Pick the right model for each delegated task

Not every task needs the largest model. A simple lookup can go to a smaller model. A hard design question may need the largest one. Choosing the model is a decision, so Rein writes it as a flow.

The first question asks whether the task is simple. A simple task goes to a subagent with a smaller model. Other tasks ask one more question about how much reasoning they need.

Delegation uses a template. A template is text with slots, such as the model and the task. The flow fills the slots, and the agent sends the result to a subagent.

Each model choice has a case. Add a new option, and the checker reports every place that still needs a case.