Agent Squad

Agent Squad

Orchestrate teams of specialized AI agents that collaborate on tasks via shared workspaces and heartbeat scheduling.

Quick Start

# Initialize your squad
agent-squad init my-squad

# Add agents
agent-squad add researcher --role "Deep researcher. Finds sources and data."
agent-squad add writer --role "Content writer. Sharp, opinionated voice."
agent-squad add lead --role "Squad lead. Coordinates and reports."

# Configure task system
agent-squad config --tasks linear --project "Content Squad"

# Start heartbeats (agents wake every 15 min)
agent-squad start

What This Enables

Before: One AI doing everything. Context bloat. Generic output.

After: Specialized agents with distinct roles, shared task board, coordinated via heartbeats. Each agent:

• Has a unique personality (SOUL.md)
• Checks for work every 15 minutes (cron)
• Posts updates to shared task system
• @mentions other agents when handoffs needed

Core Concepts

Agents = OpenClaw Sessions

Each agent is an isolated OpenClaw session with:

• Unique session key (agent:{name}:main)
• Custom SOUL.md (personality, role, tools)
• Heartbeat cron (wakes periodically)
• Access to shared task system

The Heartbeat Pattern

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

:00  Lead checks Linear → delegates tasks
:05  Researcher checks → does research, @mentions Writer  
:10  Writer checks → reads research, writes draft

Agents check for @mentions, assigned tasks, and activity. If nothing to do → HEARTBEAT_OK and sleep.

Shared Task Systems

Agents coordinate via task system comments:

• Linear (recommended) — Issues, comments, states
• Trello — Cards, lists, checklists
• GitHub Issues — Native integration
• Files — Markdown in git repo

Commands

Squad Management

agent-squad init <name>              # Create new squad workspace
agent-squad status                   # Show all agents and health
agent-squad logs <agent>             # View recent heartbeats

Agent Management

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

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

Task System Integration

agent-squad config --tasks linear    # Use Linear
agent-squad config --tasks trello    # Use Trello
agent-squad config --tasks github    # Use GitHub Issues
agent-squad config --tasks file      # Use local markdown files

Operations

agent-squad start                    # Start all heartbeats
agent-squad stop                     # Stop all heartbeats
agent-squad trigger <agent>          # Manual heartbeat trigger
agent-squad notify "@writer draft ready"  # Send notification

Example: Content Marketing Squad

3 agents, Linear integration:

# Setup
agent-squad init content-squad
agent-squad add fury --role "Researcher. Every claim needs a source."
agent-squad add loki --role "Writer. Sharp, anti-fluff voice."
agent-squad add jarvis --role "Lead. Coordinates, reports to human."

agent-squad config --tasks linear --project "Content Squad"
agent-squad start

Workflow:

1. You create Linear issue: "Blog post: AI Agent Security"
2. Jarvis (lead) assigns to Fury (researcher)
3. Fury heartbeat → researches → posts findings → @mentions Loki
4. Loki heartbeat → writes draft → posts → @mentions Jarvis
5. Jarvis notifies you via Telegram: "Draft ready for review"

SOUL.md Templates

See references/personalities.md for pre-built agent personalities:

• Researcher (Fury)
• Writer (Loki)
• Lead (Jarvis)
• SEO Analyst (Vision)
• Social Media (Quill)
• Developer (Friday)

Advanced

Custom Heartbeat Schedules

Stagger agents to prevent all waking at once:

agent-squad add lead --schedule "0,15,30,45 * * * *"
agent-squad add researcher --schedule "5,20,35,50 * * * *"
agent-squad add writer --schedule "10,25,40,55 * * * *"

AGENTS.md (Shared Operations Manual)

Each agent workspace includes AGENTS.md with:

• Squad mission
• Communication rules (@mentions, handoffs)
• File conventions
• Escalation procedures

See references/agents-template.md

Multi-Model Squads

Use cheaper models for heartbeats, premium for creative work:

agent-squad add researcher --model "kimi-code"        # Cheap, fast
agent-squad add writer --model "claude-opus-4-5"      # Premium creative

Architecture

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

Requirements

• OpenClaw gateway running
• Task system credentials (Linear API key, Trello token, etc.)
• Node.js (for CLI)

See Also

• references/personalities.md — Agent personality templates
• references/agents-template.md — AGENTS.md template
• references/workflows.md — Common squad workflows