I’m finding that using skills on Claude or a custom GPT in chatgpt with the same skill file produces very solid code. The skill pack I developed for myself always refers back to the official home assistant documentation and “requires” the AI to validate against that documentation before finalizing any coding.
Like the OP, I’m achieving so much more than I could do by myself, but AI isn’t a “just trust the output” effort by any means.
I‘ve just discovered this thread and the skill pack.
First impressions of the skill pack are very good. I use Gemini and added the repo as notebookLM external source.
A second notebook contained a copy of my main HA setup files and I just asked it to give me feedback on some of my stuff. It looks really promising
Thank you very much
The first real coding session was really good.
Using the skill pack with Gemini thinking mode was a real difference to all my prior attempts.
I created a complex heating schedule based upon school schedules, additional flags and google cal der evaluation.
The results were well structured, without any syntax error and logically 90% valid.
It still helped to be a trained software developer anyway
Glad it was useful. I’m very open to any feedback that makes it better. Some of the decisions are specific to me (like obsessive changelogs, aliases, etc.), but always trying to get better quality quicker.