I see that is backed by ARD. What is difference between using ARD on a remote registry as compared to pulling all resources locally and using agentic search instead?
AgenticResourceDiscovery spec is useful for remote registries but only when we control the server runtime can we make full use out of it, without which in remote registries, it's up to GitHub or skills.sh on what query parameters to expose.
Delegated registry agents run in a throwaway copy of the repo and return a patch. How does that work for big monorepos, is it a full copy each time or something like a git worktree?
Vercel skills.sh? Is this the same
How does the agent sharing work and is it limited to skills? How do I load them?