SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillshuashu-proofreading

huashu-proofreading

Writing & research

View repository

Updated 10-15 14:31

About this skill

AI-Flavor Review

A professional three-pass review capability that systematically lowers an article's AI detection rate and adds human flavor.

When to Use This Skill

Automatically load this skill when the following scenarios are detected:

  • The user says "give it a review," "lower the AI flavor," or "too AI-ish"
  • After the first draft is completed
  • The user mentions a "high AI detection rate" or "no human flavor"
  • The article needs a final polish

Core Goals

  • Lower the AI detection rate: target under 30%
  • Add human flavor: make the article feel experienced, warm, and like a real person talking
  • Maintain the Peanut style: practice-oriented, technically deep, easy to understand, useful and actionable

Three-Pass Review Process

First pass: content review (logic, facts, structure)

Goal: ensure the content is accurate and the logic is clear

Checklist：

  • Are the facts accurate? (data, dates, product names)
  • Is the logic clear? (no contradictions before and after)
  • Is the structure reasonable? (no going off-topic)
  • Is the information complete? (all key points covered)
  • Is there no fabrication? (all data and cases are real)

Method：

  • Read through from start to finish, marking points of doubt
  • Search to verify uncertain information
  • Check the accuracy of numbers and proper nouns

Second pass: style review (reducing AI flavor)

Goal: remove AI flavor, add human flavor

Core check: identifying 6 major categories of AI tone

1️⃣ Clichés Everywhere

❌ AI often says：

  • "In today's era," "against the backdrop of..."
  • "In summary," "all in all"
  • "It is worth noting that," "it must be emphasized that"
  • "With the development of...," "along with..."

✅ Rewrite direction: get straight to the point, don't beat around the bush

Example：

❌ "在当今AI技术飞速发展的时代，编程工具也在不断进化..."
✅ "Claude Code出了。我用了两周，确实比Cursor好用。"

2️⃣ AI Sentence Patterns (high-frequency repeated templates)

❌ AI often says：

  • "Not... but..." (appearing 3 or more times in a row)
  • "Not only... but also..." (used in stacks)
  • "On one hand... on the other hand..."
  • "Both... and..." (overused)

✅ Rewrite direction: break into short sentences, use varied expressions

Example：

❌ "Claude Code不是简单的工具，而是智能助手。它不仅能写代码，而且能理解意图。"
✅ "Claude Code不只是补全代码。它能理解你的意图，知道你想做什么。这个很关键。"

3️⃣ Written Vocabulary (strong sense of distance)

❌ AI often says：

  • "Significantly improved," "greatly enhanced" → use specific numbers
  • "Fully utilize" → "make good use of"
  • "Perform the operation" → use the verb directly ("click," "enter")
  • "Implement the function" → "get it done"
  • "Promote development" → "help..."

✅ Rewrite direction: make it colloquial and specific

Example：

❌ "通过充分利用Claude Code的能力，可以显著提升开发效率..."
✅ "用好Claude Code，你的开发速度能快不少。我测下来，平均每个项目省3-5小时。"

4️⃣ Mechanical Structure (AI-style list addiction)

❌ AI characteristics：

  • Overuse of "first, second, finally"
  • Dense meaningless subheadings (one every 200 words)
  • If it can be explained in prose, it must be turned into a list
  • "The following points", "mainly reflected in the following aspects"

✅ Rewrite direction: natural narrative, moderate use of lists

Example：

❌ "Claude Code的优势主要体现在以下几个方面：首先，它具有强大的上下文理解能力；其次，它支持多种编程语言；最后，它的响应速度很快。"

✅ "Claude Code最强的地方是什么？理解上下文。你让它改一个功能，它知道要动哪几个文件，不会漏掉。而且支持20多种语言，速度也很快。"

5️⃣ Neutral attitude (no stance, no opinion)

❌ AI characteristics：

  • Always objectively balanced: "has both advantages and disadvantages"
  • Afraid to make judgments: "depends on the actual situation"
  • Every sentence adds qualifiers: "possibly", "perhaps", "in some cases"

✅ Rewrite direction: clear attitude, dare to make judgments

Example：

❌ "Claude Code在某些场景下可能表现较好，但也存在一些局限性，具体效果取决于实际使用情况。"

✅ "Claude Code确实好用。处理大型项目时，比Cursor稳定多了。但价格贵，个人开发者可能扛不住。"

6️⃣ Missing details (abstract, not grounded)

❌ AI characteristics：

  • Many big words, few specific details
  • "many", "some", "significant", "a large number of" (without giving numbers)
  • Lacks real feelings and personal experience
  • No timeline, no scenes, no conflict

✅ Rewrite direction: add real details, specific numbers, personal experience

Example：

❌ "经过一段时间的使用，我发现Claude Code能够有效提升开发效率。"

✅ "用了两周。最明显的变化是，之前要3天的项目，现在1天半就能搞定。特别是重构老代码，Claude Code能一次性找出所有关联文件，不用我一个个翻。"

Second pass review: specific methods

  1. Check paragraph by paragraph, mark AI-sounding expressions
  2. Compare against the list of 6 major categories of AI tonerewrite
  3. Add personal experience and real details(check the personal material library)
  4. Add expressions of uncertainty(use "I think", "maybe", "perhaps" in moderation)
  5. Check whether it sounds like Huasheng speaking

Common rewriting techniques：

  • Split long sentences into short ones (split anything over 30 characters into 2-3 sentences)
  • Change abstract to concrete ("many" → "3", "significantly improved" → "saves 3-5 hours")
  • Change written language to spoken language ("perform an operation" → "click")
  • Add real feelings ("when I used it before...", "to be honest...")

