I like https://github.com/carmonalab/UCell. It calculates a score between 0 and 1 for each marker for every cell based on the ranked expression of these markers and in my hands it worked reasonably robust to get an idea which clusters harbor which celltypes. Including some negative controls make sense, so say you have PBMCs you would include some fibroblast marker signature to see what "background" scores you would get for a celltype not present in your data. The scores you "trust" should then be much higher than that background.