I always wondered if one could feed history into some abstract parser (or nowadays, maybe a LLM) to extract the most frequent "idioms":

    ls <dir> | grep <pat> | less

vs

    ls ~ | grep .bak | less
    ls .config | grep *rc | less

you get the idea