I have discovered this kind of solution.
You just have to fill in flat .csv filename into csv readers and writers.
- MetaDataMaster.grf - runs the graphs below.
- MetaDataCreator.grf - creates metadata according to csv header and write it into meta_example.fmt file
- MetaDataUser.grf - Reads csv according to created meta_example.fmt file - you can add there a reformat and use just some predefined fields.
You can run the 2nd and 3rd graph separately to test it.