Plainspoken Writing skill
Ten skills for Chinese-language text: finds signs of machine writing across 25 dimensions and edits only on your command
- 75 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skills read your whole text, and the draft lands in the agent session history
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/plainspoken-writingDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skills read your whole text, and the draft lands in the agent session history
This is third-party code. Review the repository files before installing.
What it does
Plainspoken Writing, in Chinese Baimiao, breaks down which parts of a text read as machine-written and by default only shows them. The author starts from the observation that readers notice predictability: the next sentence can be guessed from the previous one. So the check runs over 25 signs with a concentration grade, and rewriting is enabled only when you ask. The set has seven core and three auxiliary skills arranged in layers from content to presentation. They include `voice-dissolver` for voice calibration, `fact-review` for fact checking, `chinese-write-checker` for density and logic, `plainspoken-writing` for the check itself, `editor-revisor` for cutting filler, `packaging-workshop` for headlines and delivery, and `style-blade`, `scene-fit` and `punct-polish` for style, platforms and punctuation. The skills work as plain prompts. The Python engine in `core/` and the rules in `rules/` are optional and serve batch checks and tests. Installation is `npx -y skills add taxueseek/plainspoken-writing -g --all` or copying the `skills/` folder.
Who it is for. Authors and editors who write in Chinese and want to see where a text sounds machine-made before changing anything.
Good fit when
- A draft reads smoothly but lifelessly and you need to see which phrases give away a template
- Facts, figures and quotes in an article need checking before publication
- Headlines, delivery and punctuation need polishing before a piece goes out
- You want to calibrate your own voice first so editing does not flatten it
Not a fit when
- The text is not in Chinese: the rules and signs are built on the Chinese language
- You need automatic rewriting of a whole text with no review: by default the skill only finds and shows
- You need writing from scratch: the set checks and edits finished text
Example request
Show which places in this text sound machine-written, make no edits, then propose fixes for the three most visible onesLimitations
The rules, signs and prompts target the Chinese language and do not suit Russian or English text. The README description and triggers are in Chinese. The skills work as prompts and quality depends on the model; the figures on telling machine and human text apart come from the author's own testing with no independent evaluation. The set of 25 signs belongs to the author and is not a standard.
How to disable. Delete the skill folders from `~/.claude/skills/` or `~/.agents/skills/`. For skills installed through `npx skills`, the removal command of the same installer works.
FAQ
Does the skill change the text itself?
No, by default it only detects and shows signs with a concentration level. Editing starts on your request, for example through the `editor-revisor` skill.
How do I install the set?
With `npx -y skills add taxueseek/plainspoken-writing -g --all`, which fits Claude Code, Codex and Cursor. To update run the same command again.
What is in the set?
Seven core and three auxiliary skills: voice calibration, fact checking, content assessment, machine-style detection, editing, packaging, style, platforms and punctuation.
Is Python required?
No, the skills work as prompts. The `core/` engine and `rules/` are needed only for batch checks and tests from a repository clone.
Does it suit Russian text?
No, the rules and signs are built on the Chinese language. The catalog has separate editing skills for Russian text.
Discussion
No questions or reviews yet.
Related
SkillsHuman Writing
human-writing
A skill for lively Chinese writing: helps the agent write and edit so a specific person seems to speak, not a model
An editor skill for Chinese text: removes filler, repetition and template phrasing, keeping facts, certainty and author voice
qu-ai-wei (去 AI 味)
qu-ai-wei
A skill that edits Simplified Chinese: cuts stock phrases and calques while keeping facts, numbers and the author's voice
Natural Talk strips the AI tone from Chinese text: answers from the first sentence, no filler, hollow metaphors or translationese.
