[Bug] Claude Opus 1M hangs after tool execution or task completion

Bug Description
Ever since you rolled out the 1m token version of Opus, most of my sessions have now involved me having to figure out if the agent is hung or actually still doing work (I have found that at least I can usually confirm this with the btw feature), then hitting cancel and moving forward. It gets in a state where practically every time it calls a tool, it finishes successfully and then just hangs. It's better with Sonnet, but you have massively downgraded my experience with Opus.

Example: I ask Claude Code to research something and help construct a plan. It does a grep on a file, completes the grep, and then sits there "fidgeting" with the spinner forever. So I hit escape, ask if it is "thinking" and it says it got the info it needs, and will create a plan. This is different from when I actually interrupt it while it is thinking/working, in which case it recognizes that I interrupted it.

I have the same experience in either CLI or VS Code plugin.

Environment Info

• Platform: darwin
• Terminal: Apple_Terminal
• Version: 2.1.79
• Feedback ID: 560fac36-a737-41b9-9378-16abd8c0b071

Errors

[]