Transcription
If you were using Claude code, and you think downloading more GitHub repos is going to make it work better, then you need to stop what you're doing and listen. Your file structure is 10 times more important than just giving Claude more crap to use.
The problem is that Claude is getting lost in the sauce. The issue is never the tools or how you were communicating with Claude. It's the actual map that Claude's using to see the information. So, more tools equals more surface area to break. But, with the right file structure, all your tools fit in the right place, and Claude knows when and how to use each of them.
And the biggest problem that I've seen is most people complain that Claude code just doesn't listen to rules. And the best way that I can explain how this works is that each folder is a quest marker. And every Claude.md file is a waypoint. You're giving Claude access to predictable paths. Everything is separated by intent, and now every new plugin, agent, skill inherits this map the second that they spawn into your workspace.
So, one, your context will improve, you'll spend less tokens, and you'll get quicker answers. Secondly, all of your new agents inherit this new map. And most importantly, every new plugin just becomes a multiplier, and not just more crap to add to the pre-existing pile. But, don't get it twisted, because agents and skills still matter. And that's only after the foundation is built. So, keep this in mind. We're only going to use plugins to scale what's already working with Claude.
If this was helpful, follow your boy, and comment AI agent if you want access to all of my free Claude code resources.