Humanizer Stack

Humanizer Stack is a pair of Claude Code skills for writers and editors revising public-facing AI drafts. It checks sentence-level habits and the shape of a piece, rather than limiting edits to word substitutions.

• Phrasing review: The surface skill checks promotional wording, vague attribution, inflated explanations, punctuation, and repetitive sentence patterns, with guidance drawn from Wikipedia's Signs of AI Writing.
• Structural review: A second skill examines overexplained themes, overly tidy narratives, descriptions of emotion, vague references, and how the writing addresses its reader.
• Comparison across drafts: Shape checks compare a piece with earlier writing to catch repeated formulas that vocabulary edits can leave intact.
• Local scanners: Python scanners provide deterministic checks for patterns they can match and can fit into automated review workflows.

The package works through Claude Code; the Python scanners run locally. They cover only part of the editing work. Cadence, formulaic structure, and polished but empty filler still need editorial judgment, and the package doesn't promise undetectable writing.

The structural guidance draws on StoryScope's study of long fiction. Applying those findings to short nonfiction is an inference, so the same checks don't carry equal weight across genres. Included upstream material has MIT and CC BY-SA 4.0 terms, including share-alike obligations for the latter.