I know that ssh-add is "front-end" of ssh-agent.
But ssh-agent on my computer is already running (I could find it in top). When I type ssh-add, it said "Could not open a connection to your authentication agent".
How ssh-add communicate with ssh-agent in details?
My situation is
#! /bin/sh
# hello.sh
eval $(ssh-agent -s)
ssh-add
Then,
bash ./hello.sh
when I type
ssh-add some-new-key
in terminal,
It output that error message
The output of ssh-agent -s is some environment variable assignments, something like SSH_AUTH_SOCK=blahblah; export SSH_AUTH_SOCK etc. When you run eval $(ssh-agent -s), the shell executes that as code, and those variables get set in that shell. The variables there contain the information ssh-add needs to contact the agent, and they get inherited down from the shell to the ssh-add process.
But here, you're running it from inside hello.sh. The shell running the script is an independent process, distinct from the upper interactive shell that started hello.sh, and the variables don't get inherited "upwards".
Instead, if you source the script, with source hello.sh, or . hello.sh, it runs in the same shell, and the variables get assigned properly.
Though, if you're running multiple shells (multiple terminal emulators, SSH sessions, screen/tmux windows, whatever), you really only need one ssh-agent. You'll have to save the variable assignments to a file somewhere, and load them from e.g. .bashrc. But I don't know what exactly you're doing.
eval $(ssh-agent)bash- source it instead:source hello.sh