OH

op7418/humanizer-zh

Code review

This page was translated automatically. Technical details may differ from the original. View the original

21.3K stars Quality 71 Trend 85MIT

The Chinese version of Humanizer, Claude Code Skills, is designed to erase AI-generated traces in text.

Overview

- The core documents of this project are translated from the blader/humanizer - Utilities section (core rules, quick checklist, quality scoring) referenced hardikpandya/stop-slop - the original project's Signs of AI writing guide based on Wikipedia Humanizer-zh is a tool for removing AI-generated traces from text, helping you rewrite AI-generated content to sound more natural and human-like. This project is suitable for: - Editing and reviewing AI-generated content - Enhancing the humanization of articles - Learning to recognize common patterns in AI writing 1. Download the ZIP file of this project or clone it locally 2. Copy the Humanizer-zh folder to the skills directory of Claude Code: - : ~/.claude/skills/ - : %USERPROFILE%.claude\skills\skills\ This study analyzed the application of machine learning in medical diagnosis, focusing on early lung cancer screening. The study used data from 5,000 medical records from 2019 to 2023. I've been wondering how AI will change the way we work. Last week, I chatted with a few friends who work in products—some were excited, some worried about unemployment. Most likely, the truth lies somewhere boring in between. 1. Overemphasis on meaning, legacy, and broader trends 2. Overemphasis on visibility and media coverage 3. Superficial analysis ending with -ing 4. Promotional and advertising language 5. Vague attribution and vague wording 6. Outline "Challenges and Future Outlook" section 7. Overuse of "AI Vocabulary" 8. Avoid Using "is" (avoid verb avoidance) 9. Negative parallelism 10. Overuse of the three-stage rule 11. Deliberate word substitution (synonym loop) 12. False range

Overview (machine translated)

Edit empty language, repetition, and templated expressions in existing articles, comments, and documents, preserving facts, certainty, and author voice. Input is a passage of text or a file; default output is the final rewritten draft; sentences without issues can be left unchanged.

This is an editorial guide executed by Agent reads, not a standalone detection program. It cannot prove who wrote the article, nor does it guarantee it will pass through any AI detector.

Update on 2026-09-23: Rules and examples have been rewritten, clearly preserving facts, author positions, and style. Ordinary parallelism, dashes, and connectors are no longer uniformly modified. See update log for details.

Installation

This is the simplest installation method, and it will automatically install the skill in the correct directory.

Method 2: Clone via Git

Method 3: Manual installation

  1. Download the ZIP file of this project or clone it locally
  2. Copy the Humanizer-zh folder to the skills directory in Claude Code:
  • macOS/Linux: ~/.claude/skills/
  • Windows: %USERPROFILE%\.claude\skills\
  1. Make sure the folder structure is as follows:

Verify the installation

After restarting Claude Code or reloading skills, enter in the dialogue:

If installed successfully, the skill will be activated.

Usage

Paste text:

Edit file:

Only note “Suggestions provided, no document modifications” during review. If there is an author sample, it can be provided together. Skill will borrow expression habits but will not copy experiences, data, or viewpoints from the sample into the original text.

Paraphrase Example

The following are all teaching inputs; there are no supplementary materials not shown.

Product Introduction, Data Missing:

Original: This solution is stable, reliable, responsive, and easy to maintain.

Rewriting: This solution operates stably, responds quickly, and is easy to maintain.

No update on launch date, response speed, or maintenance costs. If the original text lacks evidence, you may request it from the author, but do not polish the data to supplement it.

Fault feedback, uncertainty:

Original text: The issue has been reported multiple times by the community and is believed to be related to a memory leak, but the cause has not yet been confirmed.

Rewriting: The community has reported this issue multiple times. It is believed to be related to a memory leak, but the cause has not been confirmed.

Keep “possibly related” and “not yet confirmed,” do not change speculation into causal conclusions, and do not arbitrarily attribute cause judgments to the community.

Update notes, featuring three real features:

Original: This update highlights the team’s pursuit of innovation. It adds export, search, and batch renaming. The offline editing program will be released in the next edition, but testing has not yet been completed.

Rewrite: This update adds export, search, and batch renaming. Offline editing is planned for the next edition and has not yet been tested.

Three functions are retained, not to avoid the three-stage method of deleting or adding items; Planning and testing states remain intended.

Clear text:

Original: Save the file first, then close the window. Changes that are not saved will be lost.

Rewrite: Save the file first, then close the window. Changes that have not been saved will be lost.

There is no requirement that every call will result in modifications.

Editorial Principles

This page was translated automatically. Technical details may differ from the original.

Recommended Tools

Try a different keyword or remove a filter.

Install

npx skillfish add op7418/humanizer-zh
New to MCP? Start with the mcpskill.org guide →