orchestration-patterns

Orchestration patterns let you build specialized subagents that manage other skills in isolation. Single-skill consumers focus on one task; multi-skill orchestrators coordinate complex workflows across domains—each agent maintains clean context by delegating rather than holding all logic.

A single-skill consumer creates a focused agent for one task with isolated context. A multi-skill orchestrator coordinates multiple skills across domains, using decision matrices and hierarchical delegation to prevent context bloat while handling complex workflows like full-stack feature development.

Yes. YAML-based skill declarations specify functional requirements for agent-to-skill coordination, enabling auto-loading, minimal tool footprints, and structured delegation patterns. This approach works for isolated subagents, decision matrices, and hierarchical orchestration across database, deployment, and workflow domains.

Use hierarchical orchestration when workflows span multiple domains or require deep context isolation. A top-level orchestrator delegates to specialized sub-agents for maximum focus—ideal for complex tasks like building full-stack features with frontend design, API generation, and database migration skills.

Minimal tool footprints prevent context window pollution but require careful skill design and clear responsibility boundaries. Coordination overhead increases with agent count; best practices include YAML declarations, context isolation specs, and pre-built templates to manage load and prevent coordination bottlenecks.

Orchestration patterns isolate context by separating concerns into focused subagents rather than loading all skills into one agent. Decision matrices route tasks to appropriate skill consumers; hierarchical delegation ensures each agent operates within minimal scope, preserving context for focused execution.