Managing Agent Skills without turning your repo into a junk

I've been poking around Agent Skills lately, and while the utility is obvious, I'm already hitting the same architectural walls I usually hit with any dependency management. It's a classic "how do I not mess this up" situation.

The first headache is the inevitable bloat in the .agents/skills directory. The temptation to dump every single skill related to a tech stack into the folder is massive, but every one of those skills brings a mountain of secondary resources and config files along for the ride. If I start treating this like a "just in case" library, my project root is going to look like a disaster zone within a month. I need a way to distinguish between "essential runtime logic" and "nice-to-have utility" before the file count explodes.

Then there's the versioning nightmare. I'm primarily in the JS ecosystem, so I'm used to npm install handling the heavy lifting. There is an skills package on npm that lets you install stuff directly, but if I'm manually downloading files or pulling from other sources, how do I track state? I'm not about to manually audit local files against upstream sources every time a dependency updates. That’s a massive waste of engineering cycles.

Finally, there's the portability issue. Once I actually build a decent library of skills that work, I don't want to be stuck. If I jump to a new machine or spin up a fresh repo, how am I supposed to sync that .agents/skills folder? Committing the whole folder to Git feels heavy and messy, but relying on a manual Gist of URLs feels like a recipe for broken links and outdated logic.

I'm looking for a more robust way to handle this—something that feels less like manual file management and more like a proper package manager for agentic workflows. If anyone has a cleaner workflow for syncing these or managing the local state without losing your mind, I'm all ears.