I see that people are using Custom Modes and the docs say Alt + Enter gives the option to use a skill as a custom mode, but I don’t see such option on my Linux machine with latest Cursor version (both in IDE and Agents Window)
Hey @Yoan_Babulkov, thanks for the report!
Custom Modes work on Linux the same as on other platforms, there is no platform restriction. A few things to check:
-
Custom Modes are available in the Agents Window, not the IDE
-
They are created from skills. In the Agents Window, type / in the chat input and use the arrow keys to highlight one of your skills. While the skill is highlighted you should see a hint reading “Alt+⏎ to Use as Mode”, and pressing Alt+Enter turns it into a mode. You need at least one skill (a SKILL.md with a valid frontmatter block) for this to appear, since the option only shows on skill entries, not on regular commands.
One Linux note: some desktop environments bind Alt+Enter at the window manager level. If you see the hint on the highlighted skill but the shortcut does nothing, check your system keyboard shortcuts!
If you still do not see it, can you share what shows up when you type / in the Agents Window chat input (do your skills appear in the list?), plus your exact version?
I found the hint (it’s really well hidden haha). And also it worked this time in the Agents Window, oh well, thanks.
Some additional questions:
- Why is it available only in the Agents Window? Is it planned to be added to the IDE as well?
- How does it exactly work? If i just use a normal skill, will it not be in the context for the next message? And this mode makes it so that it persists the whole chat?
Hey @Yoan_Babulkov, glad you found it!
Custom Modes were designed and built specifically for the Agents Window, so that is where the feature lives today.
When you invoke a skill normally with /, its instructions are attached to that one message only. They stay in the conversation history like anything else you’ve sent, but the agent treats them as a one-time instruction, so nothing tells it to keep applying the skill on later messages and its influence fades as the chat grows.
Using a skill as a mode makes it persistent. On the first message the skill is attached and the agent is told to follow it until you exit or switch modes. On every following message the agent gets a reminder that it is still in that mode and should re-read the skill if it has lost track of the rules. When you exit the mode, the agent is explicitly told to stop following it.
Thanks, Colin!
It’s a neat feature, hope it becomes available in the IDE in the future as well.