KubeSat: An autonomous framework for satellite constellations and swarms

I don't know enough about this to comment, but is there not an argument to be made that satellites should run the least code possible? I picture a misconfigured Kubernetes setup irreversibly taking out an entire fleet of satellites.

The mission sounds cool, and I understand its application:

> KubeSat is an open-source project for building a cognitive, autonomous framework for satellite constellations and swarms.

But couldn't the cognition layer run at a ground station? The ground station would then control the satellites using a limited control layer that's as robust and bug-free as possible. Debugging a bad Kubernetes setup in space sounds like a nightmare.