[BUG] Long prompts block model output when AskUserQuestion modal appears (VS Code extension)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

When I submit a long user prompt (one that fills most of the chat pane on first render) and the model responds by invoking an AskUserQuestion modal, the expanded prompt continues to occupy the visible area and covers the model's response context. I can't see what the model was reasoning about or why it's asking me the question. After I answer the modal, the response text often still isn't visible — the view doesn't reliably scroll to the latest output, so I'm left staring at my own prompt with no clear indication that the assistant has moved on or what it was asking me a question about.

What Should Happen?

• Long user prompts should auto-collapse (especially on first render) with an expand affordance, so the model's response and any modal context are visible by default.
• When an AskUserQuestion modal is dismissed/answered, the chat view should auto-scroll to the latest model output so I can see what I just responded to and what's happening next.
• Ideally, the first-time render of any prompt above a length threshold should be capped to N lines regardless of screen size.

Error Messages/Logs

Steps to Reproduce

1. Open VS Code on macOS with the Claude Code extension installed and signed in (Anthropic first-party auth).
2. Open the Claude Code extension panel (native panel, not a terminal).
3. In the chat input, paste a long prompt — long enough that once submitted it fills most of the visible chat pane (e.g., a multi-paragraph task description or spec; 40+ lines on a standard MacBook display reliably reproduces it).
4. Phrase the prompt so the assistant will likely respond with an AskUserQuestion modal — e.g., ask for a decision that has multiple valid approaches, or give it a task with ambiguous scope that requires a choice.
5. Submit and wait for the assistant to invoke the AskUserQuestion modal.
6. Observe: the submitted user prompt remains fully expanded in the chat view, covering the assistant's pre-modal reasoning/context and model response. The modal's question lacks the surrounding "why is it asking this" context.
7. Answer the modal.
8. Observe: the chat view does not reliably scroll to the latest assistant output. I often have to scroll manually, and sometimes the new output appears to never render into view at all.

Claude Model

Opus

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

2.1.113

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Cursor

Additional Information

_No response_