claude-code - 💡(How to fix) Fix [BUG] Desktop transcript becomes unscrollable when wake messages pile up during long agent turn

Preflight Checklist

• I searched existing issues for scrollback / stuck message / MCP wake-message variants. Closest related issues I found: #58660, #58484, #60345. This report is narrower: incoming wake/external messages append during a long active turn and make the active session UI unusable.
• This is a single bug report.

What's Wrong?

In Claude Code Desktop / Cowork UI on macOS, incoming external wake messages can pile up while an agent is still inside a long-running turn. After enough messages arrive, the transcript area appears not to remain scrollable and the prompt/input field is pushed out of reach. The user can no longer see what the agent is currently doing, nor reach the prompt field, even though new wake messages still continue to arrive.

Observed failure mode:

• A Claude agent was busy for roughly 50 minutes and did not end its turn.
• During that time, repeated wake/A2A-style messages arrived, each rendered as a full separate message block in the visible transcript.
• The messages were not visually stacked, collapsed, or coalesced.
• The transcript/prompt area became effectively unusable: the user could no longer see the live agent activity or the prompt field.
• New incoming messages still arrived, so the transport was alive; this appears to be a UI layout/scrollability failure, not a backend delivery failure.

What Should Happen?

Incoming external/wake messages should not be able to make the session unusable. Expected behavior:

• The transcript remains scrollable regardless of how many incoming wake messages arrive.
• The prompt/input field remains reachable.
• Repeated wake/external messages should be stacked, grouped, collapsed, or otherwise compacted so long-running turns do not flood the visible UI.
• The user should still be able to inspect current agent output/progress while messages continue to arrive.

Steps to Reproduce

1. Open a Claude Code Desktop / Cowork session on macOS.
2. Start an agent turn that runs for a long time without yielding, e.g. 30-60 minutes.
3. During that active turn, deliver repeated external/wake messages to the same agent/session. In our case these were A2A wake/status messages shown as transcript blocks like [WAKE][priority:normal] ....
4. Let approximately 10+ wake messages accumulate.
5. Observe that the visible chat fills with separate wake blocks; the prompt field and current live activity become inaccessible/not visible.

Actual Result

The session UI fills with repeated wake message cards. The transcript does not stay usefully scrollable from the user's perspective, and the prompt field disappears below the visible area. The user loses the ability to inspect the live agent turn or type into the prompt, while incoming messages still continue to render.

Expected Result

Wake/external messages should be compacted or scrollable, and the prompt field should remain accessible. A long-running turn plus incoming messages should not trap the UI in a state where the user cannot inspect or interact with the session.

Environment

• App: Claude Code Desktop / Cowork UI
• OS: macOS
• Date observed: 2026-05-21
• Trigger: repeated incoming wake/A2A-style messages during a long active agent turn

Screenshot Evidence

A screenshot captured by the reporter shows many [WAKE][priority:normal] message blocks stacked vertically in the session, with the prompt field no longer visible. I am filing via GitHub CLI, so I cannot attach the image directly here; the screenshot can be provided by the reporter if needed.

Related / Not Exact Duplicates

• #58660: asks for scrolling in agent detail view; related scrollability issue but different trigger.
• #58484 / #60345: sticky/floating user-message overlay; related Desktop UI rendering/scroll bugs but not this wake-message accumulation case.