Agent Coordination

This handles spawning and coordinating multiple Claude agents for complex tasks that need different specialized roles. You get 60+ agent types ranging from basic coders and reviewers to niche ones like Byzantine coordinators and SPARC pattern specialists. The routing codes are clever, mapping task complexity numbers to predefined agent teams. It's overkill for single-agent work, but if you're orchestrating something like a full feature build with architecture, coding, security audits, and testing all happening in parallel, this gives you the scaffolding. Keep teams under 8 agents for tight coordination, and agents communicate through shared memory rather than direct messaging.