If you have many samples, want to do an in depth analysis and have enough time I personally would do the following:
1) Get a basic set of RefSeq or Ensembl Bacterial genome assemblies covering all taxonomic groups. Either take all of them or perform a clever sub setting (like taking representative genomes). You can do the same for Fungi and Protists if you think you might have them in the sample. Or other Eukaryotes. I guess Archaea are not necessary, but there are not that many so you can as well add some of them. You can remove genomes from surveillance projects, as many of them can be virtually identical.
2) With those you do a first screening of all samples to roughly identify what you have. I would use something like Kraken2 or Centrifuge.
3) Now extend the groups you find with closely related genomes from RefSeq and maybe GenBank. Be careful with GenBank genomes, they could have wrong taxonomic annotation. Maybe check this. For Bacteria, Fungi, Protists and what else you have, you can do the same. Kick out the Genomes where you do not see any hits. Maybe add some/all viral RefSeq genomes.
4) Now do another screening and check if you seem to have a good representation of what is there. You can check the reads which did not get any hits with Blast or so to get indications what you are missing.
5) Improve your DB a bit more if necessary
6) Enjoy :)