Building Multiagent Systems

Building multi-agent systems

Architecture patterns for multi-agent systems where AI agents coordinate using tools.

Installation

What this plugin does

Guides you through designing and implementing multi-agent systems -- coordination, lifecycle management, production hardening. Based on patterns from real production systems.

Concepts covered

- Four-layer architecture -- reasoning, orchestration, tool bus, deterministic adapters
- Schema-first tools -- typed contracts that let sub-agents discover and validate tools
- Deterministic boundary -- separating LLM reasoning from testable execution
- Seven coordination patterns -- fan-out/fan-in, sequential pipeline, recursive delegation, work-stealing queue, map-reduce, peer collaboration, MAKER (million-step zero-error)
- Foundational patterns -- event-sourcing, hierarchical IDs, agent state machines
- Tool coordination -- permission inheritance, locking, rate limiting, caching
- Agent collaboration -- subagent spawning, tool inheritance, sub-agent as tool
- Lifecycle management -- cascading stop, orphan detection, heartbeat monitoring
- Self-modification safety -- when and how sub-agents can modify themselves or each other
- Production hardening -- checkpointing, monitoring, cost tracking across agent hierarchies

When to use

- Designing systems where multiple AI agents coordinate
- Implementing orchestrators that spawn sub-agents
- Building parallel or sequential agent workflows
- Coordinating shared resources across agents
- Managing agent lifecycle and state

Quick example

Discovery questions

Before architecting, the skill asks:

1. Starting point -- greenfield, adding to existing, or fixing current?
2. Primary use case -- parallel work, pipeline, delegation, collaboration, queue?
3. Scale expectations -- small…