I tend to follow the same principle as many stompbox-manufactorers (so there is any consistency) and at least have an output cap in my builds. Further more I tend to isolate pots with coupling caps whenever possible. Would that be a good starting point?
Yes, if there is any chance of DC across them during normal operation then it's a good idea. Generally speaking I include caps at input and output - Just In Case.
Speaking of this, it reminds me that I've build a marshall lead 12 preamp where there's no couplingcap existing at all, none in the whole design, and it works fine. Probably due to the fet-opamps in it.
Fully DC coupled designs are quite possible; it's just a case of "God help you" if you get some rogue DC on the input.
This seems to be mainly a fetish of the Hyper-Fi crowd who have a fairly irrational objection to capacitors in the signal path (yet are happy to include them in the power supply, forgetting that it is in the signal return path). Some caps do
indeed have some undesirable traits, disk ceramics for example can be microphonic, but generally speaking when it come to audio, concerns about caps are pretty grossly overcooked.
You will note that most solid-state power amps are DC coupled, and while they generally have a few caps for gain setting and bootstrapping there are usually none in the direct signal path except right at the input. A major exception of course are amp that run on a single (rather than split) supply rail and have a thumping great electrolytic coupling the output to the speaker.
Fully DC coupled makes an amp harder to design (considering the accumulation of voltage offsets and thermal drift) and harder to service.
I was wondering about how designs with different supplies and reference voltages can match? Like for example a preamp with a dual power supply and a reference of 0V and stompboxes with a single supply and a reference of say 4,5V, or stompboxes with different reference voltages. It would seem to me that they would mess with each others reference voltages and could cause a decrease of headroom and unwanted rail clipping (asymmetrical).
Do couplingcaps also take care of this problem?
Yes and yes.
Your reference point is AC "ground" but if you have a single-ended supply, such as a 9V battery in a stomp, it may not be DC ground; then you have to have DC blocking caps somewhere near the input and output or you won't be able to bias input stages, and the standing DC offset from the output stage may well cause problems downstream.
When you use split rails, such as with op-amps, then the DC conditions can be arranged to center on ground the same as the AC conditions, so you can get away with not having any DC blocking caps (or more correctly, you can design them out).
You normally bias an amplifying stage so that its output is half way across the available supply, so as to maximise headroom. If one side of the supply is ground than this "Q point" must be some non-zero voltage. But when you have both +ve and -ve supplies the idle Q-point can be at ground or zero volts and still be half the available voltage swing.