PlasmaPOS DevContainer Images

Pre-built Docker images for DevPod cloud development environments. These images are the base layer — PlasmaPOS/dotfiles handles runtime configuration (env vars, git identity, shell aliases).

Repository Structure

devcontainers/
├── Dockerfile.light            # Backend/tooling projects
├── Dockerfile.full             # App projects with mobile + web testing
└── .github/workflows/build.yml # Auto-builds on push to main

Images

light — Backend & Tooling

Registry: ghcr.io/plasmapos/devcontainers/light:latest For: dev-agent, seal, aqua

Layer	What's installed
Base	Debian (devcontainers base image)
Runtime	Node.js 22 (system-wide), Bun (latest)
Dev tools	Convex CLI (via bun global)
System	Bun symlinked to /usr/local/bin for non-interactive SSH

full — Apps with E2E Testing

Registry: ghcr.io/plasmapos/devcontainers/full:latest For: plasma, pile, roster, veil

Everything in light, plus:

Layer	What's installed
Browser	Playwright + Chromium (browsers in /home/vscode/.cache/ms-playwright)
Mobile	Android SDK 34, platform-tools, emulator
AVD	Pre-configured Pixel 6 AVD (test-device) with Google APIs x86_64
Dev tools	agent-device CLI (via bun global)
System	Java JDK (headless, required by Android SDK)

VM Sizing

Image	Projects	GCP Machine Type	vCPU	RAM
light	dev-agent, seal, aqua	n2d-standard-4	4	16 GB
full	plasma, pile, roster, veil	n2d-standard-8	8	32 GB

Full-image projects need 8 vCPU because running emulator + dev server + AI CLIs concurrently on 4 vCPU causes OOM.

Override per workspace:

devpod up <repo> --provider-option MACHINE_TYPE=n2d-standard-8

Usage in devcontainer.json

Each project repo has a .devcontainer/devcontainer.json that references one of these images:

{
  "name": "my-project",
  "image": "ghcr.io/plasmapos/devcontainers/light:latest",
  "features": {
    "ghcr.io/devcontainers/features/git:1": {},
    "ghcr.io/devcontainers/features/github-cli:1": {}
  },
  "postCreateCommand": "bun install",
  "runArgs": ["--shm-size=2g"]
}

Note: Node.js is already baked into the image — do NOT add the ghcr.io/devcontainers/features/node:1 feature (it would install a duplicate).

Building

Images are auto-built on push to main via GitHub Actions.

# Manual trigger
gh workflow run build.yml

# Build locally for testing
docker build -f Dockerfile.light -t devcontainers-light .
docker build -f Dockerfile.full -t devcontainers-full .

Design Decisions

AI CLIs (Claude Code, Codex) run locally on your machine — they connect to the workspace via ssh <project>.devpod. They are NOT installed inside the container image. This keeps the image small and avoids version drift between local AI tools and remote copies.

Why symlink bun into /usr/local/bin?

Same reason — non-interactive SSH sessions don't source .bashrc/.profile. The symlink ensures bun and bunx are available everywhere. The actual bun installation lives at /home/vscode/.bun/bin/bun.

Why Chromium only (no Firefox/WebKit)?

Chromium covers our testing needs. Each additional browser adds ~200 MB to the image and increases build time. Add others if needed:

RUN npx playwright install --with-deps chromium firefox

For Coding Agents

If you are an AI agent and need to understand the environment:

• Bun: /home/vscode/.bun/bin/bun (also at /usr/local/bin/bun)
• Node: /usr/bin/node (v22, system-wide)
• Convex CLI: ~/.bun/bin/convex (also via bunx convex)
• Playwright browsers (full image only): /home/vscode/.cache/ms-playwright
• Android SDK (full image only): /opt/android-sdk
• Android emulator AVD (full image only): test-device (Pixel 6, API 34, Google APIs x86_64)
• Default user: vscode (the standard devcontainers user)
• Runtime config (env vars, git, aliases) is handled by PlasmaPOS/dotfiles, not this image
• AI CLIs (Claude Code, Codex) are NOT in the image — run them locally, they SSH into the workspace