Assistant text before a tool call (e.g. ask_user popup) is hidden in the collapsed "Thought for…" block instead of rendering as chat

[j-alicia-long]

Describe the bug

In interactive mode, any assistant chat text written in the same turn before a tool call (most notably the ask_user question popup) is never rendered as a visible chat message. It is instead folded into the collapsed "Thought for Ns" reasoning block, where it stays hidden unless manually expanded — even after the tool call resolves.

The practical impact: when the agent writes explanatory context ("here's the quoted text / options / proposal…") and then opens an ask_user popup referring to that context, the user only sees the popup and cannot see the text it refers to.

Affected version

1.0.73 (Copilot app / workspace UI; copilot --version in shell reports 26.5.2)

Steps to reproduce the behavior

Ran a series of controlled tests with the agent in interactive mode:

1. Agent emits a chat message containing a marker + blockquote, then calls ask_user in the same turn → user sees only the popup; the text never appears, even after answering. It is only findable by expanding the "Thought for Ns" block.
2. Agent emits a text-only turn (no tool calls) → text renders normally as visible chat.
3. Agent calls a tool first, then writes text after the tool result in the same turn → text renders normally.
4. Markdown inside the ask_user question field (blockquote, inline code, bold) renders correctly, and long question text is not truncated — so the popup itself is fine; only preceding same-turn text is lost.

Expected behavior

Assistant text preceding a tool call in the same turn should render as a normal, permanently visible chat message (before or alongside the popup), not be classified as model reasoning/thinking and collapsed.

Additional context

• Operating system: macOS (Darwin, ARM)
• Surface: Copilot desktop app (Tauri workspace UI) project session, interactive mode
• The misclassification appears to be in how the UI segments a turn: everything prior to the first tool call is treated as part of the thinking block rather than as assistant output.