Ai Agent Design

This is the skill you reach for when you're building agents that actually loop, call tools, and iterate toward a goal. It covers the core loop patterns (ReAct, plan-and-execute, reflection), tool schema design, multi-agent orchestration topologies, and memory strategies from working context to vector stores. The principles are sharp: tools over knowledge, constrain scope, fail gracefully with max iterations, and evaluate trajectory quality not just final answers. You get working TypeScript examples for ReAct loops and tool schemas with Zod validation. If you're debugging an agent that hallucinates tools or loops forever, or building a multi-agent system and need to pick between sequential, parallel, or hierarchical topology, this is the reference.