• Publish PR to GitHub Packages

Hmm. I'm not really sure about this -- dev-time discovery of ATs from LIBRARYs would also work and have no effect in prod. As would dev time use of an AT that's just not bundled, or not at the default location and not declared in the mods.toml. And... etc.; I feel like at some level this isn't really necessary? After all -- if you're providing an AT, why can't you just use a mod instead of a GAMELIBRARY? The two act basically identical -- in fact, all that's needed to make a given GAMELIBRARY a mod is (a) remove the FMLModType manifest entry, (b) provide a mods.toml in the same source set with only a license, mod ID, and version (plus loader info on older MC versions), and call it a day. There's no case where this is a non-trivial change that I'm aware of. I feel like if you're using any "mod" feature, like an AT, your thing should probably be a mod to begin with?

Also as an aside, I'm not entirely against this -- after all, mixin configs are still loaded from GAMELIBRARYs for some reason? And consistent behaviour in this regard feels sensible -- but if this is implemented can we please just include the game libraries in the original loop instead of duplicating the whole thing? Feels a bit wasteful.

As noted on discord, making gander use mods instead of GAMELIBRARYs is a ~30 line change or so, so yeah, at least in that case that seems like a cleaner solution -- ATs are a mod feature after all, this solution feels kinda like, I dunno, letting @EventBusSubscriber work in GAMELIBRARYs.