What do you supply when creating a prompt agent?

Why this is the answer

Prompt agents are defined entirely through configuration — instructions, a model and tools — and Foundry hosts and runs them, so there is no application code, container or scaling to maintain. Supplying a container image describes a hosted agent, the other agent type. A fine-tuned model and evaluation set belong to a customisation workflow, not to defining an agent. Writing an explicit orchestration graph is the kind of custom logic you would move to a hosted agent for.