CD-HIT will remove all sequences that share identity above a certain threshold. If you set that threshold at 1.0 (meaning 100% identity), it will remove all identical sequences and retain only one representative:
seqkit grep can exlude a list of records via ID/name/sequence/sequence, common sequences here.
# common sequences, one record per line
seqkit seq --seq --line-width 0 common.fa -o common.fa.txt
# output dir
mkdir -p uniq
for f in dir/*; do
seqkit grep --by-seq --only-positive-strand --invert-match \
--pattern-file common.fa.txt $f -o uniq/$b
The method above compares sequence by the whole bases/amino acids (exact match), you may also use clustering methods, which might be more reasonable for proteins.