Pipelines can accept as many input files as you like, but it should be possible to run with as few as possible.

For example, pipelines should auto-generate missing reference files, where possible. So given a reference genome Fasta file a pipeline would build the reference index files. The pipeline should also be able to optionally accept the reference index files in this case, if available.