[Feature Request] Reduce unsolicited internal reasoning in outputs and add privacy controls for sensitive contexts

Bug Description
I've noticed lately that Claude is eager to write out its "thoughts" without user inputs. Validating its own ideas without searching up anything, and acting more "on its own". That's not very useful, and more importantly introduces uneccessary risks into using Claude. Leaks into the chat-window are potential security issues, beware.

Environment Info

• Platform: darwin
• Terminal: tmux
• Version: 2.1.169
• Feedback ID: a946c47d-ed9f-4b57-9c3f-c70a09a0b09b

Errors

[]