So it seems vagrant boxes build just fine on GCE instances that have nested virtualization enabled, via a gitlab shell runner.
This means it's possible for us to fully automate the build and deployment process for vagrant boxes. Here's a rough plan:
The stages for building and provisioning the nested virt shell runners could conceivably be run on a shared runner.
I do have some questions though. Is it fine to build/provision these GCE runners using the gcloud cli tool? Or does the cloud team have some infrastructure as code tool that they prefer to use instead in order to avoid vendor lockin?