Question: How is your approach better than just recursively traversing package.json files? Seems like you're downloading a lot of stuff onto your server when you could keep it all in the client.
I'm not recursively traversing package.json, just making a graph out of a single repository. External packages are only referenced by name, not on their internals.
Reading the whole repository has the advantage of making links between files in that same repository visible, instead just an opaque mass of packages depending on packages.
Also, recursively fetching thousands of package.json files will sometimes (most times, I would argue) be more resource-intensive than just downloading a single repo.
Finally, it's been done on the server because that way the generated image can be placed on GitHub's README.md.
