Summary When Enable auto-run mode is turned on in Cursor v0.49.5 under Agent chat, commands not included in the Command allowlist still require manual confirmation (⌘+Enter). Users expect all commands to auto-run when the feature is enabled and the allowlist is empty.
Environment
Cursor Version: 0.49.5
Platform: macOS (Ventura 13.x)
Mode: Agent Chat (Default new chat mode = Agent)
Steps to Reproduce
Open Cursor Settings → Features → Chat.
Check Enable auto-run mode .
Clear all entries in Command allowlist (so it appears empty).
Initiate an Agent chat and ask it to run a shell command (e.g., ls -la ).
Expected Behavior With Enable auto-run mode on and an empty allowlist, any generated commands should execute automatically without requiring manual confirmation.
Actual Behavior Commands still open in an editable prompt and require pressing ⌘+Enter (Confirm) before execution.
Workaround Manually pressing ⌘+Enter for each command or explicitly adding every command to the allowlist.
Impact Reduces automation efficiency and forces manual steps in workflows that should be fully automatic with auto-run enabled.
Suggested Fix Allow commands to run immediately if auto-run is enabled and the allowlist is empty, without manual confirmation.
As a user I have to disagree hard on this, unless there is a specific setting for your case like Allow ALL commands without confirmation and I agree to be responsible for any and all damages then an empty allowlist should never allow automatically any command to be executed even in Auto-run mode.
Auto-run mode can be reasonably filled with several sufficient commands already.
If there are common commands like test ..... or node ... you can just set the begin of command test or node to allow list if you trust it and do not need additional confirmation. You would still be responsible if things go wrong.
You do not need to add every single command to allow list.
user should have this option.
in the early version, this is the default behavior.
lots of time, agent is doing debug work, impossible to white list one by one…
I need an option to while list all, and of cause, I would be responsible for any and all damages.
thanks for understanding.
When the agent is debugging, grep log, restart service, apt install, cat, echo, etc, it is impossible to predict what action it will do. And manually confirm will only slower the process.
@T1000@deanrie when this feature will be released?
And before that, I need to downgrade, and what’s the latest version which support fully auto run?
and I agree that such a setting in Cursor to allow ALL commands would be useful.