Is anyone else experiencing the agents running python to review and edit your files?
This started a couple of days ago. I’ve not changed any rules or skills and I’ve not instructed it to use python to edit the files. The agent no longer works autonomously. It just writes something like this and waits for me to hit run, then it will do a similar statement but to write code to the file.
Recent Cursor updates removed the “Command Allowlist”, so now all host-level terminal commands (like `python`, `pytest`, etc.) are gated behind an approval layer (even with Auto-Run enabled). Auto-Run now only applies to internal agent actions, not your system shell.
To restore autonomous execution, set Auto-Run Mode to `Auto-Run in Sandbox`. This allows the agent to run Python inside Cursor’s sandboxed runtime without prompting.
Hey Colin. Below is my version information. It happened on multiple models: Composer 1.5, Codex 5.3 High. It hasn’t happened today. Next time it happens I’ll grab a request ID.