[Feature Request] Track and label advisor escalations for model training dataset improvement

Bug Description
Advisor sounds cool on paper; My concern is still the aspect of token usage. I'm a Pro user on the $20 a month plan using Claude Code in my terminal, and the problem I'm worrying about with Advisor is that it might end up defaulting to it in moments when it could've taken a better angle during a situation. Something I think needs to be accounted for if you guys are collecting training data is marking moments when the advisor needs to be called by the weaker model, as those are potentially key and prime areas where the weaker model is faltering. Just a loose estimate with my inexperience with code and how models work, but the reasoning is solid; typically workers go to their boss to consult them on an issue when they've exhausted their current means of handling the problem to their limits of tools and mental reasoning. Sonnet needing to contact Opus in this same dynamic is very similar to this positioning, and should be marked appropriately to handle the tracking of where the lapses in judgement were for future model correction means.

Environment Info

• Platform: linux
• Terminal: xterm-256color
• Version: 2.1.141
• Feedback ID: 2f0082b4-5bc1-4e5a-9d04-819cdece54ba

Errors

[]