Designs an agent's architecture — the tool surface, context strategy, control flow, and model choice — from a scoped set of jobs. Use this whenever the user is deciding how an agent should be built rather than what it should do — choosing tools, wiring up an agent loop, deciding single-agent vs multi-agent, picking a model, or planning what goes in the context window. Also use when an existing agent is unreliable and the cause looks structural rather than a prompt problem. For deciding what the agent is for, use agent-scoping first.
