add an option to disable “Ran N commands” collapsing and always show executed commands

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #39819
• #39869

Powered by Codex Action

This "feature" must be disabled by default, its the worst change since Plan mode's timer's integration - making it unusable.

This is a serious UX regression. Commands are actions executed on the user’s machine, and users should be able to see them continuously in the primary TUI without opening a separate transcript. Requiring Ctrl+T adds friction precisely where transparency and auditability matter most.

Please keep individual command activity visible, or provide a dedicated option to disable command compaction while preserving the normal rich TUI. tui.raw_output_mode = true is not an equivalent solution because it replaces the rich renderer rather than disabling this one compaction behavior.

I'm reverting Codex back to 0.148.0 until this bug is not fixed.

I am seeing the same regression with Codex CLI 0.149.0 on macOS in Warp.

Uploaded thread: 01a0234c-adf6-7d73-9267-ad7621e18775

This is not only a UX preference. Inline command visibility is a safety and user-control mechanism: it lets the user continuously observe what the agent is doing, notice an incorrect path, target, flag, or destructive operation, and intervene while the task is still running.

Ctrl+T is useful for retrospective inspection, but a separate transcript view is not equivalent to continuous chronological visibility in the main TUI. Hiding successful commands by default weakens real-time oversight precisely when the agent is acting autonomously.

The compact view can remain available, but it should be configurable. Please add a persistent config.toml option to keep every executed command visible inline, for example:

[tui]
compact_command_activity = false

An interactive toggle would also be useful, but the important requirement is a persistent opt-out from command collapsing.

While we're on the topic, it is incredibly annoying how you can't see what commands are currently running. /ps will only return the first line of a command, which is often something like starting a python interpreter, and then you can't actually see what code is being run. How am I supposed to audit what the model is doing if I can't even see the command? If we're just supposed to blindly trust the model, then maybe OpenAI should wait until their models don't make obvious blunders at least a couple of times per session.

One additional reason I need command invocations visible is agent observability. The Ran ... lines make it easy to spot redundant searches, repeated tests, retry loops, overly broad commands, or generally inefficient approaches.

There are really three useful levels:

1. Ran N commands - forced compact view since 0.149
2. Ran ... - individual commands, output collapsed, previous behavior before 0.149
3. Ctrl+T - full transcript. This is not a substitute: command invocations are low-volume/high-signal, while command output is often high-volume.

Please keep level 2 available as a TUI option.

Not having a flag or config setting to toggle this is crazy.

I have the same issue in Codex CLI 0.149.0. I do not need full command output, but I do need each executed command to remain visible inline while the agent works. Ctrl+T is useful for full retrospective output, but it is not a practical replacement for continuously seeing command names, paths, flags, retries, and test invocations. This should be shown by default. Why change what works?

Either revert or provide a persistent opt-out such as:

[tui]
compact_command_activity = false

Ideally, this would support a command-only or bounded-preview mode so users can keep monitoring execution without flooding the TUI.

I completely agree, I use the TUI exactly for its transparency and this completely made that pointless.

Consider a generalized solution where the user can expand any collapsed section by (for example) navigating up to it and hitting + to expand, - to collapse.

@cowwoc Please don't: if I wanted mouse-focused UI I would use Codex Desktop, not the CLI!

And as other commenters have noted: we want to keep track of what Codex is doing, to stop it if it's going in a wrong direction. Manually expanding every command really doesn't fit here.

@cowwoc Please don't: if I wanted mouse-focused UI I would use Codex Desktop, not the CLI! And as other commenters have noted: we want to keep track of what Codex is doing, to stop it if it's going in a wrong direction. Manually expanding every command really doesn't fit here.

I tend to agree with this. Why change something that worked rather well? I mean was there a complain about this? It wasn't as cluttering and the TUI already handled the output well by showing only a few lines and collapsing everything else. What we have now completely obfuscates the information and the direction the AI is doing. Having this 'at a glance' has been quite informative yet practical for us.

@clemisch I get what you're saying but this isn't what I was trying to propose.

I am proposing enabling users to expand/collapse individual elements using keyboard navigation alongside the change you are asking for.

Meaning, you should be able to configure the default behavior in config.toml, and make minor adjustments at runtime using keyboard navigation.

I agree that transparency is important, but even with the old behavior once you spot a problematic command execution the only way to dig into it is with CTRL+T. If the command in question is further up the history log, you are going to have a hard time finding it.

Thanks for the feedback. The "Ran N commands" change was prompted by strong user feedback that the TUI output was too verbose, especially when it comes to tool calls.

We'll revisit this decision and see if we can come up with a solution that satisfies everyone — either some middle ground or a configuration option.

https://github.com/openai/codex/pull/38921

I can reproduce this on codex-cli 0.149.1 on Linux in the interactive TUI.

I also confirmed that the installed release contains the upstream commit 1f41cc5d9272, “Compact successful command activity in the TUI” (PR #38921). The full transcript is preserved and Ctrl+T works, but the compact view hides the commands during normal operation.

For developer and operations workflows, I need to see each command inline while the agent is running so I can audit paths, flags, environment variables, and potentially destructive side effects in real time. The compact presentation is useful as an optional mode, but it should not be the only mode.

Please consider an opt-in/opt-out setting such as:

[tui]
compact_command_activity = false

or an interactive toggle/slash command. The setting should also apply when replaying history. I would prefer preserving the new grouping as the default if desired, while allowing developers who need continuous command visibility to restore the previous expanded presentation.

