Desktop app: assistant text followed by tool calls in the same turn is never rendered (swallows analysis and AskUserQuestion preambles)

Summary

In the Claude Code desktop app, assistant text messages that are followed by tool calls in the same turn are not rendered in the transcript. The turn displays only the collapsed tool-activity chips (e.g. "Ran 3 commands, edited a file ›", "Saved 2 memories ›") and the final text block that comes after the last tool call. Any substantive prose the model wrote between tool calls — or before them — is silently dropped from the UI.

Impact

This is worse than cosmetic:

1. Lost analysis: In my session, the model ran two quick verification commands, then wrote a ~700-word analysis answering my questions, then made file edits based on it. The UI showed: my question → "Ran 3 commands…" chips → the model's post-edit wrap-up ("All of round 2 is recorded and committed…"). The entire analysis — the actual answers to my questions — never appeared. I only discovered it existed because the wrap-up referenced conclusions I'd never seen, and I asked the model to resurface what it had written.

1. Blind decisions via AskUserQuestion: The same swallowing affects text that precedes an AskUserQuestion tool call. I was presented with multiple-choice decision questions whose explanatory preamble (written immediately before the tool call) never rendered — I answered product decisions essentially blind. Typing a question into the "Other" option just re-presents the questions; the model's replies are never shown while the question control is active.

Steps to reproduce

1. In the desktop app, ask something that causes the model to: run a tool → write a substantial text response → run more tools (e.g. file edits) → write a short final summary, all in one turn.
2. Observe the transcript: only the first pre-tool sentence (sometimes), the collapsed tool chips, and the final summary render. The mid-turn text block is missing.
3. Variant: have the model write an explanation and then call AskUserQuestion in the same turn — the explanation does not render before (or after) the question UI.

Expected

Every assistant text block in a turn renders in order, interleaved with the tool-activity chips — matching the CLI behavior, where mid-turn text is visible.

Environment

• Claude Code desktop app (macOS, Darwin 24.6.0 / macOS 15)
• Model: claude-fable-5
• Observed 2026-07-06, reproduced multiple times in one session (both the mid-turn-text case and the AskUserQuestion-preamble case)

(Filed with the assistance of Claude Code at the user's request.)