UX improvement recommendations for Codex CLI

[marektomas-cz]

trafficstars

What feature would you like to see?

I am a developer who used Claude "you are absolutely right" Code for a long time, and a few days ago I switched 100% to Codex CLI, because GPT-5 with reasoning is a significantly higher quality model than competing models. I also do not support the competitor’s shady practices, I will not go into details.

Now to the recommendations. I believe you already have a lot on the TODO list, but I will summarize everything from a UX point of view, as someone who has tested a number of competing CLI tools. I believe Codex has the potential to be the best CLI coding tool, but it needs many improvements.

1. Subagent support. It should be possible to define and run subagents for tasks. The benefit is that the main context window does not get cluttered, and subagents send only a summary of the work they did into the main window.

2. /history or a similar slash command. This should let a user resume previous sessions. This is a very critical requirement.

3. Control over the thinking view. Add options to hide it, show snapshots only, or show the full thinking process. I personally prefer seeing the full process, but for some people it can be distracting from a UX perspective.

4. Pasting copied text. I am not sure if this is Windows only, but when I paste a copied link, backslashes \ are removed. Another problem, when text with multiple paragraphs is pasted into the input window, only the first paragraph is inserted and the prompt is sent automatically. This needs a fix.

5. Better MCP server management. Improvements are needed (UX and TUI).

6. Background bash shell support. During development, the CLI should be able to start a background bash session, keep it alive, and during the conversation be able to modify that session, read logs, and so on.

7. Native Windows support. This may again be Windows specific, but there are issues with installing MCP servers and with functionality on Windows in general. Native support would be better than a workaround. There are still many Windows developers. Right now it feels like there is no native Windows support, only a wrapper, sandbox, or tunnel via WSL. The model thinks it is in Linux or WSL, which it probably is, and this is a problem. Please add native Windows support.

8. Add a TUI switcher for Plan and Build.

9. Overall TUI improvements. It really does not look good yet. Especially the TODO section. Please add an option for the agent or LLM to update the TODO during the session if, during its thinking process, it realizes that changes are appropriate. It would also be great to make the TODO more visible. It appears only once, then it gets lost in the thinking spam.

GPT-5 is an excellent model, but it is a pity that OpenAI does not allocate more manpower to developing tier 1 CLI tool. I believe that would attract many clients, and many would switch from the competing "you are absolutely right".

Thank you for your work.

Are you interested in implementing this feature?

No response

Additional information

No response

[nilswx]

Good list. Undo/redo would also be useful: https://github.com/openai/codex/issues/2379.

[luuk00101]

Great suggestions!

[gpeal]

Thanks for all of these suggestions! You may have noticed that the pace of updates has increased significantly in the last few weeks and you should expect that to continue (including many of the ideas you listed). Make sure to keep the CLI up to date!

Example: subagents

[sldx]

+1 for /history and generally for resuming sessions. Not having that is literally destructive.

[am-will]

yes, need to be able to run agents in parallel and to resume/call history. Custom /prompts should also be able to run bash commands and have an option to run them in project scope

[Waishnav]

i wanted to use GPT-5 through my Codex subscription in Opencode, but I don’t think that’s going to be possible…
https://github.com/sst/opencode/issues/1686

So I’m hoping this project improves rapidly over the next few months.
Here are my 2 cents, a quick list of features I think are needed:

1. /undo — to revert patches applied by the tool call (similar to the /undo command in opencode).
2. /history — to browse through my past sessions.

[0xdevalias]

0.30.0 added the start of support for --resume and --continue:

• https://github.com/openai/codex/releases/tag/rust-v0.30.0
  • https://github.com/openai/codex/pull/3135
  • https://github.com/openai/codex/pull/3185
  • https://github.com/openai/codex/pull/3218

See also:

• https://github.com/openai/codex/issues/1991
• https://github.com/openai/codex/issues/3233

Originally posted by @0xdevalias in https://github.com/openai/codex/issues/2173#issuecomment-3264145154

[etraut-openai]

Thanks for the suggestions. Most (all?) of these are being tracked by other enhancement requests in the issue tracker. If there are specific ideas that you feel are not covered in other issues, please feel free to open a new issue. Filing one idea per issue helps us track and prioritize.

I'm going to close this now to reduce duplication.

[0xdevalias]

@etraut-openai For continuity when closing something as a duplicate because other issues exist; it's helpful to link to those issues from your closing message.