Nicely done!

I do a pretty similar version of this with command-| in textmate, but I've wished for/thought about building a command like this to speed up the process!

The textmate approach does have the advantage (or perhaps disadvantage, for large enough inputs!) of emitting immutable "views" of the data at each step, rather than (presumably?) re-evaluating the pipeline each time, which is nice if the steps of the pipeline are the slow part, or to go back 2 steps and start a new pipeline. Maybe a potential future flag? :-)

Automatically emitting an `up<N>.sh` script is super clever too!