I'm a bit behind on my machine because I require Haskell's dbus package for my setup to work thus there is this downgrade.

node2nix should come from nixpkgs, but out of scope for this issue :)

Agree. I originally copied this script from some other place but as you can see it uses nix-shell -i bash -p nodePackages.node2nix so system's version of <nixpkgs>. I'm not sure how can this be fixed at the moment but it seems to me it would be better if this used node2nix from repository itself.

something like:

ROOT="$(realpath "$(dirname -- "$(readlink -f -- "${BASH_SOURCE[0]}")")"/../../../../..)"
$(nix-build $ROOT -A nodePacakges.node2nix --no-out-link)/bin/node2nix ...

This is the biggest hack and I can't really explain why is needed.

Probably since it's needed for elm-test, which is installer here as a dependency?

That's the case of elm-verify-examples as well and in that case the writeScriptBin works as expected. I have a bad feeling that this somewhat depends on how PATH is being defined and potentially in case of this package the version from node_modules/.bin takes precedence over the script. That would be very unfortunate behavior. If that's the case we need a better way to patch this.

binwrap package also provides library functions which can (and are) called from packages code (and hopefully should just check the binaries are already present and do nothing). I think if we would try to avoid installing that package in any way we would need to do a lot of patching in source files. One thing that comes to mind is to let npm install the library but remove its bin files.

Some part of me always wants to start crying every time I start considering doing hacks like this but so far I don't have idea about how to solve these issues in more robust way.

Yeah most probably need to check $PATH precedence for node_modules, but I think this is good enough to work.

just a bit of sorting