Highlighting code and using "Add to Chat" into Agent chat is really bad all of a sudden

Describe the Bug

When you highlight code and press Command+L to send it to the Agent chat, the agent very often doesn’t understand that we are focusing on the highlighted code.

Steps to Reproduce

Start a session with an agent. Some time into the session, highlight some code and send it to Agent chat with Command+L. Either add some comment, or not, and then submit.
The agent will not understand to focus on the highlighted code as in previous versions.

Expected Behavior

The Agent should understand that when you highlight code and send it to chat, it should focus on this part.

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 1.4.5
VSCode Version: 1.99.3
Commit: af58d92614edb1f72bdd756615d131bf8dfa5290
Date: 2025-08-13T02:08:56.371Z
Electron: 34.5.8
Chromium: 132.0.6834.210
Node.js: 20.19.1
V8: 13.2.152.41-electron.0
OS: Darwin arm64 24.6.0

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

Hey, thanks for the report. Could you let me know which model you’re using? Additionally, does creating a new chat resolve the issue?

Hey! “Add to Chat”-functionality has become less reliable with all models, but some more than others.

GPT-5 seems to be performing the worst. Gemini and Claude models are worse too, but not to the extent of GPT-5.

Opening a new chat can help in some cases, but also doesn’t necessarily “fix” the problem.

The models in general seem less aware of what in the context to focus on than in earlier versions, especially in the case of “Add to Chat”.

(Side note, I don’t have any rules that should affect this behavior imo)

Thanks for the additional info, we’ll look into it.