How to make more efficient prompts?

How To
1

Hi all

I’m going through usage request like they’re going out of fashion.

I was/am trying to get Cursor to build a basic blog using Nuxt3 and Directus using cursor and also get it setup ready for pushing to my live server so it’s ready to go with a push of the button.

I’ll start by saying that although I am fairly new to these packages but comfortable enough to get by. I have got them working together fine before I moved over to Docker, which I’ll hold my hands up and say I don’t understand it at all no matter how many videos or write ups I read.

I’ve got Docker working Nuxt and Directus setup but it took a lot of input from me and ignoring what cursor was writing, taking this and that from part of its code and adding my own otherwise I kept going around in circles.

I tried adding the docs for docker, nuxt and directus at the top of the page using (at docker) (at nuxt) (at directus) and I even read about adding (at web) which I also did. I also don’t get the difference between normal and agent modes. I have tried both and to be honest, I don’t get it. I think I dead read someone that agents uses a lot more requests.

Now that Nuxt and Directus is up and running, I am now trying to build a blog connecting the two but I seem to have to do a lot of manual work again.

I read posts about people creating iOS apps or websites with very little effort so I have to put this down to my prompts, I don’t know how to ask cursor.

I’ve tried things like:


You will need to help me with my Nuxt and Directus project and check for any errors or security floors based on the latest docs I have attached. I need your help setting up a development and production environment. When running on my local machine, I need to be on a development environment but when put onto my remote server I need to be in a production environment.

I wish to use all the features built into Nuxt3 unless there are better benefits on offer using 3rd party but ideallly stick with what is built into Nuxt3.

First you will ask me about my project and what libraries or securty features I require until you believe you have enough information to complete the task.

Once done, you will ask me for my full guide on my current setup, this is a long file so It may go over your limit, you need to let me know and allow me to post the guide in multiple messages until i’ve given it to you in full. Once done, you will rewrite my guide in markdown allowing me to copy and paste it into my markdown editor. My guide seems a very manual process creating directory and files i.e for nginx so if there is a better way then please update my guide and provide me with the required files.

You will then provide me with the updated code in full which will allow me to edit my files and create the project.

It works to a degree but it pretty much rebuilt my guide and not the files.

Can anyone help me with getting better and more efficient results?

I read that the differences between normal and agent modes are

Agent mode

  • Can complete tasks by itself using the command line.
  • Can determine which files are needed as context by itself.
  • Know what should do in next step.

Normal mode

  • Cannot use the command line.
  • Requires the user to specify the context.

• Wait for your command for next step.

So when do I use agent and when do I use normal?

2

Use agent mode when:

  • You need composer to perform command line actions.
  • You’re not sure what files need to be edited (or would rather the agent decide for you. Note that even if the agent gets it right, this may result in more requests and ‘noise’ in your context window).

I read posts about people creating iOS apps or websites with very little effort so I have to put this down to my prompts, I don’t know how to ask cursor.

While this is true and is possible, a couple of points:

  1. Content creators are incentivised to exaggerate how easy it is (more clicks/views).
  2. The complexity of the project matters.
  3. The user’s skill level/expertise in what they are building matters.

That is to say: the less you know about what you’re doing, the harder it will be for you to achieve a complex goal using composer or similar AI tools.

This is not to say you can’t or shouldn’t try. I think you should.

But when starting to reach outside your understanding, I think it’s better to slow down, ask the AI for more information on the areas you’re unclear on, and better understand the problem and your code base.

Not only will this help you solve the problem, but it will facilitate a better understanding of your code base and the language/libraries you’re working with. Both of which are useful.

In my experience, the more I know about the language/project I’m working on, the more I can get out of these AI tools.

1 Like
3

Thanks, that’s good advice. I’ll ask what I don’t know rather than asking for the world.

1 Like
4

Great to hear.

AI is amazing at building things for you. But it’s also the best learning tool that humankind has ever had access to.

Because it can answer your specific question (as opposed to an online article that is “kind of” relevant), is infinitely patient, and can cater to your individual learning stage.

It makes learning Docker bearable :wink:

1 Like