moai-domain-humanize

Run a humanization pass that detects catalogued AI tells—formulaic connectors, cliché openers, inflated significance—and rewrites only the flagged spans. This Skill supports Korean, English, Japanese, and Chinese with per-language tell catalogues and severity tiers.

It detects ten prose categories including focal vocabulary like delve and tapestry, negative parallelism such as "It's not just X, it's Y", rule-of-three tricolons, signposting fillers, transition overuse, cliché openers, vague attribution, and formatting abuse, plus nine marketing-copy tells.

No. Meaning preservation is the top rule: an Invariant Ledger records facts, numbers, names, dates, and certainty levels before editing, and a Delta Audit verifies survival afterward. Any drift on a supplied fact triggers a rollback of that edit.

Yes. Copy mode uses per-language copy-layer catalogues for headlines, CTAs, taglines, and landing pages, replacing the change-rate guard with a fact-anchor preservation guard. A design-copy module adds genre rules for landing pages and slide decks.

Automated AI detectors are unreliable, especially on CJK polite registers, so this Skill is a pattern-based editing tool rather than a detection oracle. Change-rate percentages are LLM estimates, not computed metrics, and are reported as such.

In prose mode, edits exceeding an estimated 30% of the text trigger a warning and cap the grade at C, while exceeding 50% forces a halt and human review. Borderline estimates are treated conservatively as over the threshold.