I'm enjoying this series, thanks very much!

This really seems like the most challenging thing with vibe coding. I am also wondering how to best utilize instruction files. For example, Laravel Boost gives a pretty detailed set of instructions. I am trying out the livewire starter kit for example, and I am using VS Code with GitHub Copilot. It adds this other thing it calls "skills", which are feature-specific instructions.

It seems to me that whenever you are vibe coding on an existing codebase, all of this context might in fact produce worse results? For example, say you want to work on something related to backend logic and queued jobs or whatever. For that particular task, you really don't have any need whatsoever of providing context regarding how the frontend works, i.e. if tailwind or livewire is being used.

Is this significant? Or does this not impact the context too much too matter? Will the agent understand that information does not apply to the task at hand? (I am now referring to the instruction files which as I understand are always provided to the agent.)

In your video you add the very basic declaration: "Wender is an AI design app". I wouldn't expect to have to add this to every new chat. That seems like a piece of information that belongs in an instruction file instead.

Sorry if I'm rushing ahead, perhaps you address this in a later video in the series!

Prompting Tips