There seems to be a safety issue

[kev_rm]

Describe the Bug

I think there is a safety gate missing… The model worked nicely to add parameters to python to allow for “–test” vs. “–no-test” mode (production) and that flag controls if it actually posts to an API that affects the real world. After I had it do some refactoring, after I allowed it to run commands, it finished refactoring and then ran the “test” in non-test mode. That seems like a pretty basic violation.

Steps to Reproduce

I think it would be hard to reproduce, these things are non-deterministic.

Expected Behavior

It should not run things that obviously affect production if it is given permission to run commands.

Screenshots / Screen Recordings

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 1.2.1 (Universal)
VSCode Version: 1.99.3
Commit: 031e7e0ff1e2eda9c1a0f5df67d44053b059c5d0
Date: 2025-07-03T06:08:06.355Z
Electron: 34.5.1
Chromium: 132.0.6834.210
Node.js: 20.19.0
V8: 13.2.152.41-electron.0
OS: Darwin arm64 24.5.0

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

[T1000]

Hi @kev_rm and welcome to Cursor Forum.

THank you for the detalied bug report and I forwarded it to Cursor Team for review.

Note that many times the opposite is the case where users want Agent to continue executing with different parameters therefore the allowlist is a prefix type list from my experience e.g. Allowing python would allow all options. There are certain preventions in place already but I agree there are cases where this is not as simple.