Good news @jessica.song! I think I solved this. 
First, thank you for sending me the PDF describing the protocol. I noticed a key statement:
- pairs of primers (Fw-Rev or Rev-Fw) had to be present in the sequence fragments
That is, this particular sequencing facility must expect mixed orientation reads. 
Therefore you need to enter in both primers for each of the --p-front-* commands (in 5' - 3' orientation). 
Like so:
qiime cutadapt trim-paired \
--i-demultiplexed-sequences paired-end-demux.qza \
--p-cores 4 \
--p-front-f CCTACGGGNGGCWGCAG GACTACHVGGGTATCTAAKCC \
--p-front-r GACTACHVGGGTATCTAAKCC CCTACGGGNGGCWGCAG \
--p-overlap 3 \
--p-error-rate 0.1 \
--p-match-read-wildcards \
--p-match-adapter-wildcards \
--p-discard-untrimmed \
--o-trimmed-sequences primer-trimmed-demux-2.qza \
--verbose > cutadapt-log-2.txt
The output from the first sample looks like this:
=== Summary ===
Total read pairs processed: 133,428
Read 1 with adapter: 133,350 (99.9%)
Read 2 with adapter: 132,063 (99.0%)
Pairs written (passing filters): 131,987 (98.9%)
Total basepairs processed: 77,399,028 bp
Read 1: 38,699,572 bp
Read 2: 38,699,456 bp
Total written (filtered): 71,548,932 bp (92.4%)
Read 1: 35,774,760 bp
Read 2: 35,774,172 bp
-Cheers!
-Mike