Multi-Agent Handoff Protocol Designer

Designs orchestrator + worker handoff protocols for production multi-agent LLM systems — triage routing, agent role boundaries, shared state schema, loop-prevention rules, escalation paths, failure recovery, and a verification harness. Turns 'we want a team of agents' into a reviewable architecture spec mapped to LangGraph, CrewAI, AutoGen, OpenAI Agents SDK, or the Claude Agent SDK — designed for the 2026 multi-agent wave where infinite-handoff loops are the #1 production failure mode