Third pass: detail polishing (punctuation, layout, rhythm)

Goal: make the article read comfortably and naturally

Checklist：

  • Is the sentence length appropriate? (mainly 15-25 characters, no more than 30 characters)
  • Are paragraphs not too long? (3-5 lines on a phone screen)
  • Is the punctuation natural? (use more periods, fewer commas connecting long sentences)
  • Is there variation in rhythm? (a mix of fast and slow)
  • Are subheadings not too dense? (one every 300-500 characters)
  • Are lists used in moderation? (if prose can be used, don't use a list)
  • Is bolding used in moderation? (1-2 places every 200-300 characters)

Methods：

  • Read aloud and feel the rhythm
  • Find long sentences over 30 characters and split them shorter
  • Check paragraph length and split those that are too long
  • Adjust the layout and remove excessive formatting

Final check

  • Read through from beginning to end and rewrite places that feel choppy
  • It feels like Huasheng himself is speaking
  • (Optional) AI detection rate below 30%

Huasheng style characteristics (must be maintained)

  • Practice-oriented: has specific cases and executable steps
  • Technical depth + easy-to-understand expression: does not show off jargon, but has depth
  • Useful and actionable: readers can act immediately after reading
  • Full of human touch: has a sense of experience and warmth, avoiding AI tone
  • Authenticity of data: All tests and cases must be real
  • Appropriate uncertainty: When you should say "I think," say it; don't pretend to be omniscient

Using the personal material library

When to use：

  • ✅ The article needs real case support
  • ✅ Needs genuine evaluations of tools/products
  • ✅ Needs personal experience as an opening
  • ✅ The article is too "AI-flavored" and needs more human touch

Quick usage workflow：

Method A: Search raw data directly (recommended as first choice)

Applicable scenarios: New topics, unorganized themes, needing the most raw and authentic expression

Use the Grep tool to search /写作参考/个人素材库/全部即刻动态.csv：

pattern: "关键词1|关键词2"
path: /Users/alchain/Documents/写作/写作参考/个人素材库/全部即刻动态.csv
output_mode: content
-n: true

Example：

  • Writing about Amap's Saostreet List → search "Amap|Saostreet List"
  • Writing about AI programming tools → search "Cursor|Claude Code|Copilot"
  • Writing about product growth → search "growth|conversion|users"

Method B: View refined materials

Applicable scenarios: Common topics (AI programming tools, product development, etc.) already have systematic organization

  1. Open the master index:/写作参考/个人素材库/README.md
  2. Find the topic: open _即刻动态主题索引.md
  3. Select materials: view the refined file (such asAI编程工具使用心得.md）
  4. Rewrite for use: expand into long-form logic, maintain an authentic attitude

Notes：

  • All materials are real and cannot be fabricated or exaggerated
  • Pay attention to the time span (2025.05-2025.10) to avoid outdated information
  • Must be rewritten into long-form logic; cannot be directly copied and pasted

Review tools (optional)

Automated detection tools

  • Keyword detection: Search for clichés such as "in today's", "in summary", "it is worth noting", etc.
  • Sentence length check: Flag long sentences over 30 characters
  • AI detection: Use tools such as GPTZero, Originality.AI (for reference only)

Manual detection methods

  • Read aloud: The places where you stumble are the places that need revision
  • Compare with past articles: Compare with Huasheng's previous excellent articles to see whether the style is consistent
  • Ask yourself: Would Huasheng say it this way?

Common mistakes and rewrite examples

Mistake 1: Beating around the bush at the beginning

❌ "在当今AI技术飞速发展的背景下，编程工具领域也迎来了新的变革。作为一款新兴的AI编程助手，Claude Code正在受到越来越多开发者的关注..."

✅ "Claude Code出了。我用了两周，确实比Cursor好用。今天聊聊为什么。"

Mistake 2: Overusing AI sentence patterns

❌ "Claude Code不是简单的代码补全工具，而是一个智能编程助手。它不仅能够生成代码，而且能够理解开发者的意图。"

✅ "Claude Code不只是补全代码。它能理解你想干嘛。比如你说'重构这个模块'，它就知道该怎么做。"

Mistake 3: Lacking specific details

❌ "经过一段时间的测试，我发现Claude Code在很多方面都表现优秀。"

✅ "用了两周。最明显的变化是，之前要3天的项目，现在1天半就能搞定。"

Mistake 4: Attitude too neutral

❌ "Claude Code有其优势，也存在一些不足，具体是否适合使用需要根据实际情况判断。"

✅ "Claude Code确实好用。但价格贵，个人开发者可能扛不住。如果你公司报销，强烈推荐。"

Reference resources

  • /写作参考/AI味审校指南.md - Complete review guide (includes more examples)
  • /写作参考/个人素材库/README.md - Master index of the personal material library
  • /写作参考/个人素材库/全部即刻动态.csv - Raw material data
  • /写作参考/个人素材库/_即刻动态使用规范.md - Material usage guidelines

Success cases

  • "Why the Best Content Is All Useless": After three rounds of review, the AI detection rate dropped to 28%
  • "How a Team Burning Billions of Tokens a Month Uses Claude Code": Added a large number of real cases, strong human touch

Last updated: 2025-11-07 Applicable projects: WeChat official account writing, video scripts Maintainer: Huasheng

Produced by Huashu | AI Native Coder · Indie Developer WeChat Official Account "Huashu" | 300K+ followers | AI tools and efficiency improvement Notable works: Little Cat Fill Light (AppStore paid chart Top1) · "Master DeepSeek in One Book"