I was wondering if it would be helpful if I posted in this forum the plans the AI gives for resolving the issues.
From a non-developer point of view I see it as useless for the most part.
Not only it has no understanding of the codebase, but you personally using it can’t offer anything that developers who understands the codebase can’t do better themselves and they can even use the same LLMs.
The one exception is when there are interminent issues that can’t be reproduced by maintainers and you are able to narrow down the specific steps to reliably reproduce it.
I think this is a good approach. AI implementations by someone who has no knowledge of the codebase is just adding more work on the maintainers.
The good thing about stash is in its plugin support, and it genuinely think that LLMs has given us a lot of functionality and niche stuff that we would never have without LLMs because real developers would need to work on it. But best to keep it to the plugins.
That’s a good idea, developing plugins with AI and only adding them to the main project if they are used by enough people and are already debugged. I also think it would be a good idea to restrict the github issues to maintainers and reach consensus about what to change in this forum.
Don’t think plugins ever need to be merged into the source. If an idea is good enough it should probably be rewritten in the correct standard by a human.