Skipping of arbitrary jobs if certain jobs are completed/failed

Summary of the feature

Is there a possibility to skip arbitrary jobs if certain dependencies are fulfilled, e.g. if a previous job has failed? If I understand the "skippable" keyword correctly, this only works for members/chunks.

Describe your use case

I would like to check if some required input files are already present. If yes, start the simulation, if no create them from scratch.

My intuitive approach uses three jobs:

A: check if input is there (lightweight, can run e.g. on the login node of the cluster)
B: create the input data (some computational power required, runs in the batch system)
C: simulation

With the straight dependencies A -> B -> C. However, if A succeeds, I can just skip B, because there is no need for B anymore.

Describe the problem

I see four solutions for this

Make a configurable switch for the user to use existing or create new files. Quite elegant for the resulting workflow, slightly more complicated for the user.
Combine jobs A and B into one job, that does the check. However, this introduces some overhead of submitting a job to the batch system that does barely anything if the files do exist.
Make job A fail if the files do not exist and set the dependency A -> B to the status FAILED. With a weak dependency of C on both A and B, C will be started as soon as A is successful or (if not), after B is successful. But B will always be started, just to the quit without doing anything. More complicated than 1 with few benefit in runtime.
>> new feature <<: make a dependency A -> B "if and only if A has FAILED" and combine it with the weak dependencies as in 2.

Is there an alternative way to do something like this, or is this already implemented and I did not find out how?