agent-system-architecture

What problem does it solve?

Designing reliable multi-agent systems is complex and error-prone: teams struggle with choosing the right orchestration pattern, defining safe state management, avoiding race conditions, and keeping observability and human-in-the-loop controls in place. This Skill captures proven patterns and hard rules so architects can produce correct, minimal-complexity designs that meet performance, safety, and maintainability constraints.

Core Features & Use Cases

• Orchestration pattern selection: Guidance for choosing sequential, parallel, hierarchical, handoff, or group-chat wiring based on task ambiguity and coordination costs.
• Wiring & State strategy: Recommendations for message-passing vs shared-blackboard, triggers, termination conditions, and partitioning outputs to avoid race conditions.
• Observability & HITL: Templates for supervisor/monitoring components that log token usage, latency, and success/failure rates, and explicit points for human approval on high-risk decisions.
• Use Case: Designing a parallel PR-review pipeline with specialist agents (quality, security, coverage), a lightweight aggregator, and a Mermaid wiring diagram plus saved architecture spec.

Quick Start

Design a parallel PR review agent system that runs quality, security, and coverage agents, outputs a Mermaid wiring diagram and an implementation-ready architecture spec.