slash commands parity in vscode extension

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

• #7949
• #5917

Powered by Codex Action

Thanks for the suggestion.

Many of the slash commands in the CLI either don't make sense or have GUI-specific implementations. This includes "/resume", "/mention", "/logout", "/quit", "/diff", "/approvals", "/model" and "/new".

We've already implemented "/feedback" and "/status".

We're working on a new GUI configuration for MCP, so an "/mcp" slash command would be redundant.

The "/compact" feature is increasingly obsolete now that we have high-quality auto-compaction implemented in the latest models, so we haven't prioritized that.

The "/undo" feature has been pretty problematic; I think we need to figure out how to fix it or replace it with something better.

I think that just leaves "/init".

@etraut-openai
Did you forget about the "$" and the usage of the skill feature in the IDE extension?

~~The skills feature doesn't introduce any new slash commands.~~ It's also still under development at this time.

Edit: I was incorrect about this. The skills feature does add a new slash command, so yes, this should be added to the list.

~技能功能并未引入任何新的斜杠命令。~目前它仍在开发中。 编辑：我之前的说法有误。技能功能确实添加了一个新的斜杠命令，所以是的，应该将其添加到列表中。

**On the contrary, I think there are some issues with adding skills to slash commands.**

Skills should be something that GPT itself decides to use based on whether the current content requires a predefined skill to solve the problem.
If a skill still needs to be manually invoked by the user, then why wouldn’t I just use a simpler and more direct custom slash command to solve the problem and streamline my workflow?

In my opinion, custom slash commands are more like clickable buttons:

• they are explicitly triggered by humans
• they are mainly about timing

This makes them better suited for defining workflows.

Skills, on the other hand, are more like tools, similar to existing ones such as:

• view_image_tool
• shell_tool
• web_search_request

Finally, why did you suddenly introduce a mechanism where prefixing a command with an exclamation mark allows it to be executed without model involvement, while still injecting the command into the context?

This behavior is not intuitive at all and is largely undocumented.

~Hey @etraut-openai , I would like to add /review for consideration in codex VSC extension. This was what I missed when working with the cli vs the extension.~
This would leave, if I summarized correctly:

• /init
• ~/review~
• Command for skill $ (new command)

EDIT: Removed /review, as this was implemented already. User error on my site.

/review is already supported in the VSC extension. It was added about a month ago.

/review is already supported in the VSC extension. It was added about a month ago.

Dang, thank you @etraut-openai for pointing out. Short addition, what went to my misinterpretation:
1) I typed /
2) A new context menu is opening, that is empty.
3) I wrongly thought the / is missing and typed /review --> No commands is displayed.

Maybe this is a UX improvement - but priority very low. :-D And if people are not coming from the CLI they will never notice.

I updated my comment from above.

Agreed with @etraut-openai! Otherwise I'd add that skills are coming to the extension very soon