Chat rendering disrupted after assistant response with excessive trailing whitespace

Summary

During an active Kit coding-agent conversation, the user reported that chat rendering broke immediately after an assistant response. The transcript visible to the agent contains an ordinary short paragraph followed by a very large tail of blank/whitespace-only lines.

This is a user-observed incident, not yet a deterministic reproduction. A special/control character is a hypothesis from the user, not a confirmed cause. The agent can observe the whitespace-heavy message in conversation context but has not inspected raw transport bytes or the rendered screen. It is not yet known whether the whitespace originated in model output, message transport, or rendering, or whether the affected rendering layer is Kit or the host client.

Environment

Complete kit --version output:

kit 0.1.126-herdr.1

• macOS 26.6.2; Darwin arm64.
• Herdr-flavored local Kit build; active interactive coding session with background compose/subagent calls.
• Exact launch command, terminal/client version, and rendering layer have not been captured.

Observed sequence

1. An extended coding-agent session was running several background subagents.
2. The user asked the agent to continue working without stopping at routine checkpoints.
3. After starting background work and reading startup scripts, the assistant sent a short progress/final response.
4. In the transcript, that response is followed by a very large run of whitespace-only/blank lines.
5. The user immediately reported a chat-rendering problem and asked whether a printed character had broken rendering.

Expected / actual

Expected: ordinary readable chat output; whitespace-heavy assistant content should not disrupt the surrounding conversation layout or navigation.
Actual: user reports broken chat rendering coincident with the whitespace-heavy response. Exact visual symptoms (e.g. excess vertical space versus redraw corruption) still need confirmation; no screenshot is attached.

Suggested minimal diagnostic probe (not yet reproduced)

Exercise the renderer with a short assistant paragraph followed by hundreds of repeated whitespace-only lines (for example repeated \n \n). Compare streamed rendering with history/replay rendering. Check whether control characters exist in the original event before attributing the issue to them.

No private prompts, full session logs, credentials, or repository artifacts are attached. Open and closed issues were searched for render whitespace, blank lines, and rendering; no matching whitespace-related report was found.

Start with the chat renderer entry point and the streamed-rendering versus history/replay paths. Run the suggested probe with a short paragraph followed by hundreds of whitespace-only lines, and inspect the original event for control characters before assigning blame. Done means the rendering behavior and affected layer are reproduced and a regression test or clear diagnosis covers the case.