Skip to main content

What is a Squad?

A squad is a team of agents organized around a business domain. Each squad:
  • Owns a specific area of responsibility
  • Shares memory and context between agents
  • Produces outputs to a dedicated location

Squad Structure

.agents/squads/research/
├── SQUAD.md           # Squad definition
├── analyst.md         # Agent: research analyst
├── writer.md          # Agent: content writer
└── reviewer.md        # Agent: quality reviewer

SQUAD.md Definition

# Research Squad

## Mission
Transform raw intelligence into actionable insights.

## Agents
- analyst: Deep research and analysis
- writer: Content creation from research
- reviewer: Quality and accuracy checks

## Outputs
- research/ directory
- Weekly research reports
- Competitive analysis documents

## Memory
Shared context about ongoing research topics,
sources, and accumulated knowledge.

Example Squad Domains

Organize squads by business function, not technical capability:
DomainExample FocusExample Agents
ResearchMarket intelligenceanalyst, writer, reviewer
CustomerPipeline & relationshipslead-finder, outreach
EngineeringCode & infrastructurereviewer, doc-writer, deployer
ContentMarketing & communicationscopywriter, seo, social
OperationsInternal processesscheduler, reporter, auditor
Start with 2-3 squads. Add more as your needs grow. Each squad should have clear ownership of a business domain.

Running a Squad

Basic Execution

# Run entire squad
squads run research

# Run specific agent
squads run research/analyst

# Preview without executing
squads run research --dry-run

Continuous Improvement Loops

Squads can run in cycles that continuously improve your codebase:
┌─────────────────────────────────────────────────┐
│              Autonomous Loop                     │
├─────────────────────────────────────────────────┤
│  1. Audit      → Find issues, create tickets    │
│  2. Solve      → Branch, fix, create PR         │
│  3. Review     → Adversarial QA, test           │
│  4. Merge      → Auto-merge if passes           │
│  5. Deploy     → Release artifacts              │
│  6. Repeat     → New audit cycle                │
└─────────────────────────────────────────────────┘
This pattern enables:
  • Self-healing codebases — Issues found and fixed automatically
  • Continuous quality — Every change reviewed and tested
  • Human oversight — Escalate to Slack/email when uncertain
See GitHub Orchestration for implementation details.