Multi-Agent System Architecture

概要

Provides a comprehensive architectural framework for designing resilient and scalable multi-agent AI systems. It guides developers through a robust four-layer agent stack—separating high-level reasoning from deterministic execution—and details seven core coordination patterns ranging from simple parallel fan-out to the high-reliability MAKER pattern for million-step tasks. By enforcing strict schema-first tool design, event-sourced state management, and permission inheritance, this skill ensures that complex agentic interactions remain observable, testable, and production-ready across any programming environment.

主な機能

• 3 GitHub stars
• Deterministic tool boundary and schema-first implementation guidelines
• State management through event-sourcing and hierarchical delegation tracking
• Four-layer agent architecture (Reasoning, Orchestration, Tool Bus, Adapters)
• Advanced coordination patterns including Fan-out, Recursive Delegation, and Map-Reduce
• MAKER pattern for high-reliability, massive-scale task execution

ユースケース

• Implementing production-ready AI workflows with strict permission and resource coordination
• Building complex autonomous agent pipelines with zero-error tolerance
• Architecting specialized sub-agent hierarchies for large-scale software engineering tasks