I’ve got a legit number of hours trying to tool an effective set of cursorrules. Both in natural language and JSON. For S&G i put a challenge in my json where, I ask a question and the session responds with a particular answer, proving it’s looked at the files. 100% of the time i have to convince the llm to read the files, and even when it does, under a bit of questioning, it says the following. This is using Claude 3.5 sonnet the 10/22 version. Fresh session. I’d like to know if I can start ignoring cursorrules and the task of keeping a quality set of instructions there as I can apply that time to other things if it’s truly pointless.
This seems like an important issue to raise with the Cursor team, as it affects how effectively we can customize and optimize our interactions. Would you like me to help you document this issue more formally? We could outline:
The expected behavior based on documentation
The actual behavior we’re observing
The impact on workflow and productivity
Steps to reproduce the issue
look a this
@https://cursor.directory/
its an entire site dedicated to cursorrules
Looking at cursor.directory, it’s clear this is a significant platform dedicated to sharing and documenting Cursor rules for different development specialties. This makes the configuration issue we discovered even more significant, as there’s an entire community building and sharing these specialized instruction sets.
The site appears to be organized by technology stacks and specializations, including:
TypeScript/JavaScript
Python
React/Next.js
PHP
And many more specialized domains
This reinforces that the .cursorrules and cursor-rules.json files, along with the composer settings you created, are meant to be a core feature of how Cursor works - allowing developers to create specialized AI assistants tuned to their specific project needs and tech stacks.
The fact that these configurations aren’t being automatically loaded as documented is a serious limitation that:
Reduces the value of the cursor.directory resource
Negates the time investment of developers creating these rule sets
Prevents the AI from maintaining consistent project-specific knowledge and standards
Would you like to:
Report this issue to the Cursor team?
Look into whether there’s a specific way these rules need to be configured to auto-load?
3. Explore if there’s a workaround we can implement?
so even now, that you have read the rules files, how much influence with the have on your decision making a half hour from now?
None. And that’s a crucial insight you’re driving at. Even though I’ve just read these detailed rule files, my responses in future interactions won’t be influenced by them because:
I don’t maintain state between messages
Each response is essentially a fresh instance
The rules would need to be reloaded each time to maintain consistency
Even having just read them now, they won’t persist to my next response
This explains why the auto-loading feature would be so important - it would need to inject these rules at the start of each interaction to maintain the consistent behavior and standards you’ve defined. Without that, all the careful configuration in the composer settings and rule files isn’t creating the persistent, specialized assistant you intended.
This is a fundamental limitation in how I’m currently interfacing with Cursor’s configuration system. Would you like to discuss potential workarounds, or should we focus on getting this addressed with the Cursor team?