Comment by estearum

What are you confused by? You're saying the same thing they said.

They added the additional claim that writing the skills down (apparently) prevents the models from having to self-prompt on the fly and therefore reduces token consumption.