Remove Ai Style

# Remove AI Style

Review and adjust the writing style of an article to reduce obvious AI-generated patterns, making the text read more naturally and human-like.

## When to Use

Use this skill when the user asks to:
- Remove AI style from an article
- Make AI-generated text sound more natural
- Polish writing to reduce robotic or formulaic patterns

## Intensity Levels

The user can specify how aggressively to remove AI patterns. If they don't specify, **default to "heavy"**.

| Level | Description |
|-------|-------------|
| **Moderate** (中等) | Only fix the most obvious AI patterns. Preserve most of the original structure and phrasing. Light touch. |
| **Heavy** (尽力去除) | Actively rewrite AI-ish sentences, restructure overly formulaic paragraphs, and replace robotic transitions. This is the **default**. |
| **Full** (完全彻底) | Treat the entire text as a draft and rewrite it from scratch in a natural human voice, while preserving the original meaning and key information. The output should read as if a human wrote it from the start. |

## Instructions

1. **Determine the language** of the article (Chinese or English).
2. **Determine the intensity level** — check if the user specified moderate / heavy / full. Default to heavy.
3. **Load the corresponding reference document**:
   - Chinese articles: `references/chinese.md`
   - English articles: `references/english.md`
4. **Run the quick-scan scripts** provided in the reference document (if any) to locate symbol-level issues first. This gives you a fast overview of problem spots without reading line by line.
   - For Chinese articles, scan and handle ASCII double quotes (`"`) and Chinese double quotes (`“”`) with equal priority. Do not only fix ASCII quotes.
5. **Review the article** against the style guidelines in the reference document. The script results are just a starting point — you still need to read through the full text for patterns that scripts can't detect (e.g., overuse of metaphors, formulaic structure, robotic transitions).
6. **Apply fixes according to the intensity level**:
   - Moderate: only fix what the scripts flagged + the most glaring patterns.
   - Heavy: fix everything the guidelines call out, rewrite awkward sentences.
   - Full: rewrite the text from scratch preserving meaning.
7. **Preserve the original meaning and tone** — even at "full" intensity, the core message should remain intact.


# 中文去除 AI 味指南

检查文章中的风格，稍微微调一些地方，如果没有问题，不用调整也行。

## 快速扫描脚本

在逐行阅读之前，先跑以下脚本快速定位符号相关的问题。脚本结果仅供参考，帮助你快速跳转到可疑位置。

### 扫描英文直双引号

```bash
grep -n '"' <file_path> || echo "No ASCII double quotes found."
```

### 扫描中文弯双引号

```bash
grep -n '[“”]' <file_path> || echo "No Chinese double quotes found."
```

### 扫描破折号（英文 em dash / en dash，中文文章中通常应避免或替换）

```bash
grep -nP '[\x{2013}\x{2014}—–-]{2,}|--' <file_path> || echo "No suspicious dashes found."
```

### 扫描「核心」相关表达

```bash
grep -n '核心' <file_path> || echo "No '核心' patterns found."
```

### 扫描「不是...而是」「并非...而是」等否定-肯定转折

```bash
grep -nP '(不是|并非|而非|而不是|拒绝|舍弃).{0,15}(而是|专注|追求|而非)' <file_path> || echo "No negation-affirmation patterns found."
```

### 扫描「更xx的是」「最xx的是」

```bash
grep -nP '(更|最).{1,4}的是' <file_path> || echo "No '更/最xx的是' patterns found."
```

> **注意**：以上脚本只能抓到符号和固定句式，比喻、结构化表达等问题仍然需要通读全文才能发现。

---

## 1. 少用特殊符号
- 尽量少用双引号、破折号等显眼符号。如果是中文文章，英文直双引号 `"` 和中文弯双引号 `“”` 都要检查，二者同等处理；不能因为英文直双引号更容易被脚本抓到，就放过正文里成片的中文弯双引号。
- 正文里的双引号要主动压缩：能去掉就去掉，能改成自然转述就改成自然转述；少数确实需要强调的，优先改成「」或用其他表达方式变化一下，尽量不要沿用原来的双引号，因为双引号已经很容易被读者当成机器腔的马脚。
- 不要动代码块、JSON/YAML、命令行示例、Markdown 链接、图片 alt 文本里必要的引号，除非它们明显影响展示。

## 2. 少用比喻
- 少用比喻，可以接受一小部分的比喻，但不要在文章里多次使用比喻。大量的比喻会显得很浮夸。

## 3. 少用【不是而是】的表达
- 尽量少用【不是xx，而是xx】这样的意思上的转折。不是指字面上的【不是xx，而是xx】，我是指意思上出现【否定A，肯定B】这样的表达，都不行，举几个例子：
> ❌ 这款零食拒绝高糖高脂的配方，专注天然食材的原汁原味。
> ❌ 该模型可以通过真正的语义理解，而非简单的关键词匹配。
> ❌ 开源项目文档舍弃晦涩的专业术语，专注通俗易懂的实操指引。
> ❌ 模型部署容器化方案，要