This is pretty nifty and I could see it being quite handy for building one-off pipelines. The whole storing a diff of your change and reapplying it in the pipeline is super clever, though I'd have concerns about how the diff algo would handle different types of source input.

It was only after squinting for a bit and mentally parsing the example code that I had the a-ha moment and understood when/why I'd want to use this though. Maybe a video or asciicinema/terminalizer recording showing how it works in practice might help illustrate