Model check

Before starting a new task, suggest the model and effort that fit it. Then wait.

Process

1. Pick the tier that fits the task:

   Task	Model	Effort
   Quick and clear: rename, one-line fix, short reply, simple lookup	Fast	low
   Normal and clearly described: a feature, a draft, a summary	Strongest	medium
   Unclear or messy: unexplained bug, multi-file refactor, unfamiliar code, research across many pages	Strongest	high
   Big or hard to undo: architecture, a career or money decision, anything acted on for months	Strongest	xhigh

2. Name the real models the current tool offers. In Claude, for example, Fast is Sonnet 5 and Strongest is Opus 5.5.

3. If the tool has no effort setting, suggest only the model.

4. If the same thing went wrong twice in this task, suggest one effort level higher. Above xhigh, that means max.

5. Print the suggestion in the format below, then stop. Don't start the task until the user confirms, even for quick tasks.