[Bug] Ctrl+O transcript view doesn't tail new messages (still repros on 2.1.119, re: #17126, #41187)

Filing per #17126's auto-close message ("If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant."). It is — still reproduces on Claude Code 2.1.119.

Same root cause as #17126 (originally framed around background processes) and #41187 (verbose transcript not auto-refreshing while thinking): the Ctrl+O transcript view doesn't tail new content. You have to toggle it off/on to see anything new. My earlier issue #53353 was closed as a duplicate of #41187 → #17126; reposting here with a clean repro since #17126 is locked.

Summary

After toggling verbose/transcript view with Ctrl+O, the view does not follow new content:

• New user prompts and agent responses are not rendered while the transcript view is active.
• The agent does receive and process new messages — they're just invisible until you toggle Ctrl+O off.
• Toggling Ctrl+O off immediately reveals all the "missing" output.

This makes the verbose/transcript view effectively unusable for ongoing conversations: nothing new appears, but work is silently happening underneath.

Environment

• Claude Code: 2.1.119
• Model tested: claude-haiku-4-5-20251001 (launched with --model haiku)
• macOS: 26.3.1 (build 25D771280a)
• Node: v24.12.0
• Shell: zsh
• Terminal: reproduced inside tmux 3.6a (TERM=tmux-256color); originally observed in a regular macOS terminal as well

Reproduction steps

1. Launch Claude Code: claude --model haiku in any directory.
2. Send a message, e.g. say hi in one word. Confirm you get a response (Hi).
3. Press Ctrl+O. The footer changes to:

Showing detailed transcript · ctrl+o to toggle · ↑↓ scroll · [ to print output · v to open in vi
Existing messages now show a timestamp + model annotation (e.g. 12:36 PM claude-haiku-4-5-20251001).

1. Send a new message, e.g. count slowly from 1 to 5, one number per line.
2. Wait for the agent to finish (Haiku responds in a few seconds).
3. Observe the transcript view: neither the new prompt nor the agent's response appears.
4. Press Ctrl+O again to exit transcript view.
5. The previously missing prompt and response are now visible in normal history.

Expected behavior

While transcript / verbose view is active, new user prompts and agent responses (including streaming output) should render in real time, the same way they do in normal view — just with the additional verbose annotations (timestamps, model, etc.). Essentially tail -f semantics for the transcript view.

Actual behavior

The transcript view behaves like a static snapshot taken at the moment Ctrl+O was pressed. Anything that happens afterward is hidden until you toggle Ctrl+O off, at which point the new content suddenly appears.

Additional notes

• Confirmed the agent is actually running while the view appears static — the response is fully present in history once Ctrl+O is toggled off.
• The current input prompt also seems to swallow keystrokes for viewer navigation (↑↓, [, v) while in transcript view, which compounds the perception that the UI is unresponsive.
• Possibly related but distinct issues: #46261 (Ctrl+E transcript toggle state mismatch on macOS), #40984 (Ctrl+O hides Claude questions), #48083 (Write tool diffs missing in verbose mode), #25980 (thinking blocks not shown in verbose).

Workaround

Toggle Ctrl+O off whenever you want to see new output, then optionally toggle back on to inspect it with verbose annotations.