claude-code - 💡(How to fix) Fix AskUserQuestion overlay obscures the assistant's previous message

Summary

When the assistant invokes AskUserQuestion, the resulting question panel (with 2–4 option buttons + the implicit "Other" option) is rendered as an overlay/modal that covers the message the assistant just wrote. If that message contains a long report — context that the user needs to read in order to choose between the options — the user can no longer see it once the picker appears.

Steps to reproduce

1. Have the assistant produce a long, paragraph-rich message that ends with a decision point (e.g. "Which approach do you want to take?").
2. Have the assistant immediately invoke AskUserQuestion after that message.
3. Observe that the option picker appears on top of (or scrolls past) the body of the previous assistant message, hiding the very context that explains the options.

I hit this in a real session: the assistant had just delivered an 18-item UX audit grouped in three priority sections, and asked me to choose which group to attack first via AskUserQuestion. The audit text was no longer reachable while the picker was open, so the question became impossible to answer without first dismissing the picker (and even then, scrolling back was awkward).

Expected

Either:

• The picker is anchored below the assistant message (so both stay visible), or
• The assistant message scrolls into view above the picker automatically, or
• The picker is dismissible non-destructively so the user can re-read context and re-open it.

Currently the most robust workaround is for the assistant to not use AskUserQuestion whenever the prior turn contains substantive content the user needs to read to answer — which makes the tool unusable in exactly the cases where structured decisions matter most.

Environment

• Claude Code CLI on Windows (PowerShell host)
• Encountered during a multi-week project session

Why it matters

AskUserQuestion is supposed to make decisions structured and quick. When it covers the context that justifies the decision, it does the opposite — the user has to dismiss it, scroll, re-find the explanation, then re-trigger the flow. For long-running coding sessions where the assistant routinely produces dense reports before asking for direction, this turns a quality-of-life feature into friction.

Reported on behalf of an end user who explicitly asked for this to reach the team. Happy to provide screenshots if helpful.