Agent Team Topologies

A quick-reference model for structuring multi-agent teams in Claude Code.

8 composable topology patterns — a nod to Team Topologies thinking, applied to how work flows through agent teams. Browse the patterns, find what fits, and adapt.

Find Your Topology Getting Started Official Docs View on GitHub


Pick Your Topology

Goal Question Topology
Understand Multiple independent questions? Parallel Explorers
  Ambiguous bug? Competing Hypotheses
Build Multi-layer feature? Feature Pod
  Many small tasks? Task Queue
  Pure coordination? Orchestrator-Only
Review Multiple lenses? Review Board
Risky change Expensive to get wrong? Risky Refactor
Any of the above Need quality enforcement? + Quality-Gated overlay

Full decision tree →


The 8 Topologies

Pattern Best For Cost
Parallel Explorers Discovery, research, codebase mapping Low
Review Board Code review with distinct lenses Low
Competing Hypotheses Ambiguous bugs, architectural decisions Medium
Feature Pod Cross-layer feature delivery Medium
Risky Refactor High-risk changes needing plan approval Low
Orchestrator-Only Pure coordination, lead never codes High
Quality-Gated Enforcing completion standards (composable) Overlay
Task Queue Many small independent tasks High

Topologies are primitives, not monoliths. Any teammate slot can itself become a topology – a reviewer in Feature Pod can spawn a Review Board, an explorer can fan out sub-explorers. See Composing Topologies for recipes.


Guides

Document What’s Inside
Getting Started Enable agent teams, install configs, run your first topology
Mental Model Teams vs subagents, core concepts, selection heuristics
Decision Tree Expanded flowchart for picking the right topology
Composing Topologies Recipes for chaining, nesting, and combining patterns
Anti-Patterns 8 things NOT to do with agent teams
Cost Guide Token economics by topology, cost reduction strategies
Best Practices Operational guidance for running agent teams

Contributing

See Contributing for how to propose new topology patterns, submit real-world examples, and improve agent definitions or hooks.