Skip to content

Instantly share code, notes, and snippets.

@opencoca

opencoca/docker-quickstart.md

Last active April 6, 2026 14:21
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save opencoca/953205010bfb77a7846f8f651c0e5e81 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save opencoca/953205010bfb77a7846f8f651c0e5e81 to your computer and use it in GitHub Desktop.
Download ZIP
gstack Docker quick-start — run Claude Code + browser skills without cloning the repo
Raw
docker-quickstart.md

gstack in Docker — From Zero to AI Pair Programmer in 60 Seconds

"One docker run and you have Claude Code, a headless browser, and 20+ AI skills. No install. No config. No kidding."

You don't need to clone a repo. You don't need Node.js. You don't even need an API key upfront. Just Docker and a minute of your time.

What you're getting

This isn't just Claude in a box. It's a full AI development environment:

  • Claude Code CLI — Anthropic's coding agent, pre-installed and ready
  • Headless Chromium — Playwright-managed, sandbox already handled (no --security-opt flags, no kernel workarounds)
  • 20+ gstack skills — /qa, /ship, /browse, /investigate, /design-review, /benchmark, /canary, and more
  • Bun runtime — fast JS/TS toolchain under the hood
  • Multi-arch — runs on amd64 and arm64 (yes, your M-series Mac too)

"It's the mass-production version of 'works on my machine.' It works on every machine."

Prerequisites

  • Docker (Desktop or Engine)

That's the whole list.

Step 1: Pull

docker pull ghcr.io/opencoca/gstack:latest

Step 2: Run

docker run -it --rm \
  -v agent-claude:/home/agent/.claude \
  -v agent-codex:/home/agent/.codex \
  -v agent-config:/home/agent/.config \
  -v agent-cache:/home/agent/.cache \
  -v "$PWD":/workspace \
  ghcr.io/opencoca/gstack:latest bash

You're in. The container drops you into a shell as user agent, with your current directory mounted at /workspace. Any skill repos with SKILL.md files are auto-discovered and wired into Claude — no setup scripts needed.

Step 3: Log in

claude login

Follow the link in your terminal — web-based OAuth is the way to go. It gives you the full interactive experience and your session persists across container restarts (thanks to the agent-claude volume).

API keys work too if that's your thing:

export ANTHROPIC_API_KEY=sk-ant-...

Step 4: Build something

claude

Or skip the chat and put it to work:

claude -p "explain this codebase and find the three biggest risks"

"From docker pull to shipping features with an AI pair programmer — under a minute. The future didn't need a setup wizard."

The one-liner (for the impatient)

Skip the shell entirely, launch Claude directly against your project:

docker run -it --rm \
  -v agent-claude:/home/agent/.claude \
  -v agent-codex:/home/agent/.codex \
  -v agent-config:/home/agent/.config \
  -v agent-cache:/home/agent/.cache \
  -v "$PWD":/workspace \
  ghcr.io/opencoca/gstack:latest claude

Your config survives restarts

Named Docker volumes keep your auth, history, and settings between runs. Nothing is lost when the container stops.

Volume What lives there
agent-claude Auth tokens, config, conversation history, automatic backups
agent-codex Codex config (for multi-AI workflows)
agent-config General app preferences
agent-cache Runtime cache (speeds up repeat operations)

The container entrypoint handles first-run setup automatically — creates directories, symlinks config files, and even restores from backups if something goes sideways. You don't think about it.

Make it stick: shell aliases

Those docker run commands are long. You should only type them once. Drop these in your ~/.bashrc, ~/.zshrc, or wherever you keep aliases:

# Interactive shell — explore, install tools, poke around
alias gstack='docker run -it --rm \
  -v agent-claude:/home/agent/.claude \
  -v agent-codex:/home/agent/.codex \
  -v agent-config:/home/agent/.config \
  -v agent-cache:/home/agent/.cache \
  -v "$PWD":/workspace \
  ghcr.io/opencoca/gstack:latest bash'

# Jump straight into Claude against your current directory
alias gstack-claude='docker run -it --rm \
  -v agent-claude:/home/agent/.claude \
  -v agent-codex:/home/agent/.codex \
  -v agent-config:/home/agent/.config \
  -v agent-cache:/home/agent/.cache \
  -v "$PWD":/workspace \
  ghcr.io/opencoca/gstack:latest claude'

Then reload your shell:

source ~/.zshrc  # or ~/.bashrc

Now it's just:

cd my-project
gstack-claude

"Two aliases. That's the entire install. gstack for a shell, gstack-claude to start building."

