A Universal Writing Skill That Helps You Produce Text Without the 'AI Smell'
Core Highlights
The WeChat public account "Digital Life Kazike" open-sourced a universal writing skill called "Living-Person Writing.skill," with a single core goal: stripping away the lingering "AI smell" from text generated by AI. Rather than piling on ornate vocabulary, it forces the model to speak like a human, using real cases, real emotions, and a real rhythm to produce copy with a tactile sense of everyday life. Put simply, this is a "counter-machine-tone" filter bolted onto large models, making machine output read more like a real person's essay. The project arrives at a moment when audiences have grown sharply intolerant of templated, soulless machine prose, and it responds to a genuine, widely felt pain point among Chinese creators who worry their output is instantly recognizable as machine-made and therefore dismissed on sight by readers who have developed a keen nose for it. The timing is no accident: as generation gets cheaper, authenticity becomes the scarce currency that separates content people finish from content people scroll past.
What Happened
The Skill works in two layers. On the content layer, it encourages users to supply their own real experiences and emotional material, letting the model expand from concrete details instead of falling back on generic templates. On the phrasing layer, it explicitly bans a batch of high-frequency AI tics and jargon, such as stiff connectives, excessive parallel antitheses, and overused internet terms like "empower," "closed-loop," and "handle." Stacking the two layers, the resulting text's voice clearly sits closer to a real human, no longer carrying that uniform, correct yet hollow machine feel. In practice, writers report that drafts produced with the Skill read as if typed by a colleague in a hurry rather than by a committee of robots, which is exactly the effect they were after when they reached for the tool in the first place, hoping to sound like a person rather than a brochure. The second layer does the heavy lifting, because most AI tells are matters of rhythm and word choice rather than of facts.
Technical Details
In terms of implementation, it is essentially a carefully designed system-prompt and constraint rule set, with embedded examples and a negative taboo list. The Skill has been adapted to mainstream domestic models such as Qwen 3.8 Max, DeepSeek V4 Pro, and Kimi K3, showing that its prompts have cross-model portability and are not tied to a single vendor. It can be used directly in domestic agent products like WorkBuddy and Qwen Office, lowering the barrier for ordinary users, who can get started without any programming background. Because the rules are plain text, a user can open the file, read the forbidden phrases, and tweak them to match their own voice in a few minutes, which is part of why the community has embraced it and begun sharing their own forks tuned to specific niches such as parenting blogs or technical tutorials. That portability is what keeps the Skill useful as models churn underneath it.
Comparison with Competitors
Most off-the-shelf "de-AI" solutions rely on post-processing polish or simply swapping models, treating the symptom rather than the cause. Compared with plugins that only do "tone fine-tuning," Living-Person Writing corrects the model's expressive inertia at the level of writing paradigm, more like training a "plain persona." It is open-source and cross-platform; against closed-source paid writing assistants, it is friendlier to Chinese creators and easier to modify, and the community can keep adding new taboo words and samples. This openness also means the Skill improves fastest where the users are, rather than waiting on a vendor's release cycle that may never prioritize the nuances of Chinese everyday speech, where a single stock phrase can betray the machine instantly. The contrast with paid tools is less about quality and more about who controls the evolution of the voice.
Industry Impact and Use Cases
In scenarios that heavily depend on "authenticity," such as content creation, self-media, and e-commerce copywriting, removing the AI smell directly determines conversion and trust. The opening of this Skill lowers the cost for ordinary people to produce natural Chinese, and also reflects that the domestic agent ecosystem is moving from "usable" to "good to use and human-like." For operators and writers, it is a handy deodorizer always within reach, and it reminds the industry that what matters more than parameter scale is the ineffable human warmth in the text. As AI writing floods feeds, that warmth may become the only durable differentiator left when every competitor has access to the same underlying models and the same flashy feature set. The strategic implication is clear: in a world of infinite draft text, the brands that win will be the ones readers still believe are written by someone who cares. It is a small, practical answer to a large, vague anxiety about the soullessness of machine text, and its popularity suggests that writers care about voice as much as they care about speed.