Agent Architecture

Agent Architecture Designer
If Claude Code is the engine, Agent Architecture Designer is the dashboard - design the team, watch it work before you spend a single token, then export exactly what Claude Code needs to run it for real.

Looking for the visual designer? Open index.html directly in your browser, or use the live version on GitHub Pages - no install, no build step. Everything below also documents the Claude Code plugin (60 subagents + 62 orchestration presets) generated from that same app - see Claude Code integration.

Why this exists

Agent Architecture Designer is primarily an educational and developmental tool. It is not a production orchestrator - it is a place where you can slow down and study multi-agent systems the way you would study a complex machine: one moving part at a time.

The goal is that after using it, you should be able to:

- Understand what every single agent actually does - its role, inputs, outputs, anti-patterns, and failure modes
- Understand how agents talk to each other - who hands off to whom, which phases they live in, where the friction is
- Understand why a given preset looks the way it does - why it has 6 agents and not 12, why a Five Minds debate sits in the middle, why the HITL gate lives where it lives
- Recognize the architecture patterns behind the scenes - ReAct, Orchestrator-Worker, Blackboard, Routing, and seven others, with real examples pulled from the agents on the canvas
- Understand exactly when Claude Code will run your own hooks, across all 27 documented lifecycle events, before you write a single line of hook config
- Understand the cost and context budget of a multi-agent system before you ever spend a token

You can use it as a visual designer, but the real value is the Encyclopedia behind every agent and…