Multi-Agent Coordination Architecture — Production Orchestration Patterns

Multi-agent coordination is the most talked-about capability in AI infrastructure for 2026 — and the most underdocumented. Most operators running multi-agent setups discover the hard way that agents step on each other's context, duplicate work, lose state at handoff points, and fail silently when a subagent crashes mid-task.

This SKILL.md contains production-validated coordination architecture covering: role assignment without overlap, task handoff state capture and acknowledgment chains, shared vs. isolated context rules, subagent failure recovery and coordinator takeover patterns, and three orchestration models (sequential, parallel, hybrid) with cost trade-offs for each.