This is a client-side TUI behavior; no proxy layer is involved.

@etraut-openai Sorry for the added noise, what's the proper channel to provide strong counter-user feedback for this new behavior? Upvote this issue? Because this new lack of transparency of what's being run is making codex substantially less steerable (simply don't have the info to steer it—"ctrl+t to view transcript" is hardly a solution since if "TUI output was too verbose" before, reading the ctrl+t transcript is hellish due to very poor highlight/delimitation) these past couple of days and noticeably affecting my success rate with it.

@zmwangx pin version 0.148 for now

what's the proper channel to provide strong counter-user feedback for this new behavior

@zmwangx I think given that we got a dev response three days after this issue was opened, this counts as an effective (as can be, at least) feedback channel?

I've experienced much slower reactions to seemingly much stronger issues, so tbh I'm pleasantly surprised.

It's disappointing (and a bit baffling) that such a major UX change like this was shipped without an option to disable.

This seems to apply to subagent threads as well. Either that, or the v1 schema has again be revised.

Absurd changes require absurd solutions: You can work around this by telling Codex to end each command with exit 1 and they still show.

<img width="540" height="350" alt="Image" src="https://github.com/user-attachments/assets/140c8463-75bf-4ce6-90a0-47278b3d3814" />

Also, the auto-approve passing conclusion is not shown; I cannot even tell if something escalated in these "commands ran".

Exactly, this is a hostile change and must be immediately reverted. Deliberations about the NEW feature can be done later.

Anyways, just revert to older version and don't update.

Common sense says NEW features affecting UX must be opt-in. I don't know why this change was done.

<img width="483" height="33" alt="Image" src="https://github.com/user-attachments/assets/cd028f08-7dac-44ea-a40f-4ca65b804d83" />

This is the default user experience in Codex CLI now. And view transcript is just a wall of text. Unusable.

+1

Absurd changes require absurd solutions: You can work around this by telling Codex to end each command with exit 1 and they still show. <img alt="Image" width="540" height="350" src="https://private-user-images.githubusercontent.com/133508953/640603073-140c8463-75bf-4ce6-90a0-47278b3d3814.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3ODc3MzQxMzMsIm5iZiI6MTc4NzczMzgzMywicGF0aCI6Ii8xMzM1MDg5NTMvNjQwNjAzMDczLTE0MGM4NDYzLTc1YmYtNGNlNi05MGEwLTQ3Mjc4YjNkMzgxNC5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjYwODI2JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI2MDgyNlQwODQzNTNaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT1jM2VlNjc0ZjAwYmViOTk3NGVlODUwNjYzM2M4ZDdlYTk5ZWYyOTYyNDNmZDIwYjcyMjljYjRlYTlmYTcyNDI4JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZyZXNwb25zZS1jb250ZW50LXR5cGU9aW1hZ2UlMkZwbmcifQ.UX2GROqQUkPg1jjB7rSEZCQD4CR68JpOZ4_JbZbzIUo">

This is genius actually 😁

I've added a hook that appends exit 1 to every bash command, and it works like a charm! I explained the purpose of it in AGENTS.md so that Codex doesn't stress over it, and now I can see all commands again. Feels great.

I've added a hook that appends exit 1 to every bash command, and it works like a charm! I explained the purpose of it in AGENTS.md so that Codex doesn't stress over it, and now I can see all commands again. Feels great.

It may feel great for now, but the purpose of AGENTS.md is not to hold every little thing you do, but to hold some highly critical information and instructions, but mostly to be a router to all your harness: workflows, prompts, guidelines, important docs to follow, scripts, etc. Otherwise, you may find yourself in a cluttered, ugly file, holding every little thing you do in your environment, taxing your agent reasoning.

> I've added a hook that appends exit 1 to every bash command, and it works like a charm! I explained the purpose of it in AGENTS.md so that Codex doesn't stress over it, and now I can see all commands again. Feels great. It may feel great for now, but the purpose of AGENTS.md is not to hold every little thing you do, but to hold some highly critical information and instructions, but mostly to be a router to all your harness: workflows, prompts, guidelines, important docs to follow, scripts, etc. Otherwise, you may find yourself in a cluttered, ugly file, holding every little thing you do in your environment, taxing your agent reasoning.

Of course. It's just an ugly workaround, and I really hope they fix the squashing soon

A very user hostile UX change. I hope that at least an option is provided soon.
This is a transparency/supervision issue.

Before every shell invocation, including local-file content reads and searches, print only the exact command in a visible commentary `bash` block, include every explicitly set environment variable and argument, and add no narration or invented commands. Omit visibility for direct file-edit/apply-patch operations that do not invoke a shell, as well as `git add` and `git commit` commands; all other shell invocations remain visible.

This is what im using in my AGENTS.md as a temporary thing until this is fixed
Its getting ridiculous using codex as a harness

Please don't do this OpenAI. I'm sure I'm not the only one who stopped using Claude and switched to Codex when Anthropic decided to start doing this with Claude Code in an effort to hide all forms of feedback from the user. I know that hiding things from the user is not your intent like it was for them, but it's not a direction this product should move in at all. If you want a mode that keeps your chat history window compressed, it should be opt in. Codex CLI is your most power-user oriented product, it should remain that way.

This is such a bad UX change and I literally thought something was up with my codex cli, there should be an option to make it work how it was like before.