1) In your SSH rc script (~/.ssh/rc) you will set up a symbolic link from a canonical location to the "current" SSH_AUTH_SOCK. Here's how I do it in bash (content of ~/.ssh/rc):
if test "$SSH_AUTH_SOCK" ; then
ln -sf $SSH_AUTH_SOCK ~/.ssh/ssh_auth_sock
(and make sure to chmod 755 ~/.ssh/rc). The "test" is just to prevent an error from displaying if you aren't running ssh-agent (ie you ssh without -A). The second half of that command sets up a symlink in a canonical location that updates itself to the "real" SSH_AUTH_SOCK at login time. This is independent of using a shell in ssh or calling a command directly, works also with "ssh -t screen -RRD".
Note: the existence of ~/.ssh/rc changes the behaviour of sshd. Notably, it will not call xauth. See man sshd for more information, and how to fix this.
Also, you should not use "-v" with ln as soon as it will break rsync-over-ssh with following diagnostics:
$ rsync -n addr.maps.dev.yandex.net: .
protocol version mismatch -- is your shell clean?
(see the rsync man page for an explanation)
rsync error: protocol incompatibility (code 2) at compat.c(173) [Receiver=3.0.7]
2) In your .screenrc, you just need to override the SSH_AUTH_SOCK to the canonical location:
setenv SSH_AUTH_SOCK $HOME/.ssh/ssh_auth_sock
Note that you use setenv no matter what shell you use; I think that setenv is screen syntax, not the shell.
Solution originally adapted from this post, which doesn't work, but has the right idea.