dsh-subagent-model-picker

Routing subagents to a cheaper model has been solved many ways in recent days, yet every existing approach shares one trait: it decides for you. Tool-based routing lets the model pick, auto-routing leans on policy classification, and command-style input offers no visible state at all. dsh-subagent-model-picker takes a different path. It places a dropdown right beside the main model seat so you choose visually, and every selection applies deterministically to all subagent paths in that session—no guessing involved.

Under the hood it covers subagent, subagent_fork, and workflow fan-out children at any depth. Models that expose reasoning-effort levels get a matching Thinking effort menu alongside the model picker. The choice is per-session state that survives a restart, and the default behavior is simply to inherit the main model, with no hidden routing. It composes cleanly with auto-routing plugins: keep a policy router as the default, then override a specific session manually—or do the reverse.

This is for developers who want precise control over subagent inference cost without delegating the routing decision to a model or a policy rule. If you want to know, at a glance, exactly which model every subagent is running on before you hit send, and keep that state visible after the fact, this is the tool you have been looking for.