AskUserQuestion truncates preceding text output in terminal viewport

Description

When a skill (or Claude itself) outputs long narrative text immediately before calling AskUserQuestion, the preceding text gets pushed above the terminal viewport once the interactive option picker renders. The user loses visibility of the context/recommendation that was just displayed — the very information needed to make an informed choice.

Steps to Reproduce

1. Use a skill that outputs detailed context before each AskUserQuestion call (e.g., gstack skills like /office-hours which output a Re-ground + Simplify + Recommend block before every question)
2. The skill outputs ~3-5 lines of narrative text including a recommendation
3. The skill then calls AskUserQuestion with options that have long descriptions
4. The interactive option picker renders and pushes the narrative text above the visible viewport

Expected Behavior

The terminal UI should ensure that text output immediately preceding an AskUserQuestion call remains visible (or at minimum easily scrollable to), so users can read the context before selecting an option.

Actual Behavior

The narrative text (context, recommendation, completeness scores) gets cut off / pushed above the fold when the AskUserQuestion picker renders. Users have to manually scroll up to read the recommendation before making a choice.

Screenshot

The screenshot below shows text being truncated at the bottom where options are displayed. The recommendation and option descriptions are partially cut off:

• The "RECOMMENDATION: Choose A because..." line and analysis are visible but the option description text at the bottom is cut off by the viewport

Why This Is More Apparent with Certain Skills

Most simple AskUserQuestion calls (e.g., "Which approach? A or B?") don't trigger this because there's minimal preceding text. Skills that follow a structured format — outputting context, analysis, and recommendations before each question — expose this issue consistently.

For example, gstack skills mandate a format that outputs:

1. Re-ground — project/branch/task context
2. Simplify — plain English explanation
3. Recommend — explicit recommendation with completeness scores
4. Options — lettered options with effort estimates

This is useful information that the user needs to see while making their selection, but the current rendering pushes it out of view.

Environment

• Claude Code CLI (macOS, Terminal)
• Happens consistently across sessions

Possible Solutions

• Auto-scroll the viewport so that the text preceding the AskUserQuestion + the question itself are both visible
• Render the preceding text as part of the question UI component (e.g., as a collapsible "Context" section)
• Allow skills to pass a context field to AskUserQuestion that gets rendered above the options but within the same UI component