My thoughts exactly. I also kept cross-hardware work in mind when testing this idea. Rather than giving a private ssh key to your claude worker and telling it to work across two machines, allowing two separate workers access the same underlying unified memory/context might go a long way
This is a really cool idea. In terms of interoperability, FLUX should probably warn users that FLUX specific features may not apply when sending it outside it's protocol, but should allow users to send it anyway.