[BUG] Feedback After Cursor v0.50.4

Bug Reports
1

Hi everyone :waving_hand:

I’ve been vibe coding quite a bit lately and ran into some usability hiccups after the recent update. Thought I’d document them here in case others are facing the same issues—or if the Cursor team wants examples to look into.

Below is a breakdown of what I’ve observed along with the workarounds I’m currently using.

.mdc Rules & Agent Behavior

:lady_beetle: Issues

  1. Missing basic Markdown editing behavior
    Inside .mdc files, standard markdown behaviors like pressing Tab to indent a line don’t work.
  2. Cursor Editing Features don’t work
    I can’t use common actions like Cursor Tab or Cmd+K inside .mdc files.
  3. No partial accept of agent suggestions
    When the agent suggests edits to .mdc files, there’s no way to accept only some changes—I have to accept all or nothing.
  4. File references don’t resolve properly
    Using inline tags like @todos.mdc doesn’t work. Cursor Agent fails to resolve them and sometimes literally looks for a file named @todos.mdc :thinking:
  5. Duplicate files created unexpectedly
    When the agent updates a rule, it occasionally creates a duplicate file with the same name in the project root instead of modifying the original in .cursor/rules.
  6. Metadata not applied on rule creation
    If I ask the agent to create a new rule (e.g., tech_stack.mdc) with Rule Type=Always, it creates the file and fills in the content correctly, but it skips setting the Rule Type. I have to go in and set it manually every time.

:hammer_and_wrench: My Workarounds

  • For 1 & 2:
    I temporarily rename the file to .md, make my edits, then switch it back to .mdc.
  • For 4:
    I avoid using tag-style references (@file.mdc) and instead use the full relative path, e.g., .cursor/rules/todos.mdc.

:brain: gemini-2.5-pro (Thinking) Agent

:lady_beetle: Issues

  1. Agent stops prematurely
    Sometimes, the agent just… gives up. It’ll read a file or run a pre-step and then stop before getting to the main task.
  2. Fails on large outputs
    If the output grows too long, the agent stops halfway and doesn’t pick back up or provide a continuation.

:hammer_and_wrench: My Workarounds

  • For 1:
    I add a single character (like .) to the end of the prompt to “nudge” the agent into action. Works, but burns credits unnecessarily.
  • For #2:
    I revert to an earlier state, start a new chat, and give the same prompt again.

If others are seeing the same, I’d love to hear your workarounds or ideas. And if the Cursor team has suggestions—or needs more context—happy to share!

Thanks for building something this useful :raising_hands:

1 Like
2

100%… its gone brain dead since the latest update, pretty much unuseable. My experience is even with the most simple tasks its attempting to re-structure non-related code, creating whole new components for no reason, deleting & failing at re-factoring, importing non-sense that doesnt exist etc etc. it was working really well prior to the latest update. ill likely cancel my sub if this keeps up

1 Like