Get live statistics and analysis of Elie Steinbock — oss/acc's profile on X / Twitter

Since I open sourced my .mdc rules for Cursor I've been getting a lot of questions. Here are all the answers in one place: 1. Isn't 20 mdc files too much information for Cursor? No. This is what .mdc files solve. Cursor Rules file is now deprecated. You no longer provide all your information in one go, but in small chunks (.mdc files). Each of my .mdc files is <50 lines. 2. Do you need to manually tag Cursor? There are 4 options for rules: a. Always - the rule is part of every chat b. Auto attached - e.g. auto attach whenever you're chatting about ".tsx" files, or "tailwind.config" file. c. Agent requested - the agent can request to read the file. This is what I use most of the time. The agent decides whether to request a rule based on the description you set for the rule d. Manual - only applied when tagging it. I mostly use c. And I'll often tag the a rule file for Cursor myself as I'm not certain it will always read it otherwise. 3. How do I write the rules? I have a cursor-rules.mdc that writes rules. Easiest way to do this: * Share a snippet of code with Cursor (command-shift-L on relevant text) * Tag the .cursor-rules.mdc file * Ask Cursor to create a new .mdc file for this pattern Pro tip" Cursor a little bit buggy here. I prefer to do this in Ask mode and copy the result into an mdc file. Sometimes you'll see Cursor show empty files when I do this in Agent mode. 4. Does it work? Yes. Watch it in action at the end of the video (linked in next tweet). If you let the AI do its own thing it will invent new patterns. A lot of the code we write is the same thing over and over (e.g. add a form, make an LLM call, add a server action that updates the db, etc.). There's no one right way to write code. But in your project there is. You want the code to be consistent. The code for each form should look more or less the same. Rules help Cursor do this effectively. (Tagging other code examples in your codebase is another way to do this, but .mdc is cleaner and less noise). 5. Is it worth the effort to set up rules? Yes. It's a minimal, one-time effort to set up a rule. Especially once you have a base to work off of. And they provide value forever. Cursor Agent is my no 1 employee. Giving it the context it needs to be as effective as possible is well worth the effort. 6. What rules do you use? So far my rules can be broken down into a few buckets: a. Code patterns (e.g. add a form, server action, or test) b. General knowledge (e.g. project structure) c. Features (e.g. this is what the inbox cleaner does and how it works) Most of my rules fall under "code patterns". But I'm steadily adding more "feature" rules. I'm often repeating myself over and over to LLMs. I might be using Claude or Anthropic and I can copy paste in these notes now. And of course I can tag these notes in Cursor too. If you want to see my rules in action, check out my YouTube channel or Inbox Zero GitHub. Links below.

I open sourced my Cursor .mdc rules! Here's how to use .mdc's to be 10x more productive 👇

Single best Cursor Rule I have: 1. Tell Cursor the feature you want implemented 2. Tag the rule (@ task-list.mdc) 3. Watch the AI work its magic. Don't believe me? Copy and paste the rule into your repo and see how you get on. Link below.

I added a project structure .mdc for Cursor. It explains the core project structure with comments for important folders. Very helpful for an LLM (or human) to get a quick grasp of the entire project. Link to rules below.

Don't waste time writing Cursor rules. AI generated 90% of mine.

Here are my "meta" Cursor rules: 🥇 cursor-rules.mdc - an mdc to create other mdc's. Super useful. 🥈 task-list.mdc - how Cursor should manage a todo list file and handle tasks one at a time, marking them completed as it progresses. 🥉 cursor-rules-analytics.mdc - tells Cursor to update an analytics .md file each time it uses a rule. Fun experiment, but will delete this one soon. All my rules are open and available at the link below.

In my last tweet I described a flow that massively improves Cursor by giving it a task list and memory file to work with. I've now converted this into a .mdc file to speed up development. Link to the repo with the full task-list.mdc file below 👇

Here's the no BS guide to creating Cursor Rules. I've been asked this a lot recently. So here's a 6 minute vid on how you can create them (and that it's okay to not have rules). Link to the repo in the first reply.

Here's how to add memory using a Cursor rule. Memory is a good way to stop Cursor making the same mistakes again, by committing them to memory (the learned-memories.mdc rule). You could use a simple Cursor rule to achieve a similar goal, but this works well as a quick scratchpad without having to create a full rule for it. Link to full code example below 👇

Big news. I'm now a Cursor Ambassador. Any suggestions or questions for the Cursor team? Send them over and I'll get them over.

Who works Sundays? Share a link. Love to see

Watch out for this scammer on TrustMRR. We did a project for him a few years ago, and he didn't pay.

Do founders think it's cost-effective to pay for these ad slots? I've never seen less clickable ads in my life.

Been following Sappy Seals for a while. Got a little friend for my pengu NFT

Got an mfer a few days ago. Is mfer follow mfer a thing?

Since I open sourced my .mdc rules for Cursor I've been getting a lot of questions. Here are all the answers in one place: 1. Isn't 20 mdc files too much information for Cursor? No. This is what .mdc files solve. Cursor Rules file is now deprecated. You no longer provide all your information in one go, but in small chunks (.mdc files). Each of my .mdc files is <50 lines. 2. Do you need to manually tag Cursor? There are 4 options for rules: a. Always - the rule is part of every chat b. Auto attached - e.g. auto attach whenever you're chatting about ".tsx" files, or "tailwind.config" file. c. Agent requested - the agent can request to read the file. This is what I use most of the time. The agent decides whether to request a rule based on the description you set for the rule d. Manual - only applied when tagging it. I mostly use c. And I'll often tag the a rule file for Cursor myself as I'm not certain it will always read it otherwise. 3. How do I write the rules? I have a cursor-rules.mdc that writes rules. Easiest way to do this: * Share a snippet of code with Cursor (command-shift-L on relevant text) * Tag the .cursor-rules.mdc file * Ask Cursor to create a new .mdc file for this pattern Pro tip" Cursor a little bit buggy here. I prefer to do this in Ask mode and copy the result into an mdc file. Sometimes you'll see Cursor show empty files when I do this in Agent mode. 4. Does it work? Yes. Watch it in action at the end of the video (linked in next tweet). If you let the AI do its own thing it will invent new patterns. A lot of the code we write is the same thing over and over (e.g. add a form, make an LLM call, add a server action that updates the db, etc.). There's no one right way to write code. But in your project there is. You want the code to be consistent. The code for each form should look more or less the same. Rules help Cursor do this effectively. (Tagging other code examples in your codebase is another way to do this, but .mdc is cleaner and less noise). 5. Is it worth the effort to set up rules? Yes. It's a minimal, one-time effort to set up a rule. Especially once you have a base to work off of. And they provide value forever. Cursor Agent is my no 1 employee. Giving it the context it needs to be as effective as possible is well worth the effort. 6. What rules do you use? So far my rules can be broken down into a few buckets: a. Code patterns (e.g. add a form, server action, or test) b. General knowledge (e.g. project structure) c. Features (e.g. this is what the inbox cleaner does and how it works) Most of my rules fall under "code patterns". But I'm steadily adding more "feature" rules. I'm often repeating myself over and over to LLMs. I might be using Claude or Anthropic and I can copy paste in these notes now. And of course I can tag these notes in Cursor too. If you want to see my rules in action, check out my YouTube channel or Inbox Zero GitHub. Links below.

I added a project structure .mdc for Cursor. It explains the core project structure with comments for important folders. Very helpful for an LLM (or human) to get a quick grasp of the entire project. Link to rules below.