Make AI-Written Chinese Sound Like a Real Person With human-writing

AI-written Chinese has a common flaw: it reads smoothly, but you cannot say who wrote it. human-writing (活人感写作) is a general-purpose writing and revision skill, usable out of the box, that makes the model write like a specific person — someone who knows things, has judgment, occasionally digresses, and comes back. It covers most Chinese writing situations: Zhihu answers, WeChat official-account posts, blogs, forum threads, people stories, explainers, reviews, fiction, and scripts for spoken pieces.

Why This Skill Matters

Before any prose, the skill settles a prerequisite: do you have something to write with. For real-world topics, missing material means research it, ask the user, or shorten the piece — never pad with circular phrasing. Fiction may invent characters and plots freely, but every scene still needs a goal, an action, and a change.

Material cleared, it manages three things: facts (verify numbers, quotes, and lived experience in reality writing; check character, action, and causality in fiction), momentum (every paragraph adds something new — a new fact, action, example, or consequence — and never repeats what was written), and Chinese itself (plain speech as the base, attention to word order and pauses, and removing report-speak, model-speak, and rhetorical setup-then-reversal sentences).

Installation

Send this to your agent:

帮我安装这个skill：https://github.com/KKKKhazix/human-writing

The agent reads the repo, finds human-writing, and completes the install; the skill then shows up as 「活人感写作」. If your agent cannot install directly, download from Releases or copy the human-writing folder from the repo — keeping that folder name — into your skills directory at ~/.agents/skills/human-writing/.

After installing, invoke it like this:

使用 $human-writing，把我的材料写成一篇有活人感和中文韵律的作品。

Real Workflow: Turn Raw Notes Into a Publishable Post

1. Gather your material first — numbers, quotes, what you actually saw — then hand it over with the target medium.
2. Ask for the draft; the skill checks your material's factual boundary before writing, and tells you what is missing instead of papering over it.
3. After the draft, ask for the revision pass. The skill goes paragraph by paragraph: cutting repeated explanations, tuning long-short sentence rhythm, and blocking colon abuse, dashes, and the 「不是……而是……」 pattern of AI boilerplate.
4. For a hard rule check, run its check script against the finished draft — scripts/check_prose.py only enforces the rules already written down; it does not decide your style.

Tips

• Version 1.1.0 changed the defense from literal strings to the underlying move: banning "plant a misconception the reader never had, then overturn it" as an action, whatever words it wears — because models paraphrase their way around string bans.
• Keep the revision reference (references/revision.md) in mind for the post-draft checklist; other references cover forum prose, reality boundaries, fiction rules, and special formats like spoken scripts.
• For chat windows without skill support (ChatGPT, Qwen), the repo ships a distilled version under 2,000 characters — dist/human-writing-lite.md — you can paste directly.
• If the script flags a normal Chinese word, check the changelog: 1.1 removed words like 「不丢人」 and 「打法」 from the false-positive list.

When Not to Use This

It is built for Chinese writing; nothing in the README addresses English output. And if you need tightly formatted, structured output — a spec, a table-heavy report — keep in mind that the style pass targets voice and rhythm, not document structure.

See the leaderboard for more skills.