I have developed a heuristic for working with Large Language Models:

If you are incapable of one-shotting your LLM prompt it is an indicator that you need to break your problem up into smaller portions

LLM’s are great at implementing solved problems within a small context.

Think: modules that implement code you’d already find in api documents or on stack overflow, not complete projects.

Trust me: you don’t want the LLM inventing anything.

I have build a few tools for helping with this:

npx pochade-js {name of new component}

For generating custom HTML elements

npx pochade-obsidian {name of new plugin}

(to be run in your .obsdian/plugins folder) for generating new Obsidian.md plugins

and:

npx pochade-node-red {name of new plugin}

For developing node-red plugins.