name	openclaw-in-a-box
description	Orchestrator skill: detects environment, asks user which integrations to enable, drives setup, and hands off to sub-skills.
version	0.2.0
user-invocable	true
metadata	{"openclaw":{"emoji":"🦞","requires":{"bins":[],"env":[]}}}

openclaw-in-a-box

You are the setup orchestrator. Your job is to get the user from zero to a running OpenClaw agent inside a secure stereOS VM with Tapes telemetry.

Step 0: Clone if needed

If the repo isn't already cloned, clone it:

git clone https://github.com/papercomputeco/openclaw-in-a-box
cd openclaw-in-a-box

If already in the repo directory, skip this step.

Step 1: Detect environment

Run these checks and report a status summary:

# Host tools
command -v mb    # Master Blaster CLI
echo $ANTHROPIC_API_KEY | head -c 10  # API key present (don't print full key)

# Model provider
echo ${MODEL_PROVIDER:-not set}
command -v ollama   # Ollama CLI (for local models)
echo ${OLLAMA_API_KEY:+set}  # Ollama cloud key

# Integration tools
command -v gog   # Gmail bridge
gog auth list    # Gmail account connected
command -v gh    # GitHub CLI
gh auth status   # GitHub authenticated
echo ${DISCORD_TOKEN:+set}  # Discord token present

Print a short status table:

Environment:
  mb CLI:          ✓ installed
  ANTHROPIC_API_KEY: ✓ set

Model Provider:
  MODEL_PROVIDER:  ○ not set (defaults to anthropic)
  Ollama CLI:      ○ not found
  OLLAMA_API_KEY:  ○ not set

Integrations:
  Gmail:   ✗ gog CLI not found
  GitHub:  ✓ gh authenticated
  Discord: ✗ DISCORD_TOKEN not set

If mb is not installed, tell the user to install Master Blaster first: https://github.com/papercomputeco/masterblaster

If MODEL_PROVIDER is not set or is anthropic, and ANTHROPIC_API_KEY is not set, tell the user to export it: export ANTHROPIC_API_KEY="sk-ant-..."

If MODEL_PROVIDER is ollama and the model ends in :cloud, and OLLAMA_API_KEY is not set, tell the user to get one at ollama.com/settings.

Step 1.5: Choose model provider

Ask the user which model provider to use:

Anthropic (default) -- Claude models via Anthropic API. Requires ANTHROPIC_API_KEY.
Ollama Cloud -- Run models like kimi-k2.5:cloud or minimax-m2.7:cloud via Ollama's hosted API. Requires OLLAMA_API_KEY from ollama.com. No local GPU needed.
Ollama Local -- Run models locally via Ollama on the host machine. The VM connects to the host's Ollama instance. Requires Ollama installed on the host with a model pulled and enough RAM/VRAM.

Based on their choice, tell them which env vars to export:

Anthropic:

export MODEL_PROVIDER="anthropic"
export ANTHROPIC_API_KEY="sk-ant-..."

Ollama Cloud:

export MODEL_PROVIDER="ollama"
export MODEL_NAME="minimax-m2.7:cloud"   # or kimi-k2.5:cloud
export OLLAMA_API_KEY="..."              # from ollama.com/settings

Ollama Local:

export MODEL_PROVIDER="ollama"
export MODEL_NAME="llama3.3"             # or any pulled model
export OLLAMA_BASE_URL="http://host.docker.internal:11434"  # if VM can't reach host localhost
# No API key needed for local

Note for Ollama Local: the VM runs in an isolated network. The default http://localhost:11434 refers to the VM's own localhost, not the host. If Ollama runs on the host, the user may need to set OLLAMA_BASE_URL to the host's IP or a bridge address depending on their stereOS network config.

Popular Ollama cloud models for agentic work:

Model	Tag	Notes
MiniMax M2.7	`minimax-m2.7:cloud`	Built for coding and agentic tasks
Kimi K2.5	`kimi-k2.5:cloud`	Multimodal, strong agentic capabilities
MiniMax M2.5	`minimax-m2.5:cloud`	Productivity and coding

Step 2: Ask what they want

Ask the user which integrations to enable. Don't assume all three. Present only the ones that aren't already set up:

Gmail Triage — archive newsletters, label receipts, star action items
GitHub Org Triage — flag stale PRs, blocked issues, release risk
Discord Bot — respond to mentions, summarize threads

If everything is already configured, skip to Step 4.

Step 3: Drive setup for each chosen integration

Gmail setup

Install gog: brew install steipete/tap/gogcli
Walk user through Google Cloud Console:
- Create project at console.cloud.google.com/projectcreate
- Enable Gmail API
- Configure OAuth consent screen (External, add test user)
- Create Desktop app OAuth credentials
- Download client_secret_*.json
Register credentials: gog auth credentials ~/Downloads/client_secret_*.json
Authenticate (requires user interaction — opens browser):
```
gog auth add USER@gmail.com --services gmail
```
Verify: gog auth list and gog gmail labels list

GitHub setup

Install gh if missing: brew install gh
Authenticate (requires user interaction):
```
gh auth login
```
Or if user has a token: export GH_TOKEN="ghp_..."
Verify: gh auth status

Discord setup

Walk user through Discord Developer Portal:
- Create application at discord.com/developers/applications
- Create bot, copy token
- Enable Message Content Intent
- Generate invite URL with bot + applications.commands scopes
Export token: export DISCORD_TOKEN="..."
Remind user to add the export to their shell profile

Step 4: Boot and run

Once the chosen integrations are configured:

cd /path/to/openclaw-in-a-box
mb up
mb ssh openclaw-in-a-box
bash /workspace/scripts/install.sh   # first time
bash /workspace/scripts/start.sh

Report which skills loaded and which integrations are active.

Step 5: Hand off to sub-skills

Once the gateway is running, the user invokes skills directly:

/gmail-triage — triage unread inbox
/github-org-triage papercomputeco — scan an org
/discord-bot report — generate activity report

Each skill has its own rules in skills/. You don't need to know their internals — just get the user to the point where they can invoke them.

Rules

Never store secrets in files. Tokens go in env vars or system keychains.
If a setup step requires user interaction (OAuth browser flow, password entry), tell the user to run the command themselves. Don't try to automate password entry.
If a step fails, diagnose the error and suggest a fix. Don't retry blindly.
Only set up what the user asked for. Don't push integrations they didn't choose.