[CLI/TUI] Resumed threads ignore command collapsing and flood the view with historical tool output
What version of Codex CLI are you using?
codex-cli 0.150.1
What subscription are you using?
ChatGPT Plus
What model are you using?
gpt-5.6 models
What platform is your computer using?
Linux 6.18.33.2-microsoft-standard-WSL2 x86_64 x86_64
What terminal/shell are you using?
VS Code integrated terminal 1.135.0, WSL2 Ubuntu 22.04. No tmux, screen, or Zellij.
What issue are you seeing?
When I resume an existing, multi-turn Codex CLI thread, the loaded TUI is dominated by a very large amount of historical command/tool output and other low-value transcript text. The actual user prompts and assistant replies are sparse, difficult to find, or effectively obscured by the wall of text.
This appears to be a recent regression in how resumed threads are rendered/loaded. It looks as though the normal presentation of executed commands (including the newer command-collapsing behavior) is not being preserved or applied consistently when the thread is reconstructed from history.
This report is not:
- a request to show only the latest turn;
- a request to disable command collapsing; or
- primarily a performance complaint about terminal repainting.
The problem is that the resumed conversation view presents a huge amount of irrelevant/raw historical tool text instead of a readable conversation with clear user/assistant message boundaries.
Steps to reproduce
- Use Codex CLI for a multi-turn session that includes agent commands/tools and normal user/assistant messages.
- Exit the session.
- Resume it with
codex resume --last(or resume the thread by ID). - Inspect the restored TUI transcript.
What did you expect to happen?
The resumed thread should show the same readable transcript structure as the live session: user and assistant messages should be easy to identify, while historical command/tool activity should remain appropriately collapsed or otherwise subordinate to the conversation.
What actually happened?
The restored view shows a huge wall of command output, transcript fragments, and other historical text, with very little visible user/assistant conversation. The result is difficult to navigate and makes the resumed thread look like raw terminal/log output rather than the conversation that was saved.
Additional notes
The behavior is visible in resumed sessions and seems temporally related to the recent thread-history and command-display changes, but I cannot identify the exact introducing commit. Related reports include #37635 (resume-time transcript repainting) and #39903 (command collapsing), but this issue is about the incorrect/irrelevant content presentation and loss of a usable conversation view, not those narrower problems.
The screenshot from the affected session contains local filesystem paths, so it is not included here in raw form.