question about agent forwarding

Although OpenSSH does not appear to fully implement
https://tools.ietf.org/html/draft-ietf-secsh-agent-02 it does seem like it
implements that draft document for agent forwarding.

Assuming that that's correct than I have a few questions. If you have to
"shell" channels and want to be able to do agent forwarding on either
channel that you'd have to do the "auth-agent-req" channel request for each
one of those channels?

I saw one implementation that created a channel for which a single channel
request was ever sent - auth-agent-req. That channel was never closed and
then another channel - a "shell" channel - was created. And forwarding
seemed to work for that channel. ie. when an attempt was made to connect to
another machine through the shell a channel open request for a "auth-agent"
channel was sent to the client. Of course I'm thinking that this particular
implementation just works by dumb luck and that it is, in fact, an
incorrect implementation. Is that correct?

Thanks!