Support strong/medium/weak model configuration for workflow agents

[Michaelliv]

Right now the workflow/subagent model selection is likely too coarse. For generated workflows, we probably want a way to configure tiers like:

• BIG_MODEL / strong model for planning, architecture decisions, hard reasoning
• MEDIUM_MODEL for normal edits and synthesis
• SMALL_MODEL / weak model for cheap repetitive operations, classification, scanning, formatting, etc.

The main motivation is cost/control: users may not want an expensive frontier model updating thousands of files when a cheaper model can handle many workflow steps.

Open questions:

• Should model tiers be global config, workflow-level config, or both?
• Should the JS workflow DSL expose model selection per step/subagent?
• Should defaults map to strong/medium/small aliases rather than concrete provider model names?
• Do we want aider-compatible naming like strong_model / weak_model, or our existing BIG_MODEL / MEDIUM_MODEL / SMALL_MODEL style?

Potential API shape:

await agent.task("scan files", { model: "small" })
await agent.task("design migration", { model: "big" })
await agent.task("apply changes", { model: "medium" })

This should make dynamic workflows more practical for large repos and long-running subagent tasks.

[joshuacc]

I like the idea of having this as an option that I can set defaults on, but I also need to be able to set arbitrary models on specific agents. I have a branch over here that allows this and it's been working great for me.

Because of that need, I'd suggest that "small", "big", and "medium" be set via a modelTier or modelClass config property instead of the model property.