I am wondering if there might be a way to allow the Cursor Agent to execute Command Palette commands. This question comes up, as I am constantly having to restart the TS server as the agent adds new code or refactors existing code. This repetitive task is highly mundane, and given the agent (if allowed) could run very dangerous commands at a terminal, I am curious why it cannot invoke command palette commands.
The latest versions of Cursor introduced an allow list for terminal commands. This is a welcome improvement over always ask or always allow, and it has actually saved my ■■■■ over the last few days when a rogue agent (dunno if it was just GPT-5, or all the changes that came with Cursor 1.4) tried to run some very dangerous commands!
I figure if an allow list approach was used with giving the agent the ability to invoke Command Palette commands, that could allow us to develop cursor rules that instruct the agent to bounce the TS server (or similar) on its own, after making changes that might lead to (or verifiably do lead to, i.e. syntax errors for missing types, import files, etc.) issues with the TS compiler.
With an allow list, the agent wouldn’t be able to arbitrarily execute any CP command it wants.
Would be a welcome addition! (Although, fix current bugs first, seems there are some fairly serious bugs that need attention these days.)