Multi Agent Orchestration skills for AI agents

Coordinate multiple specialized agents on a single task — when to hand off, what context to transfer, and how to avoid the failure modes specific to multi-agent systems. Covers supervisor/worker patterns, peer collaboration, and the blackboard model. Use when building systems where one LLM agent isn't enough.

Design the tools that an LLM agent uses. Covers naming, parameter design, return formats, error handling, and the difference between tools that LLMs use successfully and ones they consistently misuse. Use when building or refining the tool set for any agent.

Build the eval suite that tells you whether changes to your agent improve it or regress it. Covers eval-set construction, the metrics beyond pass/fail, the cost of running evals, and the practice of treating evals as the source of truth for agent behavior. Use when building any agent system you intend to ship and iterate on.

Use LangGraph (or equivalent state-machine frameworks) to express agent workflows as explicit graphs of nodes and edges. Covers the state-machine model, conditional edges, persistence, and the patterns that distinguish well-structured agent workflows from spaghetti Python with control-flow scattered everywhere.