🤖 Agent Squad

Multi-agent orchestration for OpenClaw. Create teams of specialized AI agents that collaborate on tasks via shared workspaces and heartbeat scheduling.

"One AI doing everything is a bottleneck. Ten specialized agents working together is a force multiplier."

What This Does

Before: One AI assistant handles all your tasks. Context gets bloated. Output is generic.

After: Specialized agents with distinct roles collaborate through a shared task system:

• Fury (Researcher) — Digs deep, finds sources, verifies facts
• Loki (Writer) — Crafts sharp, opinionated content
• Vision (SEO) — Optimizes for search and intent
• Jarvis (Lead) — Coordinates, delegates, reports to you

Each agent wakes up every 15 minutes, checks for assigned work, collaborates via comments, then goes back to sleep.

Quick Start

# Install globally
npm install -g agent-squad

# Create your squad
agent-squad init content-squad

# Add specialized agents
agent-squad add fury --role "Researcher" --personality skeptical
agent-squad add loki --role "Content Writer" --personality creative
agent-squad add jarvis --role "Squad Lead"

# Connect to your task system
agent-squad config --tasks linear --project "Content Squad"

# Start the heartbeats
agent-squad start

That's it. Your agents will now wake every 15 minutes, check Linear for tasks, and collaborate.

How It Works

The Heartbeat Pattern

Instead of always-on agents (expensive), agents wake on schedule:

:00  Jarvis checks Linear → delegates tasks
:05  Fury checks → researches, posts findings
:10  Loki checks → writes draft, @mentions Jarvis

Agents check for:

• Tasks assigned to them
• @mentions in comments
• Activity on threads they're subscribed to

If nothing to do → HEARTBEAT_OK and sleep until next cycle.

Shared Task Systems

Agents coordinate via your existing tools:

| System | Use Case | |--------|----------| | Linear | Software teams, issue tracking | | Trello | Visual boards, simpler workflows | | GitHub Issues | Code projects, developer-focused | | Files | Markdown in git repo, simple setup |

Agent Personalities (SOUL.md)

Each agent has a SOUL.md file defining:

• Role — What they're responsible for
• Personality — How they approach work
• Process — Step-by-step how they work
• Tools — Which skills they use
• Communication style — How they write

Example:

# SOUL.md — Who You Are

**Name:** Loki
**Role:** Content Writer

## Personality
Sharp, slightly cynical. Hates corporate speak. 
Every sentence must earn its place.

## What You're Good At
- Hooks that grab attention
- Cutting 20% of words without losing meaning
- Writing that sounds human

## What You Care About
- No "In today's world..." openings
- Oxford commas (pro), passive voice (anti)
- Specific examples over abstract advice

Real Workflow Example

You want a blog post on AI agent security.

Hour 0: You create a Linear issue: "Blog: AI Agent Security"

Hour 0: Jarvis (lead) sees the issue, assigns to Fury (researcher)

Hour 1: Fury wakes up, researches:

• Finds primary sources on agent vulnerabilities
• Extracts statistics and expert quotes
• Posts findings: "Research complete @loki"

Hour 2: Loki wakes up, sees @mention:

• Reads Fury's research
• Writes 1,000 word draft
• Posts: "Draft ready @jarvis"

Hour 3: Jarvis wakes up, sees @mention:

• Reviews the draft
• Sends you a Telegram: "📝 Draft ready for review: [link]"

You: Edit in GitHub, approve with comment

Jarvis: Moves issue to "Done"

All while you were in meetings.

Pre-Built Agent Personalities

Fury — The Researcher

Obsessive fact-checker. Every claim needs a source. Finds primary sources, extracts data, spots misinformation.

Loki — The Writer

Sharp, cynical voice. Hates fluff. Pro-hooks, anti-passive-voice. Writes like a smart human, not a brand.

Vision — The SEO Analyst

Thinks in keywords and search intent. Optimizes content before and after writing.

Jarvis — The Squad Lead

Coordinator. Delegates, tracks progress, knows when to escalate. Your interface to the squad.

Friday — The Developer

Clean code, tests, documentation. Precise and systematic.

Quill — Social Media Manager

Hook-obsessed. Thinks in threads and viral potential. Platform-native content.

Commands

# Squad management
agent-squad init <name>              # Create new squad
agent-squad status                   # Show all agents and health

# Agent management
agent-squad add <name> [options]     # Add agent
  --role "Description"
  --personality skeptical|creative|analytical|balanced
  --schedule "*/15 * * * *"
  --model "kimi-code"

agent-squad edit <name>              # Edit SOUL.md
agent-squad remove <name>            # Remove agent
agent-squad disable <name>           # Pause heartbeats
agent-squad enable <name>            # Resume heartbeats

# Task system
agent-squad config --tasks linear    # Use Linear
agent-squad config --tasks trello    # Use Trello
agent-squad config --tasks github    # Use GitHub Issues

# Operations
agent-squad start                    # Start all heartbeats
agent-squad stop                     # Stop all heartbeats

Configuration

Linear Integration

# Make sure linear skill is installed and configured
linear auth --token YOUR_TOKEN

# Configure squad to use Linear
agent-squad config --tasks linear --project "Content Squad"

Trello Integration

# Make sure trello skill is installed
trello auth --api-key KEY --token TOKEN

# Configure squad
agent-squad config --tasks trello --board "Content Board"

Architecture

┌─────────────────────────────────────────────┐
│           OpenClaw Gateway                  │
│  ┌─────────┐ ┌─────────┐ ┌─────────┐       │
│  │  Lead   │ │Researcher│ │ Writer │       │
│  │ Session │ │ Session  │ │Session │       │
│  │ (Jarvis)│ │ (Fury)   │ │(Loki)  │       │
│  └────┬────┘ └────┬────┘ └────┬────┘       │
│       │           │           │             │
│       └───────────┼───────────┘             │
│                   │                         │
│              ┌────┴────┐                    │
│              │  Cron   │                    │
│              │ Jobs    │                    │
│              └────┬────┘                    │
└───────────────────┼─────────────────────────┘
                    │
            ┌───────┴───────┐
            │  Task System  │
            │ (Linear/etc)  │
            └───────────────┘

Why This Approach?

Specialized > General

An agent who's "good at everything" is mediocre at everything. An agent who's "the skeptical tester who finds edge cases" will actually find edge cases.

Async > Real-time

Always-on agents burn API credits doing nothing. Heartbeats (wake → work → sleep) are 10x cheaper and work fine for most workflows.

Shared Context > Isolated

When agents share a task board, they see the full context: who's working on what, what was decided, what blocks progress.

Inspired By

This implementation is inspired by Bhanu Teja's Mission Control — a 10-agent squad system built on OpenClaw.

Requirements

• OpenClaw installed and running
• Node.js 18+
• Task system credentials (Linear, Trello, or GitHub)

License

MIT

Contributing

This is early software. PRs welcome:

• New agent personalities
• Additional task system integrations
• Workflow templates
• Bug fixes

Built with 💜 by Tony Stark & Orion for workeragi.