I shared this so everyone can see it. I is not a canonical nucleotide, and as far as I know q2-feature-classifier may not be set up to handle it. (Although you won't be able to build a tree, that step in Sidle won't be able to handle the I.)
There's a conversation about this issue here:
you might also check the BOLD database that's described here that will give you more ideas.
I'm sorry I dont have any COI experience, so I'm not sure about this particular issue.
To extend @jwdebelius's response, in particular to the I nucleotide.
Yes! In fact you can read one of my old posts here:
I do not think you have too many degeneracies. If anything, you'd be more likely to run the risk of extracting more spurious sequences than extracting less sequences. Given that these are the same primers as reviewed in Porter et al., from Gibson et al., Folmer et al., I'd think they are fine.
Alternatively, many researchers remove primer sequences prior to submitting to online repositories. For example, if they amplified and sequenced the same region as you, then they most likely removed the primer sequences before doing so. Which is a good thing. However, this means that PCR primer searches will not work to extract a given region.
You can try the extract-seq-segments approach of RESCRIPt to extract your amplicon region from sequences that might be missing the primer sequences.