Multiplayer Cloud Workspace & Homelab GPU Agent Template

Cloud development environment with real-time multiplayer (Browser Live Share, Zed, or tmux) connected to a 32GB VRAM Homelab GPU cluster via Tailscale Funnel + LiteLLM, with automatic failover to OpenRouter Free and Gemini—plus up to 20 parallel GitHub Actions subagents and the full Workstation Code Craft & Review Suite.

---

🎮 Multiplayer Modes (No Cursor Required)

1. Browser Co-Editing (Zero Installs): Click Open in GitHub Codespaces above -> Click the Live Share icon on the left sidebar -> Send the invite link to friends. Everyone gets independent cursors, shared terminals, and shared localhost preview ports in their browser.
2. Terminal Multiplayer (tmux + Neovim / Helix / Pi / agy): Host runs pair-join inside the Codespace terminal. Friends connect from their own terminal via:

   gh codespace ssh
   pair-join

3. Zed Native Multiplayer: Run gh codespace ssh --config >> ~/.ssh/config, connect via Zed Remote Server, and share the project channel.

---

🍽️ On-Demand Code Craft & Review Hub (doordash)

Pre-installed inside every Codespace terminal. Enforce senior engineering rigor and automated reviews without prompt bloat:

# View available kitchens
doordash menu

# Order specific reviews on your working tree or pull request
doordash order ponytail              # Minimal diff & YAGNI audit
doordash order poteto                # Domain modeling & State Transition Matrix
doordash order thermonuclear         # Structural audit & complexity elimination
doordash order create-verification   # Scaffold [REQ-XX] Grounded Verification tests
doordash order unslop                # 31-rule AI tell scanner

# Run the full review combo
doordash order combo

---

📦 Built-In Agent Skills & Tools

Skill Directory	Command	Purpose
.agents/skills/craft-hub	doordash / craft-hub	Late-binding DoorDash API for code reviews, state matrices, and verification.
.agents/skills/skills-hub	skills-hub	Meta-tool gateway: dynamically search (skills-hub search) and run tools without context bloat.
.agents/skills/graph-memory	graph-memory	Hybrid BM25 + embedding RAG over project notes, with exact-quote LLM distillation.
.agents/skills/markitdown-cli	markitdown-cli	Universal document converter to Markdown (PDF, DOCX, XLSX, PPTX, CSV, HTML).
.agents/skills/gh-action-subagent	gh-subagent	Dispatch up to 20 parallel cloud subagents inside GitHub Actions VMs via PRs.
.agents/skills/openrouter-free	openrouter-free	Multi-model fallback (Nemotron, Qwen Coder, Gemini 2.5 Flash).
.agents/skills/tailscale-ops	tailscale-ops	Private mesh networking and Tailscale Funnel management.
.agents/skills/litellm-vram-switcher	vram-rig-ctl	Dynamic GPU VRAM switcher and LiteLLM gateway.
.agents/skills/cdp-browser	cdp-browser	Headless Chrome DevTools Protocol browser automation.

---

⚡ Connecting Your Homelab GPU Rig to the Cloud Workspace

1. On your home GPU PC, expose LiteLLM (:4000) over Tailscale Funnel:

   ./scripts/host_expose_gpu.sh

2. Add Secrets: Copy the printed HOMELAB_LITELLM_URL and HOMELAB_LITELLM_KEY into your GitHub Repo's Codespaces Secrets and Actions Secrets.
3. Automatic Bridging: Inside the Codespace, the bridge is automatically configured on shell startup:
   • When your PC is awake: Cloud agents and human collaborators run inference against your dual RX 9060 XT local models (8080–8090).
   • When your PC is off or gaming (evict-all): Requests seamlessly fail over to openrouter/free and gemini-2.5-flash.
   • Test the bridge anytime with:

     ./scripts/connect_homelab_gpu.sh

---

📁 Repository Structure

agent-multiplayer-template/
├── .devcontainer/
│   ├── devcontainer.json        # Universal container (Node 22, Live Share, port forwarding)
│   └── bootstrap.sh             # Auto-symlinks doordash, skills-hub, and sets up pair-join
├── .github/workflows/
│   └── subagent.yml             # Parallel subagents executing tasks via Pull Requests
├── .agents/skills/              # Curated workstation & cloud skills
│   ├── craft-hub/               # DoorDash code review hub
│   ├── skills-hub/              # Meta-tool discovery gateway
│   ├── graph-memory/            # Local hybrid RAG & distillation
│   ├── markitdown-cli/          # Document to Markdown converter
│   ├── gh-action-subagent/      # Parallel cloud subagents
│   ├── litellm-vram-switcher/   # LiteLLM failover manager
│   ├── tailscale-ops/           # Tailscale mesh & funnel routing
│   ├── openrouter-free/         # Free multi-model inference fallback
│   └── cdp-browser/             # Headless browser automation
└── scripts/
    ├── host_expose_gpu.sh       # Expose home PC LiteLLM securely over Tailscale Funnel
    ├── connect_homelab_gpu.sh   # Bind Codespace environment to homelab GPU
    └── publish_template.sh      # Publish & configure repository as a GitHub Template