Configuring Cursor with custom instructions

Lessons learned 10/11 on Cursor after 15 years in software

Once you've identified the main prompt formulation issues with Cursor, it's time to consolidate your working environment.

Cursor's memory

Cursor has a memory area where it adds frequently recurring information. For example:

• What type of DB you use
• How you name your tests
• Your commenting preferences

This allows it to adapt to your style gradually.

Getting ahead with custom instructions

If you don't want to repeat the same things multiple times — especially in writing — you can get ahead and define your own instructions in settings.

These instructions will be added by default to each of your prompts and stored as a .mdc file in the .cursor/ folder, so committable for your colleagues.

In this file, don't hesitate to be verbose: you'll save a lot of time for your future instructions.

Configuration example

The stack is Svelte for the frontend, Go for the backend, DB is PostgreSQL

Business logic is centralized in backend/business/
Frontend calls are centralized in frontend/src/lib/api.ts

Use objects rather than functions
Refactor as duplications appear
Avoid components that are too large by creating sub-components when needed
Follow SOLID principles as much as possible

Why it matters

Without these instructions, you'll spend your time correcting the AI. It will choose different conventions with each conversation, mix styles, and force you to repeat the same instructions over and over.

With a good rules file, the AI starts each conversation with the right context. It knows your stack, your conventions, your preferences. It's the difference between a new intern every time and a colleague who knows the project.

Invest 15 minutes to write your rules. You'll save hours.

---

Originally published on LinkedIn.