Desktop: markdown is not rendered in question/elicitation text (renders fine in normal messages)

Environment

• OpenCode desktop app 1.18.21 (ai.opencode.desktop), macOS 26.4 / arm64
• Latest release at time of writing

Summary

Markdown in the question text of an elicitation prompt is displayed literally.
Markdown in ordinary assistant messages renders correctly in the same build, so this
looks specific to the question/elicitation surface rather than to the renderer.

Repro

1. Normal message — renders correctly.

Prompt:

Reply with a normal message that contains: a ## heading, **bold text**,
`inline code`, a bulleted list, and a fenced code block containing an
HTML comment. Do not use any tools.

Result: heading, bullets, inline code chips and the fenced block all render as
expected.

2. Question tool — does not render.

Prompt:

Use your question tool to ask me a SHORT question whose text contains
**bold** and `inline code`. Give me two options.

The agent asked:

Should I **review** or `commit` the changes?

Result: displayed verbatim, asterisks and backticks included.

Why I think this is worth fixing rather than intended

There is a reasonable argument for showing question text verbatim: when a prompt asks
you to approve content that will be written to a file byte-for-byte, rendering the
markdown would hide the actual bytes — fences disappear, ## becomes a heading, an
HTML comment marker can vanish entirely.

But that argument does not fit what I see. The failing case above is a single line of
trivial markdown with no payload to protect, and it fails the same way. So this reads
as "not wired up" rather than "deliberately verbatim."

Context

I hit this with an agent whose question body was a few thousand characters of markdown
(a proposed file block being submitted for approval). Rendered as one run of flat
proportional text it was essentially unreadable. The short repro above is the reduced
case.

Whether or not agents should put long content in that field, they do in practice, so
a secondary suggestion: give long question bodies a monospace, scrollable or
collapsible container. That helps even if verbatim display is kept deliberately.

As far as I can tell this content reaches the UI through the question tool → ACP
elicitation bridge (#42104, with follow-up #42158 still open), so this surface is new
and this may simply not be implemented yet.