When attempting to followup with a cloud Agent, all repository-level skills, commands, and rules are unusable.
Steps to Reproduce
In Agents window, run a cloud agent
When following up with agent, attempt to use a top-level repository-based Skill
Expected Behavior
I would expect to be able to use Skills, Commands, and Rules that I spent the last year fine tuning for my workflow before Cursor 3 botched this all up.
This is at least the 6th or 7th critical breakage in my well-established workflow in Cursor 2, using Cursor as an IDE. Was there any consideration for quality control on Cloud agent-based workflows before forcing your user base onto the Cursor 3 Agents window??
Hey, thanks for the detailed report. I see the screenshot. This is a known issue where slash commands skills/commands don’t show up during follow-up chats with cloud agents in Glass UI.
We fixed a similar bug for initial cloud agent runs, thread: Skills in Cloud Agent, but the follow-up case is still broken. The team is aware. No ETA yet, but your report helps us prioritize it.
As a temporary workaround, try describing the skill you want in your follow-up message, for example, “use the fix-flaky-tests skill”, instead of using the slash menu. The backend loads skills onto the VM for each step, so the agent should still have access to them even if the UI doesn’t show them in the slash menu.
