[BUG] Assistant text preceding an AskUserQuestion call in the same turn is never displayed (terminal CLI)

Environment

• Claude Code version: 2.1.202
• Platform: macOS (Darwin 25.5.0), terminal CLI (zsh)
• Model: claude-fable-5
• Mode: plan mode active during the session

Bug Description

When the assistant emits a markdown text block and then calls the AskUserQuestion tool in the same turn, the text block is never displayed to the user. Only the question widget is rendered. The text does not appear before the widget, and it also does not appear after the user answers the question.

How this manifested

In a real session, the user asked for an investigation. The assistant (per plan-mode workflow guidance, which instructs ending turns with AskUserQuestion or ExitPlanMode) repeatedly wrote substantial findings/reports as text and then ended the turn with an AskUserQuestion call. The user saw only the questions, never the findings — across at least four consecutive turns. From the user's perspective the assistant was "asking questions without ever answering", leading to escalating frustration ("You haven't told me anything. I don't see a report!") even though the transcript contained full reports before each question.

The problem was only discovered when the assistant ended a turn with plain text and no trailing tool call — that message rendered fine.

Steps to Reproduce

1. Start claude in a terminal (plan mode makes it easy to trigger, but any AskUserQuestion use should do).
2. Prompt something that makes the model write a substantive text answer and then ask a clarifying question via AskUserQuestion in the same assistant turn.
3. Observe the rendered conversation.

Expected Behavior

The assistant's text emitted before the AskUserQuestion tool call is rendered above the question widget (and remains in the scrollback after answering).

Actual Behavior

Only the AskUserQuestion widget is shown. The preceding assistant text is silently dropped from the visible conversation.

Impact

Silent information loss. This is especially damaging in plan mode, where the built-in workflow explicitly instructs the model to end every turn with AskUserQuestion or ExitPlanMode — so any analysis, findings, or reports written in the same turn as a clarifying question are systematically hidden from the user. It also breaks the documented pattern that answers/reports should be in the final message of a turn.

Related

• #74237 (VS Code: AskUserQuestion modal visually covers the conversation — different surface; this report is about the terminal CLI never rendering the text at all)