Hello Shangzhong,
I'm not sure about the format of NCBI files, but there are two things I've learnt about how cufflinks handles annotations:
1. You can take only fields labelled "exon" in column 3 of the GTF file
2. You should make sure that gene_id and transcript_id are set correctly for each exon, and that they cover all of the features described in the annotation.
E.g. you can parse the file for "exon" only entries, then sort and find the number of unique gene_ids and transcript_ids
If that matches the number you would expect from the annotation (e.g. from counting unique "gene" and "transcript" entries), then you are good to go.
Many other features (CDS. UTR, start/stop codon, etc) are irrelevant for expression quantification, so there's no need to worry about them.
I hope this helps
cheers
-- Alex