Feature Request: Delay AskUserQuestion modal until user has finished reading streamed response
Problem
When Claude streams a response that includes an AskUserQuestion tool call, the modal/dialog appears as soon as the tool call is parsed — often while the response is still streaming and before the user has finished reading.
This creates a disruptive experience:
- I'm reading the streamed response, typically only partway through
- The AskUserQuestion modal pops up, covering roughly half the screen
- I now have to choose between:
- Dismissing the modal to finish reading (losing easy access to the questions)
- Answering the questions without full context of what Claude wrote above them
The modal effectively blocks the reading experience at exactly the moment I need to absorb context to answer the questions well.
Affected Surfaces
This primarily impacts interfaces with visible token streaming:
- VS Code extension (where I notice it most)
- claude.ai web app
- Desktop app
- Likely other IDE integrations and agentic surfaces as well
Note: The Claude Code CLI does _not_ visibly stream tokens — the full response arrives at once, so the modal and the text appear together. This may explain why this issue hasn't been widely reported, since CLI users (who are most likely to file issues on GitHub) don't experience the disruption.
Suggested Solutions
A few possible approaches (not prescriptive, just ideas):
- Inline rendering: Render the question/options inline at the end of the response rather than as an overlay modal, so the user naturally encounters it after reading
- Scroll-aware delay: Defer displaying the modal until the user has scrolled to (or near) the bottom of the response
- Collapsed state: Show a minimal, non-blocking indicator (e.g., a small banner or badge) that the model has a question, and let the user expand it when ready
Impact
This is a relatively frequent interaction pattern — Claude often asks clarifying questions before proceeding, which is good behavior. But the current modal presentation penalizes the user for that good behavior by interrupting their reading flow.
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