Feature request – “container” execution mode for Claude Code

[mulyoved]

Why

1. Zero local toolchain fuss
   A Docker image can bundle Claude Code, Node, Python, pip, npm, etc. so every dev machine — laptop, CI runner, or cloud VM — sees the same stack. No more “works on my machine” when a teammate upgrades Node.
2. Run anywhere you have a Docker daemon
   Point DOCKER_HOST (or a Docker Context) at an on-prem box, a beefy EC2/GCE VM, or even a Kubernetes worker. Your local Claudia UI streams stdout/stderr exactly as today, but the CPU/RAM live remotely.

What

• Add an executionMode: "process" | "container" setting.
• DockerExecutor wraps the current Command interface:

  docker run --rm \
    -v "${projectPath}:/workspace" -w /workspace \
    -e CLAUDE_API_KEY \
    asterisk/claude-cli:<tag> <args>

• Images can be per-project (docker build -f Dockerfile.claude .) or a shared tag.

Nice side-effects

• Windows friendliness – users still need Docker Desktop (WSL 2/Hyper-V backend) but no manual WSL install of Claude, Node, etc.
• Reproducible builds – image tag pins language/tool versions.
• Security – combines Claudia’s existing sandbox with the kernel isolation of containers.

Sandbox ≠ Dependencies

Claudia’s sandbox hardens file/network access (seccomp, ACLs) but does not freeze tool versions; containers compliment that by pinning runtimes

Implementation

I guess the most practical approach would be to:

• Create a Docker abstraction layer that implements the same interface as the current command execution
• Modify the command creation functions to use Docker API instead of direct process spawning
• Handle file project access, as need also remote docker, volume mount is not enogh
• Adapt the streaming output handling to work with containerized processes

Does it make sense?

[git-git-hurrah]

I need this one as well. All my coding agents are just running in devcontainers. Most people I know don't work on the host machine either, so I think this would be a needed feature to really use Claudia.