An approach that uses substr, map and grep rather than regexen.

use strict;
use warnings;

open my $inFH, q{<}, \ <<EOF or die $!;
 > >2L type=chromosome_arm; loc=2L:1..23011544; ID=2L;  dbxref=REFSEQ:
+NT_033779,GB:AE014134; MD5=bfdfb99d39fa5174dae1e2ecd8a231cd; length=2
+3011544; release=r5.54; species=Dmel;
CGACAATGCACGACAGAGGAAGCAGAACAGATATTTAGATTGCCTCTCAT
TTTCTCTCCCATATTATAGGGAGAAATATGATCGCGTATGCGAGAGTAGT
GCCAACATATTGTGCTCTTTGATTTTTTGGCAACCCAAAATGGTGGCGGA
TGAACGAGATGATAATATATTCAAGTTGCCGCTAATCAGAAATAAATTCA
TTGCAACGTTAAATACAGCACAATATATGATCGCGTATGCGAGAGTAGTG
CCAACATATTGTGCTAATGAGTGCCTCTCGTTCTCTGTCTTATATTACCG
CAAACCCAAAAAGACAATACACGACAGAGAGAGAGAGCAGCGGAGATATT
TAGATTGCCTATTAAATATGATCGCGTATGCGAGAGTAGTGCCAACATAT
TGTGCTCTCTATATAATGACTGCCTCTCATTCTGTCTTATTTTACCGCAA
ACCCAAATCGACAATGCACGACAGAGGAAGCAGAACAGATATTTAGATTG
CCTCTCATTTTCTCTCCCATATTATAGGGAGAAATATGATCGCGTATGCG
AGAGTAGTGCCAACATATTGTGCTCTTTGATTTTTTGGCAACCCAAAATG
GTGGCGGATGAACGAGATGATAATATATTCAAGTTGCCGCTAATCAGAAA
TAAATTCATTGCAACGTTAAATACAGCACAATATATGATCGCGTATGCGA
GAGTAGTGCCAACATATTGTGCTAATGAGTGCCTCTCGTTCTCTGTCTTA
TATTACCGCAAACCCAAAAAGACAATACACGACAGAGAGAGAGAGCAGCG
GAGATATTTAGATTGCCTATTAAATATGATCGCGTATGCGAGAGTAGTGC
CAACATATTGTGCTCTCTATATAATGACTGCCTCTCATTCTGTCTTATTT
TACCGCAAACCCAAATCGACAATGCACGACAGAGGAAGCAGAACAGATAT
EOF

do { my $discard = <$inFH> };
my $seq = join q{}, map { chomp; $_ } <$inFH>;

close $inFH or die $!;

my $ct      = 0;
my @kmersGG = do {
   my %seen;
   grep { not $seen{ $_ } ++ }
   grep { q{GG} eq substr $_, 21 }
   map  { substr $seq, $_, 23 }
   0 .. length( $seq ) - 23;
   };

printf qq{>crispr_%d\n%s\n}, ++ $ct, $_
   for @kmersGG > 1000 ? @kmersGG[ 0 .. 999 ] : @kmersGG;
[download]

The output.

>crispr_1
CATTTTCTCTCCCATATTATAGG
>crispr_2
ATTTTCTCTCCCATATTATAGGG
>crispr_3
TATTGTGCTCTTTGATTTTTTGG
>crispr_4
GATTTTTTGGCAACCCAAAATGG
>crispr_5
TTTTTGGCAACCCAAAATGGTGG
>crispr_6
TTGGCAACCCAAAATGGTGGCGG
>crispr_7
ACGACAGAGAGAGAGAGCAGCGG
>crispr_8
AAATCGACAATGCACGACAGAGG
[download]

I hope this is of interest.