writable `$HOME/work` mount, remain the main security boundary.
 
 Enabling passwordless sudo has this advantage:
 
-- Codex and other automation can install packages, update the guest, and fix
-  system configuration without waiting for a password prompt.
+- Codex, Claude Code, and other automation can install packages, update the guest,
+  and fix system configuration without waiting for a password prompt.
 
 It also has these disadvantages:
 
@@ -154,10 +171,10 @@ It also has these disadvantages:
 
 ## FAQ
 
-> Why not running Codex directly on the host?
+> Why not run Codex or Claude Code directly on the host?
 
-Auditing all commands that codex wants to run is not productive. Instead, by
-running it in an isolated VM, you can run codex in yolo mode.
+Auditing all commands that coding agents want to run is not productive. Running
+them in an isolated VM lets you give them full permissions inside the guest.
 
 > Why not using one VM per project?
 
@@ -174,10 +191,10 @@ Lima provides:
 - It's an open source [CNCF project](https://www.cncf.io/projects/lima/), while Docker Desktop is a proprietary product.
 - The Docker Desktop feature [Enhanced Container Isolation](https://docs.docker.com/enterprise/security/hardened-desktop/enhanced-container-isolation/), which "prevents malicious containers from compromising the host system" is restricted to Docker Business.
 
-> I don't want to run codex with full privileges, I think it is dangerous!
+> I don't want to run coding agents with full privileges, I think it is dangerous!
 
-Nobody forces you to run codex with full privileges. You can still run it in the
-`buddy` VM for improved security _and_ customize its permissions.
+Nobody forces you to run Codex or Claude Code with full privileges. You can still
+run either in the `buddy` VM for improved security _and_ customize its permissions.
 
 ## License
 

buddy.yaml

@@ -16,7 +16,7 @@ ssh:
   forwardAgent: false
   loadDotSSHPubKeys: false
 
-# Codex can elevate to root inside the VM
+# Codex and Claude Code can elevate to root inside the VM
 # without receiving or storing the guest user's password.
 user:
   passwordlessSudo: true

justfile

@@ -29,11 +29,14 @@ delete: stop
 delete-file path:
     limactl shell "{{ vm }}" bash -lc 'rm -f "{{ path }}"'
 
-login: login-codex login-datadog login-fastly
+login: login-codex login-claude login-datadog login-fastly
 
 login-codex:
     limactl shell "{{ vm }}" bash -lc 'codex login --device-auth'
 
+login-claude:
+    limactl shell "{{ vm }}" bash -lc 'claude auth login'
+
 login-datadog:
     cargo run --quiet -- login-datadog "{{ vm }}"
 
@@ -54,9 +57,11 @@ logout-fastly: (delete-file (config_dir / "fastly.env"))
 # a full upgrade would make startup slower, less predictable, and could install kernel updates requiring another reboot.
 upgrade:
     limactl shell "{{ vm }}" sudo apt-get update
+    limactl shell "{{ vm }}" sudo apt-get install -y socat
     limactl shell "{{ vm }}" sudo apt-get upgrade
     limactl shell "{{ vm }}" bash -lc 'brew update && brew upgrade --yes'
     limactl shell "{{ vm }}" bash -lc 'curl -fsSL https://chatgpt.com/codex/install.sh | CODEX_NON_INTERACTIVE=1 sh'
+    limactl shell "{{ vm }}" bash -o pipefail -lc 'curl -fsSL https://claude.ai/install.sh | bash'
     # stop the old running app-server and start it again using the newly installed binary.
     limactl shell "{{ vm }}" bash -lc \
         'pkill -x codex || true; exec codex app-server daemon bootstrap'

provision/system.sh

@@ -3,7 +3,8 @@ set -Eeuo pipefail
 
 apt-get update
 
-# Codex uses bubblewrap for its Linux sandbox. This is needed if you want to run codex withot full access.
+# Codex and Claude Code use bubblewrap for their Linux sandboxes.
+# Claude Code also uses socat to proxy sandbox network traffic.
 apt-get install -y \
   build-essential \
   bubblewrap \
@@ -20,6 +21,7 @@ apt-get install -y \
   python3-venv \
   ripgrep \
   rsync \
+  socat \
   unzip \
   zip
 

provision/user.sh

@@ -35,5 +35,8 @@ brew install fastly/tap/fastly
 # Install or update codex
 curl -fsSL https://chatgpt.com/codex/install.sh | sh
 
-# The Codex installer adds ~/.local/bin to .bashrc
+# Install or update Claude Code
+curl -fsSL https://claude.ai/install.sh | bash
+
+# Make both native CLIs available to login shells, including SSH sessions.
 append_line_if_missing "export PATH=\"\$HOME/.local/bin:\$PATH\"" "$HOME/.profile"