Guide · Agent skill
How to use the Humanizer-zh 中文文本润色 agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Humanizer-zh is an agent skill that edits existing Chinese text so it reads more naturally, removing empty filler, repetition and formulaic wording. It is for writers, editors and anyone preparing a Chinese draft for publication who wants a light rewrite that keeps the original meaning. After this guide you will know what an agent skill is, how to install humanizer-zh, how to ask for help in natural language, and where its limits are.
Humanizer-zh 中文文本润色op7418/Humanizer-zh · MIT Download the skill Quick start
- 1Download the skill. Download the ZIP from this page or clone the repository, so you have the humanizer-zh folder with SKILL.md.
- 2Unzip into a skills folder. Place the folder in ~/.claude/skills/ for personal use, or in .claude/skills/ inside a project.
- 3Or upload in the Claude apps. Open Settings, go to Capabilities, then Skills, and upload the skill.
- 4Paste or point to your text. Provide the Chinese article, comment or document you want polished.
- 5Describe the editing job. State the text type and what must stay, such as tone, facts or formatting.
- 6Review the result. Compare the rewrite with your original, checking facts, uncertainty and attribution.
What an agent skill is and what humanizer-zh does
An agent skill is a folder with a SKILL.md file that gives an AI agent specific instructions for a task. When the task matches the skill, the agent follows those instructions. The skill is only guidance the agent reads and applies; it is not a standalone program.
Humanizer-zh tells the agent to review Chinese articles, comments and documents and tidy up empty filler, repetition and formulaic expressions. The source instructs it to keep the original facts, claims, uncertainty, attribution and the author's tone, and not to add invented details or experiences. It only changes wording where there is a real problem, and it treats its list of 31 patterns as hints rather than a blacklist.
- Works on existing text: polishing, not writing new content or adding evidence.
- Keeps facts, numbers, negations, conditions, time, attribution and stated uncertainty.
- Does not judge whether a text was written by AI and does not guarantee passing AI detectors.
Install humanizer-zh
You can install the skill by downloading its ZIP from this page and unzipping it into a skills folder. Use ~/.claude/skills/ for your personal skills, or .claude/skills/ inside a project to make it available only there. The folder should contain SKILL.md, README.md and CHANGELOG.md.
If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the skill there.
- Personal install: unzip into ~/.claude/skills/ so the skill folder is available across projects.
- Project install: unzip into .claude/skills/ inside the project folder.
- Claude apps: Settings → Capabilities → Skills, then upload the skill.
~/.claude/skills/How to trigger the skill with natural requests
You do not need a special command. Ask in plain language that describes an editing job: polish a Chinese draft, reduce template-like wording, or review a document for filler. The skill's stated trigger is editing or reviewing Chinese text to remove formulaic expressions.
State what must stay the same. The source says the skill keeps the author's tone and does not change the editing scope unless you ask. Mention the text type, such as a blog post or product description, so the agent can match the register.
- Ask for a polished final version of a pasted Chinese draft.
- Say which tone to keep, for example a technical review or a personal essay.
- For files, ask the agent to edit the file; the source says it only writes back when you ask it to modify a file.
Examples of what the agent returns
For a pasted text, the default output is the final rewritten text, with a short note only if you ask for one. The source's example removes an empty promotional sentence while keeping three features, a planned release, the testing status and the feedback evidence, and it adds no invented numbers or release dates.
Here is a request you can adapt: ask the agent to polish a Chinese article and keep the technical tone. The agent returns the revised text without drafts, pattern checklists or self-scores.
- Input with 'the best ever' praise: the praise is removed, factual features are kept.
- Input saying a result 'may' help: the uncertainty stays as 'may'.
- Input attributing a claim to unnamed experts: the attribution and its limits stay in place.
What it needs and safety notes
The skill requires nothing beyond the agent itself according to the source. It does not run scripts and does not use the network. Its allowed tools are Read, Write, Edit and AskUserQuestion, so the agent reads and edits text rather than running programs.
When editing files, the skill edits only the prose body by default and leaves code blocks, inline code, commands, paths, URLs, link targets, explicit IDs, YAML front matter and data unchanged. Headings are kept by default so anchors do not break.
- Requires: nothing listed in the source.
- Runs scripts: no. Uses network: no.
- Licence: MIT.
- Always review the output, especially for fact-heavy or legal text, because the agent may still shift tone.
Tips for better results
Give the agent the full text and say whether it is an essay, a product description or a technical document. The skill adjusts its behaviour by genre: technical and product text keeps terms and operational status, while business and academic text keeps formal register and argument structure.
If the source material lacks a detail, the skill keeps the wording general rather than inventing data. Check the output against your original for any changed negation, time, scope or attribution.
- Provide a writing sample if you want the output to match your own sentence length and punctuation habits.
- Ask for a light edit if you want minimal changes; the source says it does not force edits into text that already reads well.
- Keep a copy of the original before overwriting a file.
What people use it for
- Polishing a Chinese blog post or essay before publishing without changing its opinions.
- Tidying a product description so it has less filler while keeping its listed features.
- Preparing a clean final version of a Chinese draft with no extra commentary.
- Light editing of an existing document, with headings and code blocks left as they are.
Frequently asked questions
Is humanizer-zh free to use?
The skill is released under the MIT licence, as stated on this page. Check the repository licence file for the full terms before reusing it.
What does humanizer-zh require to run?
The source lists no requirements. It does not run scripts and does not use the network, so it works as instructions that the agent reads.
How do I install humanizer-zh in Claude?
Download the ZIP and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. In the Claude apps, you can instead go to Settings, Capabilities, Skills and upload it.
Can humanizer-zh detect AI-written text or pass AI detectors?
No. The skill states it does not judge authorship or guarantee passing AI detectors. It is for editing existing text, not verifying where it came from.
Will the skill change the meaning of my article?
It is instructed to keep facts, claims, uncertainty, attribution and tone, and not to add invented details. Still, the agent may shift tone, so review the output before publishing, especially for legal or fact-heavy text.
How is this different from simply asking an agent to proofread?
The skill gives the agent a defined set of editing rules, a priority order that protects meaning, and 31 pattern checks treated as hints. It returns the final text by default, without commentary or self-scores.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Humanizer-zh suits Chinese writers and editors who want a light, meaning-preserving polish of existing blog posts, essays and product descriptions, and who will review the result themselves. It is not suited to fact-heavy or legal text where any wording change matters, to writing new content, or to checking whether text was written by AI.
Download the skill