Alias to `/` to `$` when an existing command does not resolve
Resolved 💬 3 comments Opened Apr 24, 2026 by banagale Closed May 31, 2026
💡 Likely answer: A maintainer (etraut-openai, contributor)
responded on this thread — see the highlighted reply below.
Allow the TUI to automatically resolve toward known skills when /[token] is unmatched.
Pass-through complete autocomplete experience enjoyed from proper $[token] use.
Reasoning:
- Presumably data shows skills vastly outweigh command use at this time.
- If a user is being captured from CC, they will not know about
$. Muscle memory is attuned to/. And the memory is strong. - Current TUI rejects CC-valid skill invocation as per #17796
- This rejection actually makes it look like the product doesn't know about skills! At very least, output should provide a hint toward skill invocation, as again data probably shows the amount of failed "command" invocations are largely skill invocation attempts.
I am a long time codex user (as much as that can be) and did not know about it until @etraut-openai specified here.
3 Comments
Our priority is to unify the prompt experience across all of our Codex surfaces (TUI, web, app, IDE Extension). We've landed on
@as the unified way to mention files, skills, plugins, etc. We will also support$as a way to invoke skills and plugins only, without files appearing in the picker list. We don't currently have any plans to add support for/as a third option.Thank you for the additional feedback.
I did not know about @ as a skill invocation either and have been 100% using
/to invoke skills with no idea that the the TUI provided autocomplete for @ or $.If the team is to hold the course on these invocation methods, I would strongly recommend "tip:" feedback for the command not identified error and apparent attempts to invoke skills. (The model understands my existing / usage though!)
I want to reflect my experience with codex surfaces you identified:
I use terminal for virtually all code interaction now. I do sometimes submit large context code review requests to chatgpt.com--but my CLI AI skills are designed for the power of the terminal and invalid in the web browser.
After over a decade of loyalty to Pycharm, I chose not to renew my pycharm license this year. I was the kind of person to file an discuss bugs and feature requests. However, I do not forsee a need to move back toward IDEs at least as they are understood today.
What I am doing is constantly evaluating CC and codex by putting them in one of five worktrees represented by tabs in
iterm2and running them all day (and sometimes at night).I was initially aghast at CC's decision to drop "real" commands and say ~everything is a skill, but after so many months and building multiple Skill + CLI combos I am completely sold on this idea and convinced it was the right move.
I do not think codex should be chasing the product behavior of CC and mean no disrespect in comparing the product in this way.
However, I would encourage a re-evaluation of the TUI's behavior in response to skill entry. It might seem like a small thing, but if flipping users is the game, this would be an easy get.
If it would be helpful, I can screenshare a session of usage with a team member that illustrates the types of skills and invocations I'm doing.
This feature request hasn't received enough upvotes, so I'm closing.