For point 1, the best AED cutoff to use is quite arbitrary. For one the last genomes that I annotated we had a set of high quality genes identified based on synteny with genes in closely related genomes. We plotted the distribution of AEDs for those genes and found that a cutoff of 0.28 captured 98% of the high quality genes. This value would vary based on the evidence provided. I’ve used 0.5 in the past as a more permissive filter.
For point 2, these is a accessory script in the MAKER bin called quality_filter.pl. It has an option (-a) that allows you to put in an AED cutoff and it will filter the gff3 file based on that cutoff.
For general processing of GFF3 files, there is a perl library called GAL that is useful if you write code in perl.