Just for kicks, here's a (terribly ugly) pipeline that will display the contents of .nls and `ls`, with duplicates removed:

ls | cat .nls - | column -t -s: | sort -r | perl -ane 'print unless $x{$F[0]}++'

and then obviously just echo "dirname: description" >> .nls to append to the file.

I would be interested in seeing a better pipeline, one that doesn't need perl/awk/ruby/what have you to remove the duplicates correctly.

The join(1) command is designed to do exactly this:

join -a2 -t: <(sort .nls) <(ls | sort)

(I think this is different from your pipeline, in that your pipeline will print files in .nls that aren't actually in the directory.)

Awesome.

Oh hey, this is awesome! I thought about doing something like this but then opted for the (ultimately more flexible) python script.

Is fine too, but you wont get non-annotated files. nls was designed to work within your normal workflow, i.e. doesn't require you to think about getting file notes in order to get them.