Agent Orchestrator for 🦞OpenClaw Skill

For 🦞OpenClaw Clawdbot Users (Recommended)

Let 🦞OpenClaw Molty set this up for you. Run this prompt in your OpenClaw chat:

Clone https://github.com/NMC-Interactive/agent-orchestrator and link the skill for me

OpenClaw will:

Clone the repository to your workspace
Link the skill to ~/.openclaw/skills/agent-orchestrator
Confirm the dashboard is ready to use

Once linked, you can initiate the skill via /command, then control everything via chat:

/ao start                    # Launch dashboard
/ao spawn builder codex      # Start implementation // truly flexible via prompt
/ao spawn reviewer claude    # Parallel review
/ao status                   # Check what's running

Quick Start for Developers

If you just want the dashboard binary:

# Download and install (macOS/Linux)
curl -L https://github.com/NMC-Interactive/agent-orchestrator/releases/latest/download/ao-dashboard-$(uname -s | tr '[:upper:]' '[:lower:]')-$(uname -m) -o ao-dashboard
chmod +x ao-dashboard
sudo mv ao-dashboard /usr/local/bin/

# Or use Go
go install github.com/NMC-Interactive/agent-orchestrator/cmd/ao-dashboard@latest

What is Agent Orchestrator?

It’s two things that work together:

Dashboard — A terminal UI showing all your agents at once
Skill — Chat commands (/ao) that control the dashboard through OpenClaw

You can use the dashboard alone. But with OpenClaw, you get chat-based control—spawn agents, check status, retry failures, all without leaving your conversation.

Using the Skill

Once installed (via OpenClaw or manual), these commands work in any OpenClaw session:

Command	What It Does
`/ao start`	Launch the dashboard in daemon mode
`/ao spawn <name> <agent>`	Start a new agent session
`/ao status`	Show running/completed/failed counts
`/ao kill <session>`	Stop a running session
`/ao retry <session>`	Retry a failed session
`/ao clear-failed`	Clean up failed sessions

Example Workflow

You: /ao start
Clawdbot: Dashboard running. 0 active sessions.

You: /ao spawn architect claude
Clawdbot: Spawned architect (claude). Status: running.

You: Design the API for a user service
[Claude works in background]

You: /ao spawn builder codex
Clawdbot: Spawned builder (codex). Status: running.

You: /ao status
Clawdbot: Running: 2 | architect (claude) - 8 turns | builder (codex) - 3 turns

You: /ao spawn reviewer claude
Clawdbot: Spawned reviewer (claude). Waiting for builder...

The reviewer waits automatically. When builder finishes, reviewer starts with fresh context.

When to Use Which Agent

Task	Agent	Why
Architecture & design	Claude	Thinks through trade-offs, asks good questions
Implementation & refactoring	Codex	Fast, follows patterns, doesn’t overthink
Legacy code analysis	Kimi	1M token context swallows entire repos
Security review	Claude (fresh)	Second instance, no shared bias
Experiments	OpenCode	Hackable, inspectable, yours

The Dashboard (Terminal UI)

The skill controls a terminal dashboard that looks like this:

┌─ Agent Orchestrator ─────────────────────────┐
│ Running: 3 | Completed: 12 | Failed: 0       │
├──────────────────────────────────────────────┤
│ architect   │ claude  │ running │ 8 turns    │
│ builder     │ codex   │ running │ 12 turns   │
│ reviewer    │ claude  │ waiting │ 0 turns    │
└──────────────────────────────────────────────┘

Keys: (r)efresh  (:)command  (T)retry  (C)clear  (q)uit

Key features:

Static UI — Press r to refresh. No distracting auto-updates.
Activity log — See what changed without noise.
Auto-compact — Rotates context at 10 turns to prevent drift.
Cross-platform — macOS, Linux, Windows, WSL.

Installing the Dashboard (Manual)

Dashboard is designed to increase the process controllability. It is yet to mature at this stage. Use a separate terminal to start your dashboard. If you didn’t use OpenClaw to set this up:

Option 1: Download Binary

# macOS
curl -L https://github.com/NMC-Interactive/agent-orchestrator/releases/latest/download/ao-dashboard-darwin-amd64 -o ao-dashboard
chmod +x ao-dashboard
sudo mv ao-dashboard /usr/local/bin/

# Linux
curl -L https://github.com/NMC-Interactive/agent-orchestrator/releases/latest/download/ao-dashboard-linux-amd64 -o ao-dashboard
chmod +x ao-dashboard
sudo mv ao-dashboard /usr/local/bin/

# Windows
# Download from: https://github.com/NMC-Interactive/agent-orchestrator/releases/latest

Option 2: Go Install

go install github.com/NMC-Interactive/agent-orchestrator/cmd/ao-dashboard@latest

Option 3: Build from Source

git clone https://github.com/NMC-Interactive/agent-orchestrator.git
cd agent-orchestrator
go build -o bin/ao-dashboard ./cmd/ao-dashboard
sudo mv bin/ao-dashboard /usr/local/bin/

Using the Dashboard (Standalone)

Without OpenClaw, you run the dashboard directly:

# First-time setup
ao-dashboard -init

# Start daemon
ao-dashboard -daemon

# In another terminal, use the helper script to spawn agents
./scripts/ao-cmd.sh spawn architect claude
./scripts/ao-cmd.sh spawn builder codex

Dashboard keys:

r — Refresh status
: — Command mode (type kill architect or retry builder)
C — Clear all sessions
T — Retry all failed
? — Help
q — Quit

What’s in the Repository?

agent-orchestrator/
├── cmd/ao-dashboard/        # Dashboard binary source
├── skills/
│   └── agent-orchestrator/
│       └── SKILL.md         # This skill definition
├── scripts/
│   ├── install.sh           # Unix installer
│   └── install.ps1          # Windows installer
└── README.md                # Full documentation

Two parts:

cmd/ao-dashboard/ — The Go binary (dashboard)
skills/agent-orchestrator/ — The OpenClaw skill (/ao commands)

You can use either independently. Together, they give you chat-based control of your agent swarm.

Requirements

Dashboard: macOS, Linux, Windows, or WSL. No Docker.
Skill: OpenClaw (Clawdbot) installed and running.
Optional: Go 1.21+ (if building from source).

Next Steps

OpenClaw users: Run /clone https://github.com/NMC-Interactive/agent-orchestrator and link the skill for me
Developers: Download binary or go install
Read more: What is Agent Orchestrator?

License: MIT
Repository: github.com/NMC-Interactive/agent-orchestrator
Questions? Open an issue on GitHub or ask in OpenClaw chat.