Setting the status line for all projects in Codex

OpenAI's Codex CLI version 0.146.1 restricts the TUI status line to a fixed list of 26 built-in item identifiers, unlike Claude Code which accepts any executable command, according to a developer's analysis of the Codex source code. The global configuration file ~/.codex/config.toml allows ordering items such as "model-with-reasoning", "current-dir", and "git-branch", but custom scripts are not supported. The author found that codex doctor does not validate status line item names, as a test with a bogus item still reported a valid configuration.

After writing about setting Claude Code status line with Ruby https://allaboutcoding.ghinda.com/you-can-use-ruby-to-set-the-status-line-in-claude-code/ , I wanted to see if the same thing is possible for Codex and dig a bit more. First, all 3 let you set the status line globally, at the user level, or locally per project. Second, the way they work is different, and only Claude accepts a command as I described it in the previous article https://allaboutcoding.ghinda.com/you-can-use-ruby-to-set-the-status-line-in-claude-code/ . Here is a summary of the findings so far: | Tool | Global file | Setting | What it accepts | |---|---|---|---| | Claude Code | ~/.claude/settings.json | statusLine | a command, so any executable | | Codex | ~/.codex/config.toml | tui status line | an ordered list of built-in item names | | opencode | ~/.config/opencode/tui.json | plugin | a TUI plugin with a JS or TS entry | Thus, a Ruby script can be used directly only in Claude Code, while Codex only allows specific items to be displayed there, and opencode will need a JavaScript entry point that can run a Ruby script if you want it to. Codex will read the global config from ~/.codex/config.toml and then load it if it exists in .codex/config.toml from the project folder where it runs. The settings can be defined as: tui status line = "model-with-reasoning", "current-dir", "five-hour-limit", "weekly-limit", "git-branch", "context-used", "context-window-size", "fast-mode" status line use colors = true You cannot define a custom script or maybe I could not find it quickly, but as Codex is open source, you can dig more if you want in their source code , so you have to pick from what Codex allows you to choose and then put those in the order that you want to see them. The config documentation https://learn.chatgpt.com/docs/config-file/config-reference says this about the status line: Ordered list of TUI footer status-line item identifiers. null turns off the status line. I could not find in the documentation the exact list of identifiers, so I dug it in the Codex source and found this list of possible status line items at StatusLineItem https://github.com/openai/codex/blob/rust-v0.146.1/codex-rs/tui/src/bottom pane/status line setup.rs L50-L135 from version 0.146.1 with 26 items: approval-mode branch-changes codex-version context-remaining context-used context-window-size current-dir fast-mode five-hour-limit git-branch model model-with-reasoning permissions project-name pull-request-number raw-output reasoning run-state task-progress thread-id thread-title total-input-tokens total-output-tokens used-tokens weekly-limit workspace-headline Please consider this a discovery from the source code. Since there is no official documentation for this, there is no interface contract to keep it as it is I made a config with a non-existing name totally-bogus-item : tui status line = "current-dir", "git-branch", "totally-bogus-item" and then I run the doctor command: CODEX HOME=/tmp/codexhome codex doctor --summary And the configuration section came back happy: Configuration ✓ config loaded This may be a bug, but I don’t know. But you should not rely on codex doctor to verify that the names you added there are correct. Just check what the status line looks like, and may I suggest adding items incrementally, one by one. I like the Claude Code approach more in this case than Codex because you can do a lot more with status line processing than Codex can by default. For example, having the session id there, I already changed my statusline.rb file to read the session and tell me the last tool that was executed, how many prompts I have written so far, and the context used in the last turn. Codex does not give this option. You mostly get what the team already implemented. Of course, as it is open source, you can always fork and implement what you need and then make a PR back to the upstream. What they offer is more performant than computing things from the session file. I would still like to have something like status line command in Codex that behaves similarly to how Claude Code does.