I came across this post so I decided to contribute (of course, I am no expert and Nick might correct me if I am wrong). As RT is inherently dependent on your chromatography parameters, I believe that you will always need to provide some benchmark signals, like iRTs, for a prediction tool to work and to provide a RT prediction in minutes. However, if you don't have any spiked standards in your sample, you could provide a set of any peptides from your sample. All you need is a number of peptides (8 - 20) for which you have good IDs and signals, then you could create a "normal" RT-predictor in Skyline from them and provide it to Prosit as a reference. If you have a sample of eukaryotic origin (such as HeLa etc.), Prosit might be able to work with the so-called CiRTs, common iRTs which are peptides shared across eukaryotic species. You can read more about CiRTs here (Parker et al., MCP 2015):
Disclaimer: I have actually never tried this with Prosit myself, with real data in Skyline this would quite certainly work. However, CiRTs are included as a ready-to-go set within Skyline, so you could automatically add them to your document and hope that you will be able to find them in your real data. Then, build a spectral library including your actual peptides of interest, for which you want RT predictions, and choose CiRts (or any other predictor with "iRTs" of your choice) for the library's set of standard peptides.
I hope that this is not too confusing, feel free to ask again!