[Bug] Claude model performance degradation after extended usage session

Bug Description
After 12 hours of work, opus 3.7 feels / seems dumber. It's more verbose, but i have to do much more verbal steering, and being explicit. Similarly, it's lost recognition of previously working commands and instructions like "/batch" and "use agents in a team". Periodically when I ask for either of them, the response is "we don't have that skill" (paraphrasing). It used to work perfectly and now feels like a step back from models even 2 months a ago. Please bring back 4.6 (12/2025 version)! I'm disappointed even though I know y'all are working hard.

Environment Info

• Platform: linux
• Terminal: zed
• Version: 2.1.112
• Feedback ID: 08f22e3a-44bc-4dc4-944e-7fb7e00aaa6d

Errors

[]