Input: a long piece of text you paste in, plus a window size (how many sentences per pass) and an overlap (how many sentences repeat between passes).
Outcome: the text split into overlapping windows, each with its own ready-to-copy extraction prompt — you run each one externally, paste the notes back, and get a final prompt that stitches every window's findings together.
Useful for: working through documents too long to trust to one pass, the same way a sliding window scans a long DNA sequence without ever needing the whole thing in view at once.