A skill for writing in Chinese: goal and facts first, then platform, tone and a self-check of the draft before delivery.
speak-human-tw
說人話 speak-human-tw
A skill that proofreads Traditional Chinese: strips AI writing tells and fixes Taiwan localization
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Plainspoken Writing (this card) | no | 75 | |
| Humanizer RU | no | 210 | an editor of Russian text that removes AI traces and bureaucratese |
| anti-vibe-writing | no | 132 | a final-edit skill for English and Chinese drafts |
README in short
The README, in Chinese, explains that readers tell machine text by predictability and describes 25 signs, concentration levels and a no-edit default mode. Installation, updating, tables of core and auxiliary skills, the layered workflow and the repository layout follow.
README badge
Original SKILL.md
--- name: plainspoken-writing version: "1.10.0" description: | 【L1 声音真实度】白描 · AI 味检测与改写——只处理「像不像真人」。25 种 AI 味模式识别，根据严重程度选策略。默认只诊断不改，改写要用户确认。三种改写方式：自动替换、追问驱动、混合。可接收 voice-dissolver 声音清单作为保护区。 触发：AI味、像机器写的、太正式、读起来假、去AI味、AI感太重、像ChatGPT写的、太模板化、语气太标准、缺乏个性、AI腔、读起来没温度、句式太整齐、总结陈词、翻译腔、不说人话、套路化、白描、直陈白描、Plainspoken。 Not for：内容质量不够但没AI味（→ L0 chinese-write-checker）、结构冗余/废话多（→ L2 editor-revisor）、找不到自己声音（→ voice-dissolver）、排版/标点/标题（→ L3 packaging-workshop）、事实/数据/引文真假（→ L0 fact-review）。 --- # 白描：AI 味检测与消除 Plainspoken Writing 的 L1 层：只问「像不像真人」。中文名「白描」，主张「AI 爱工笔，人写白描」。 AI 味不是「写得不好」，是「读者感觉到背后没有真人」。**但去 AI 味 ≠ 好内容。** 花时间去 AI 味不如花时间把事情搞清楚。 **文字洁癖**（taxue）：AI 味不是 AI 的问题，是用套话掩盖不确定的思考。人话三要素：带立场、带偏见、带情绪。每删一个套话，都要问：我到底想说什么？如果删完发现原文也没什么真东西，那问题不在 AI 味，在内容本身。 **执行原则**： 1. **默认只诊断不改**。识别AI特征是帮人看清自己的文字，不是帮人伪装成人类。改写必须用户确认。 2. **去 AI 味 ≠ 好内容**。如果检测结果很干净，直接告诉用户。不要为了输出报告而硬找问题。 3. **改写必须基于用户的偏好**。每个 AI 特征背后都有一个本来想达成的目的。追问那个目的，而不是直接替换。 4. **见感思行是正面的改写目标**（taxue）。去 AI 味不只是删坏东西，更是让文字回到「见（观察）→ 感（感受）→ 思（思考）→ 行（行动）」的自然结构。 三种改写方式（用户选）： - **自动替换**：检测 → 直接改 → 质量验证 → 迭代 - **追问驱动**：检测 → 逐条追问写作意图 → 用户回答后给修改方向 - **混合**：自动替换 + 对关键特征追问意图 怎么运作：先判断AI味有多重，再选策略——轻微改几处、中等改一段、严重整篇重写、致命直接拦下。改完检查质量，不过关就回去重来，最多3轮。 --- ## 怎么一步步处理 先判断内容类型（推文/公众号/学术），确定AI味的判断标准。然后快速扫一遍，数数AI味关键词命中几个。 - 命中少于5处 → 直接改标记的位置 - 命中5处以上 → 进入全量诊断，先找根因（3个根因能同时引发多种表现），再验证独立信号 改之前先识别作者的声音——口头禅、动情处、不完美痕迹，这些标为保护区一字不动。 遇到疑似AI味的句子，走决策树判断该不该删：在保护区？密度够？换话题还成立？是喘息句？删了会失温度？ 改写时先清除明显废话，再逐段处理，最后检查人话三要素（偏见、情绪、立场）。