You can read it here -
http://cs.iit.edu/~khale/docs/diver-ross19.pdf. It mentions OSv and capstan. The authors lay out ambitious plans to create specialized OSes build and deployment ecosystem - driver.
Regarding capstan, one of the features it is missing is the ability to build images out of the manifest files, the same ones created by OSv ./sccript/*manifest*.py scripts. It should be pretty trivial to use.
I believe one of the authors is on the mailing list ;-)
Waldek
PS. This is one of the interesting statements that strikes a cord in me:
"Despite their benefits, SOSes still face several challenges. The designers must make the decision whether or
not the kernel interface will retain POSIX compatibility
(or binary compatibility with, e.g. Linux), pick the right
abstractions for the target workload(s), and decide on
the right level of protection, among other issues. Specialization for its own sake is not necessarily a good idea,
and as work in the architecture community shows, striking a good balance between domain-specific design and
general-purpose abstractions can pay off [35]. Some of
these design points can (and should) be based on foundational principles, but others require experimentation and
design iteration."