Human-writing is an agent skill that makes Chinese text generated by AI read like it was written by a specific person rather than a machine.
The tool addresses a fundamental problem with AI-generated Chinese writing: it often reads smoothly but lacks the distinctive voice and personality of human authorship. The approach works by enforcing a structured writing process that begins with material preparation, then manages three core aspects during composition. For factual writing, it verifies facts, numbers, quotations, and firsthand accounts; for fiction, it checks character consistency, actions, and causality. During drafting, each paragraph must advance the narrative with new information, actions, examples, or consequences without repetition. For the language itself, the tool prioritizes conversational tone while paying attention to word order and pacing, removing report-style phrasing, model-generated patterns, and rhetorical reversals. After the initial draft, a revision pass checks for circular reasoning, eliminates redundant explanations, adjusts sentence rhythm, and blocks common patterns like colon overuse, em-dashes, and formulaic constructions that signal AI authorship.
The tool suits writers working across most Chinese writing scenarios including Zhihu answers, WeChat articles, blogs, forum posts, personal narratives, popular science, reviews, fiction, and spoken content. It is designed as a ready-to-use skill that integrates with agent systems, though it can also be manually installed by copying the skill folder to a local directory.
The project shows active development with recent updates that refined its approach to removing AI characteristics. Rather than maintaining lists of forbidden phrases, the latest version targets the underlying rhetorical patterns that readers recognize as machine-generated, such as the structural opposition found in constructions like "you thought X, but actually Y," which appears across multiple surface-level variations.
