Desktop app replaces assistant text before a tool call with a generated summary, original text unreachable

Summary

In the desktop app (Code tab), assistant text written before a tool call in the same turn is replaced by a generated summary paragraph under a collapsible header. The assistant's own wording is not shown, even when the header is expanded. Text that ends the turn shows in full, so this only hits explanations the model writes before a tool call, most visibly before an AskUserQuestion call.

Reproduction

1. Desktop app, Code tab, Claude Code 2.1.258, macOS 26.6.
2. Have the model write a few paragraphs, then call AskUserQuestion in the same turn.
3. The transcript shows a header such as "Rewrote the question-field rule to answer first, then ask >" and under it a summary the model did not write (own wording, em dashes). The model's paragraphs are not in the transcript. The dialog shows only its question field.

Expected

The assistant's text shows as written. A summary on top is fine as long as the original is reachable by expanding it.

Impact

Any explanation the model gives before asking a question is lost. The workaround is stuffing the explanation into the question field, which renders without line breaks or bold, so it turns into one long paragraph in the dialog header.