Slack Cursor integration declined to implement the request

Bug Reports
1

Where does the bug appear (feature/product)?

Background Agent (GitHub, Slack, Web, Linear)

Describe the Bug

Chat ID is bc-f15c7053-ab1c-4b7f-b7aa-73712c4325ce

My PM sent @cursor a request to make a simple UI change (basically changing the z-index on a tooltip). The Slack bot responded with:

To help me locate the relevant files, could you please tell me which framework the [app] is built with (e.g., React, Angular, Vue, plain HTML/CSS)? Also, if you know the file paths for the orders page or the CSS/JS files related to the table header and the search bar tooltip, that would be very helpful.

Cursor should figure that out itself. My PM should not need to provide that information.

When that information was provided, the background agent printed some generic suggestions about how to solve this category of problem. It did not open a pull request as it usually does, or evidently look at the code to determine a tailored solution.

When I clicked “Open in Web” and tried to click “Send feedback” at the upper right, the button opened a new blank tab but otherwise did nothing. So I am coming here to report this instead.

Steps to Reproduce

Ask @Cursor to do something on Slack

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Slack / Background Agent

Does this stop you from using Cursor

Yes - Cursor is unusable

2

hi @IsaacAtAsh thank you for your bug report.

Before we send the request to our engineers, could you please clarify if the PM has explained where the change needs to be? This is not clear from your post but AI would need to know at least some reference to where the change needs to be.

3

The initial prompt was:

On the orders page of the [app], the table has an anchored header. However, the search bar tooltip is blocked by the header. Fix the tooltip so it appears in front of the anchored header.

with a screenshot attached. That should have been plenty of context to figure out where to fix the issue.

4

Usually it should be sufficient, however it depends also on the model used.

We do not see a systematic issue but I forwarded your issue to the team for internal review.

5

Thanks. The model was GPT-5.

We tried this prompt 4 times, and it worked 1 out of those 4 times, although the other failure cases were more clueless than misbehaved.

6

Thank you for the additional details, do you have any AGENT.md or .cursorrules in your repo that may influence the AI focus?

7

We do not.

8

This topic was automatically closed 22 days after the last reply. New replies are no longer allowed.