How do you chain several actions to one tap?

The phone runs the chain

Each step is performed by the phone in turn, using whatever the automation system exposes.

That is different from a scene in a home system, where the system performs the group and the phone only asks. A chain lives on the phone and a scene lives in the system.

The practical consequence is that a chain is personal to that phone and a scene is not, and that a chain can mix things a home system knows nothing about.

Where the steps are all home devices, a scene is the better structure. Where they are a mixture, a chain is the only option.

The weakest step decides

A chain of five steps where four are certain and one needs a network is a chain that fails whenever the network does.

A step requiring confirmation stops the whole sequence until somebody responds, which defeats the purpose of a single tap.

So the reliability of the chain is the reliability of its worst member, and adding a step can only reduce it.

List the steps and mark which ones can fail. That list is the design.

Failing halfway

This is the failure mode that matters, and it is worse than nothing happening at all.

Nothing happening is obvious and you do it by hand. Three of five things happening is invisible, and you now have to establish which two did not.

So put the steps most likely to fail first, where their failure is visible, rather than last where it is silent.

And prefer chains whose steps are independent, so a missing one does not leave the rest in a strange state.

Designing a chain that holds

Three or four steps. Beyond that, the odds of one failing on any given day stop being negligible.

Only steps you want every single time, since a conditional step is what makes people abandon the whole routine.

Confirmation visible at the end, so you know it completed rather than assuming.

And a manual route left intact for every step, because a chain that removed the ordinary way of doing something is a chain whose failure has no fallback.