Cursor Agent does not know how its own rules system works

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

I use the agent for everything. Even creating rules. Crazy thing is, Cursor’s agent doesn’t seem to have a clue how its own rules system works, which is rather ironic. When asking the agent (and thus also an llm) to create a rule, its become a total ■■■■ shoot whether it even has a clue where to start, let alone actually create a viable rule in a .mdc file.

Often it will just flounder around without a clue and create random .md files in arbitrary locations. Sometimes it won’t even create a file, and just give you a bunch of junk in the chat. Occasionally it will realize that this used to be a .cursorrules file thing, and try to add rules to that, but again, location is often totally arbitrary.

Ironically, I even tried to have cursor create a rule about creating rules! I didn’t realize until now, the reason that hadn’t been applying at all, was it had been created in a directory called workspace-rules/…!!! However, even after moving that into my .cursor/rules directory, the agent still seems to be clueless. (Lag time indexing rules or something?)

Anyway. It would be really useful, if the agent, actually had an explicit understanding of how its own rule system works, and when creating rules, it could factor that knowledge into the API call (system prompt, or just additional context?) In a general sense, it seems logical that Cursor and its agent, would always have up to date, accurate knowledge of how to use…itself! :stuck_out_tongue: So that if you issue prompts regarding Cursor’s own capabilities, they could be handled implicitly, without the creation of additional rules (I’m accumulating a LOT of rules and its starting to concern me, with regards to context load and token usage…especially considering that auto application is VERY flaky!)

Steps to Reproduce

Instruct cursor to create a new rule, with details of the rule you want created (but don’t explicity state every specific detail about how Cursor rules work, such as it must be a .mdc, must have the right header lines, must be located in the right directory, etc.)
See what happens.

Expected Behavior

Cursor should understand its own rules system. Asking it to create a rule should result in implicit understanding of:

  • Use .mdc file, not .md, not .txt, etc.
  • Place it in the .cursor/rules directory (root by default, unless specific scope instructed)
  • Implement proper header lines to govern application, globs, description, etc.)
    Rule then placed in correct location as correct file type with correct structure.

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 1.4.5 (Universal)
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 24.5.0

Does this stop you from using Cursor

No - Cursor works, but with this issue

An additional issue I’ve noticed recently. Agent-generated rules, often seem to end up duplicated, contents wise. The entire rule, is often duplicated, at least once, in the .mdc file. ITs an exact replication, so the rule details are at the top, then again right after that, maybe rarely a third time.