I would submit that the point of the GUI was necessarily "anti-keyboard" as more "pro-discovery", plus adding gestures that would be difficult on a keyboard.

Keyboard interfaces can be, but not necessarily, quite opaque. There's a reason in the old days companies shipped keyboard overlays and function key templates to help users (or went all in with custom keyboards built for the application).

In contrast consider something like Wordstar where you pretty much just needed to memorize the three prefix keys, and if you pressed one and waiting, you'd get a menu describing all of the options.

Obviously, things like early Smalltalk and, probably, Xerox (having never used any of the early Xerox systems), relied heavily on the mouse, but even it had command accelerators.

The CUA standard was that crossover that Windows, and MOTIF, adopted. Much of that work is still with us today, and it worked well both on GUIs and TUIs (witness the old Turbo Pascal/C++ IDEs, and other applications). Those character based applications had to assume there was no mouse, but offer the flexibility of being good citizens should a mouse be present.

This can happen if the applications exposes single keys as shortcuts. If they are "hidden" behind the different modifiers (i.e. Ctrl, Cmd, Alt), then mis-clicks should not be possible or should be more tolerated.

The "intuitiveness" argument is also related. You cannot rebind well-known shortcuts to different actions and expect the user to not get frustrated.

Hm, you should generally be able to hit Return to close a message box on Mac. Or, more specifically, to activate the default action in a dialog (the button whose background is the tint color of the application). Spacebar activates the button that has the focus ring around it, which may or not be the default button, but you can use Tab to move the focus around.

At least, this is the way it's always been on OS X and later. I'm not sure about classic Mac OS.