I would generalize and propose a customizable pattern definition that could work both ways: to define the naming of newly downloaded files (pdfs) and to help importing those already existing into the library.
It should be flexible at several levels, namely:
- pattern definition: ex: some beginner mode vs. regex for advanced features
- flat or directory based
- optional different formats for different types of sources (journals, books...) or media (pdf, html, video...)
- warn about files not indexed because of not fitting in any of the defined patterns
- more than identifying duplicates, trying to merge the files (ex: a pdf) into more complete entries one might already have (those nearly completely filled based on the journal web page but that usually lack the main article itself)

I also stand for a partnership, as both interfaces approaches are good and useful. Indeed there are occasions (usually the harvesting phase) I prefer a browser plugin and others I'd rather use an all-in-one program. But so far none alone has convinced me completely and my library is less tidy and useful than I wished it to be...