Support user-defined local slash commands in interactive Codex sessions

[nevertoday]

What variant of Codex are you using?

CLI

What feature would you like to see?

I would like Codex to support user-defined slash commands in interactive sessions that can run local actions directly, without sending anything to the model.

Today there are partial workarounds, but none of them solve this cleanly:

• custom prompts / skills still invoke the model and consume usage
• hooks are event-driven, not user-invoked
• shell aliases/scripts require breaking out of the Codex conversation flow

What I want is something like:

• /open -> open the current working directory in Finder
• /reveal README.md -> reveal a file in Finder
• /copy-path -> copy the current workspace path
• /run-local-task -> run a deterministic local shell/MCP action

The key requirement is that these commands should be treated as local commands, not prompts:

• no model invocation
• no token / usage cost
• fast deterministic execution
• visible output / confirmation in the current session

A possible design would be a dedicated config surface for local slash commands (for example ~/.codex/commands or config.toml), where a command can map to:

• a shell command
• an MCP prompt/tool
• an app action

This would make Codex much more usable for day-to-day interactive workflows, especially for users coming from tools like Claude Code where lightweight slash-driven local workflows are common.

Additional information

Related but not the same as existing requests:

• Support Slash Commands in codex exec Mode #3641 is about slash commands in codex exec mode
• Support for custom commands (custom slash commands) #7480 is about custom commands in the VS Code extension

This request is specifically about interactive Codex sessions: while I am actively chatting with Codex, I want to be able to run a local slash command such as /open without context-switching to a separate terminal and without paying model usage for something deterministic and local.

[etraut-openai]

We previously supported user-defined slash commands, but they were deprecated and removed in favor of skills, which provide a superset of the functionality. We therefore don't have any plans to reintroduce slash commands.

[jeevanpillay]

@etraut-openai thats straight bs. there are use-cases for slash commands that work outside of skills.

skills run whenever the agent think it needs to run it.
commands run when dev wants it run.

this is forcing a dev to have less configurability in workflows.

e.g we run /research_codebase before doing anywork in claude. and run /create_plan after. with slash commands these primitives dont exists.

[etraut-openai]

@jeevanpillay, you can specify that a skill should not be run automatically. See here for details. This was added to address the point you're raising.

[shafty023]

@etraut-openai a nice thing about a slash command is it allows you to define multiple skills in a single workflow. I can define a slash command in claude that specifies multiple skills to automatically inherit. I can't do that with the current skills support in codex.