Level up: optional extras

Mount your fonts (sharper screenshots)

Chromium renders pages with whatever fonts are available. For screenshots that look like they came from a real browser, mount your system fonts:

macOS:

docker run -it --rm \
  -v agent-claude:/home/agent/.claude \
  -v agent-config:/home/agent/.config \
  -v "$PWD":/workspace \
  -v /System/Library/Fonts:/usr/share/fonts/system:ro \
  -v /Library/Fonts:/usr/share/fonts/local:ro \
  ghcr.io/opencoca/gstack:latest bash

Linux:

  -v /usr/share/fonts:/usr/share/fonts/system:ro \
  -v /usr/local/share/fonts:/usr/share/fonts/local:ro

Pass API keys via environment

Web login is recommended, but if you're running headless or in CI, env vars work:

docker run -it --rm \
  -e ANTHROPIC_API_KEY \
  -v agent-claude:/home/agent/.claude \
  -v "$PWD":/workspace \
  ghcr.io/opencoca/gstack:latest claude

Or drop them in a .env file:

docker run -it --rm --env-file .env \
  -v agent-claude:/home/agent/.claude \
  -v "$PWD":/workspace \
  ghcr.io/opencoca/gstack:latest claude
Variable What it unlocks
ANTHROPIC_API_KEY Claude (primary)
OPENAI_API_KEY Codex / GPT features
GEMINI_API_KEY Gemini features
ANTHROPIC_AUTH_TOKEN Alternative Claude auth
ANTHROPIC_BASE_URL Custom/proxy endpoint

Install gstack without Docker

Prefer skills on your bare metal? Clone from opencoca/gstack — the fork with the safety harness:

Claude Code (global — available everywhere):

git clone --depth 1 https://github.com/opencoca/gstack.git ~/.claude/skills/gstack
cd ~/.claude/skills/gstack && ./setup

Claude Code (project-local — shared with your team via git):

git clone --depth 1 https://github.com/opencoca/gstack.git .claude/skills/gstack
cd .claude/skills/gstack && ./setup --local

Codex / Gemini / Cursor:

git clone --depth 1 https://github.com/opencoca/gstack.git .agents/skills/gstack
cd .agents/skills/gstack && ./setup --host codex

The ./setup script builds everything, creates skill symlinks, and asks if you want short names (/qa) or namespaced (/gstack-qa). Pass --no-prefix or --prefix to skip the prompt.

Why opencoca? Same skills, better defaults. The opencoca fork adds a safety harness — manipulation detection, local-first privacy enforcement, and a one-switch no-egress mode to hard-disable all hosted services. You audit your tools; your tools don't audit you.

Cheat sheet

# Pull the image
docker pull ghcr.io/opencoca/gstack:latest

# Interactive shell with persistent config
docker run -it --rm \
  -v agent-claude:/home/agent/.claude \
  -v "$PWD":/workspace \
  ghcr.io/opencoca/gstack:latest bash

# Jump straight into Claude
docker run -it --rm \
  -v agent-claude:/home/agent/.claude \
  -v "$PWD":/workspace \
  ghcr.io/opencoca/gstack:latest claude

# With API key (headless/CI)
docker run -it --rm \
  -e ANTHROPIC_API_KEY \
  -v "$PWD":/workspace \
  ghcr.io/opencoca/gstack:latest claude

# Log in (inside the container)
claude login

"AI coding assistants shouldn't require a PhD in DevOps to set up. docker run and go."

Happy hacking.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment