MetaClaw Setup Architect

A MetaClaw-branded OpenClaw skill that generates complete multi-agent setups from natural language descriptions. Describe what you want, and the MetaClaw Setup Architect produces every configuration file — agents, skills, workflows, memory, and installation steps — zero effort.

---

What Is This?

The MetaClaw Setup Architect is an OpenClaw skill that acts as a setup factory. Instead of manually configuring agents, writing SKILL.md files, and wiring up MCP servers, you describe your use case in plain English and the skill:

1. Discovers your requirements through 4-6 targeted questions
2. Designs a multi-agent architecture with the right tools and patterns
3. Generates every file: AGENTS.md, SOUL.md, MEMORY.md, skills, workflows, openclaw.json
4. Installs the configuration with copy-pasteable commands

Who This Is For

• Founder-operators: go from idea to a multi-agent operating setup in one guided workflow.
• Technical founders shipping fast: turn rough product intent into structured agent architecture and execution files.
• Early-stage founder teams: standardize AI operations without custom setup chaos.

Quick Start

Install the Skill

# Install directly from GitHub repo
npx skills add mverab/metaclaw --skill metaclaw-setup-architect

# Optional: verify what the repo exposes
npx skills add mverab/metaclaw --list

# Local development (before pushing)
npx skills add /absolute/path/to/metaclaw --list

# Install from local repo copy
npx skills add /absolute/path/to/metaclaw --skill metaclaw-setup-architect

Use It

Send your OpenClaw agent a message like:

"I want to create a multi-agent setup for YouTube clipping viral short-form content from long-form videos. Consider the entire pipeline, tools, editing, and publishing."

Or:

"Help me create an OpenClaw setup for a community assistant that helps Skool community owners manage questions and find relevant lessons."

MetaClaw Setup Architect takes over from there.

---

Project Structure

skills/metaclaw-setup-architect/
├── SKILL.md                          ← Main skill (the orchestrator)
│
├── knowledge/                        ← Reference knowledge base
│   ├── file-system.md                ← Complete OpenClaw file system reference
│   ├── tool-catalog.md               ← 80+ tools/MCPs categorized by domain
│   ├── agent-patterns.md             ← 5 multi-agent architecture patterns
│   └── skill-templates.md            ← Domain-specific SKILL.md templates
│
├── templates/                        ← File generation templates
│   ├── AGENTS.md.tmpl                ← Agent operating instructions
│   ├── SOUL.md.tmpl                  ← Agent personality
│   ├── USER.md.tmpl                  ← Human profile
│   ├── MEMORY.md.tmpl                ← Semantic memory
│   ├── TOOLS.md.tmpl                 ← Environment config
│   ├── HEARTBEAT.md.tmpl             ← Periodic schedule
│   ├── IDENTITY.md.tmpl              ← Agent identity card
│   ├── BOOTSTRAP.md.tmpl             ← First-run setup
│   ├── SKILL.md.tmpl                 ← Custom skill template
│   ├── workflow-AGENT.md.tmpl        ← Workflow algorithm
│   ├── openclaw.json.tmpl            ← Gateway configuration
│   └── MOC.md.tmpl                   ← Map of Content
│
└── examples/                         ← Complete end-to-end examples
    ├── youtube-clipper.md             ← 3-agent YouTube clipping pipeline
    └── community-assistant.md         ← 2-agent Skool community manager

Knowledge Base

File	What It Contains
file-system.md	Every file in OpenClaw's architecture — paths, purposes, formats, loading behavior
tool-catalog.md	80+ tools organized in 12 categories: Web, Filesystem, Video, Comms, AI Models, Data, Productivity, Marketing, Commerce, DevOps, Health, Community
agent-patterns.md	5 architecture patterns: Pipeline, Hub-and-Spoke, Autonomous Loop, Human-in-the-Loop, Supervisor
skill-templates.md	Ready-to-customize SKILL.md templates for 7 common domains

Multi-Agent Patterns

Pattern	Best For	Example
Pipeline	Sequential processing (A → B → C)	Content scrape → write → publish
Hub-and-Spoke	Coordinator + specialists	Lead dev + frontend + backend + QA
Autonomous Loop	Self-running recurring tasks	SEO monitor that runs 24/7
Human-in-the-Loop	Agent proposes, human approves	Content review before publishing
Supervisor	Quality control over other agents	QA agent that reviews all outputs

Examples

YouTube Viral Clipper (3 agents)

• Scout: Monitors channels, scores viral potential
• Editor: Downloads, transcribes, identifies clips, cuts vertical video
• Publisher: Human approval queue → publish to Shorts/TikTok/Reels

Skool Community Assistant (2 agents)

• Community Brain: Answers questions, finds lessons, builds knowledge base
• Engagement Ops: Welcomes members, tracks metrics, weekly reports to owner

---

Docker Testing

Spin up an isolated OpenClaw instance with the skill mounted — no local install required.

Prerequisites

• Docker Desktop (or Docker Engine + Compose v2)
• At least 2 GB RAM

Quick Start

# 1. Copy env file and add your API key
cp .env.example .env
# edit .env — add ANTHROPIC_API_KEY (minimum required)

# 2. Start gateway
make up

    # 3. Enforce low-cost model policy (no Opus)
make enforce-model

# 4. Open Control UI
open http://127.0.0.1:18789

# 5. Run smoke tests
make test

What the Smoke Test Checks

1. Gateway health — GET /healthz responds 200
2. Skill mounted — SKILL.md exists inside the container
3. Knowledge base — all 4 knowledge files present
4. Templates — all 11 template files present
5. Examples — both example files present
6. Model policy enforced — Sonnet primary + Haiku fallback, Opus excluded
7. Skill discovered — appears in openclaw skills list
8. Agent trigger — agent correctly activates Discovery phase on a test prompt

Useful Commands

make logs    # tail gateway logs
make shell   # bash session inside the container
make enforce-model  # set Sonnet+Haiku policy, excluding Opus
make down    # stop containers (data preserved)
make clean   # hard reset — wipes the volume
make pull    # update to latest OpenClaw image

How It Works

The docker-compose.yml uses the official ghcr.io/openclaw/openclaw:latest image (no local build) and bind-mounts the skill as read-only into the container's shared skills directory:

./skills/metaclaw-setup-architect  →  /home/node/.openclaw/skills/metaclaw-setup-architect (ro)

The test workspace (./test/workspace/) is also mounted so the test agent's AGENTS.md and SOUL.md are available without any manual configuration.

---

Extending

Add New Tool Categories

Edit knowledge/tool-catalog.md to add new tools or MCP servers.

Add New Architecture Patterns

Edit knowledge/agent-patterns.md to add new multi-agent patterns.

Add New Skill Templates

Edit knowledge/skill-templates.md to add domain-specific SKILL.md templates.

Add New Examples

Create new files in examples/ following the existing format.

---

License

MIT