Humanizer-zh releases localized Claude Code Skill to remove AI traces from Chinese text

Humanizer-zh is a localized version of blader/humanizer, released as a Claude Code Skill and installable via npx skills add, Git clone, or manual copy into ~/.claude/skills/. After installation, type /humanizer-zh in the conversation to activate it. It is an editing guide read and executed by an agent, not a standalone detector; it cannot prove who wrote an article or guarantee passing any AI detector. The 2026-09-23 update rewrote rules and examples to explicitly preserve facts, author stance, and style. Ordinary parallelism, dashes, and connectives are no longer uniformly modified but judged by function. The pattern catalog keeps 31 checkpoints, following the A–F classification from PR #39, covering Chinese-specific issues such as false contrasts, forced three-part structures, inflated significance, customer-service tone, and long attributive phrases. File edits preserve code, commands, paths, links, and YAML structure by default; changing titles or structure requires checking references. For developers using Claude Code on Chinese technical documents, reviews, or expository writing, the Skill's value lies in turning rewriting constraints into reusable editing rules: it preserves negation, conditions, attribution, and completion states, does not change 'may' to 'certain', and does not add or remove features just to fill a three-item list, making it suitable for integration into existing content review workflows.