multi-agent-composition

Multi-agent composition uses a decision framework to select the right component for each task. Skills handle focused domains, Sub-Agents manage independent reasoning threads, MCP Servers provide external integrations, and Slash Commands trigger quick actions. The Core 4 Framework guides you through architectural tradeoffs and integration paths.

Scout-plan-build is a transformation pattern for orchestrating Claude Code agents across multiple stages. Scout gathers information, plan structures decisions, and build executes solutions. This pattern manages context flow and coordinates component interactions in complex multi-agent workflows.

Build scalable multi-agent systems by applying the decision framework to compose Skills, Sub-Agents, MCP Servers, and Slash Commands. Use orchestrator workflows with lifecycle management and observability hooks to coordinate interactions, progressive disclosure to manage complexity, and clear integration paths for each component type.

Multi-agent anti-patterns include poor context management across agents, incorrect component selection, and inadequate observability. Real-world case studies in this Skill reveal common mistakes in orchestration design, context leakage between Sub-Agents, and misaligned lifecycle management that lead to scaling failures.

MCP Servers are one of four component types in Claude Code composition. They integrate external tools and services into multi-agent systems. Whether you need them depends on your architecture—the decision framework helps you evaluate MCP Servers against Skills, Sub-Agents, and Slash Commands for your specific use case.

Observability hooks provide lifecycle visibility into multi-agent systems, enabling you to monitor component interactions, track context flow, and debug orchestration issues. They support progressive disclosure and technical requirement satisfaction by exposing state changes and integration points across your composed agents.