eighttrigrams/us-vs-them provides line-level provenance for text under agentic editing, answering "who wrote this line, us or them?" by deriving authorship from version history rather than file markup. Available as a library or CLI tool, it addresses a critical problem: human-authored text should be treated as sacred, while agent-generated "slop" remains editable.

The tool identifies "islands" of human-authored lines within a "sea" of machine-generated text using simple diffing on git history. Each revision carries a provenance marker, enabling evaluation without special markup. Output shows ranges with scores: 1.0 for fully human, 0.46 for partially agent-modified, and 0.00 for fully agent-authored.

Usage requires bbin for local install. Run `us-vs-them --ours [email protected] README.md` inside a git repo to classify authorship. The `--ours` flag defines humans (shorter list); `--theirs` defines agents. Passing both is rejected. Behavior details are in `caution_test.clj`.