based on https://stackoverflow.com/a/48509425/2654518, which is based on
http://rabexc.org/posts/pitfalls-of-ssh-agents
Current code has a few issues: depends on `SSH_AUTH_SOCK` and
`SSH_AGENT_PID` env variables, which might not be available in every
shell session; and tries to create a new socket for agent-forwarding by
checking `SSH_AUTH_SOCKET` instead of `SSH_AUTH_SOCK`.
Also, it's safer to create the env file with 066 flags and in the user
home directory.
