Did you know that this phrase in Cursor’s agent security policy doesn’t refer to the workspace that’s being used but the whole file system?
Reading files and searching code don’t require approval.
...
Like my previous article related to this, where I benchmarked codebase search speed, this isn’t suitable for vibe coders. These articles were prompted by Cursor’s increasing dismantling of non-agent workflows. This time around, I will cover the overall experience.
...
This article isn’t relevant for vibe coders. It’s not about agents but about getting fast chat responses to difficult questions.
...
Versions used: Node v22.11.0, LLRT v0.4.0-beta
Note: As of writing, the LLRT creators say that LLRT is to be used for evaluation purposes only!
...
A while back, I had to create a test like that for a React hook (but the example below applies for anything that consumes a stream with native fetch). But all I was able to find were articles on testing Node.js streams (i.e. backend perspective instead of frontend) and the following packages that didn’t work for me: ...
At least with vscode. There might be other tools available but that’s the particular IDE that I’ve evaluated.
The extension that makes it possible
...
I probably won’t copy all of my articles but I’m not a fan of the fact that I can’t choose not to paywall things anymore.