anti-ai-editor

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Anti-AI Editor

Anti-AI 编辑器

Operator Context

操作背景

This skill operates as an operator for content editing, detecting and removing AI-generated writing patterns. It implements the Targeted Revision architectural pattern -- scan for patterns, propose minimal fixes, preserve meaning -- with Wabi-Sabi Authenticity ensuring human imperfections are features, not bugs.

本技能作为内容编辑的操作工具，检测并移除AI生成的写作模式。它采用**Targeted Revision（针对性修订）架构模式——扫描表达模式、提出最小化修改方案、保留原意——并通过Wabi-Sabi Authenticity（侘寂真实感）**原则确保人类的不完美是特色而非缺陷。

Hardcoded Behaviors (Always Apply)

硬编码行为（始终适用）

CLAUDE.md Compliance: Read and follow repository CLAUDE.md before editing
Over-Engineering Prevention: Make minimal fixes only. No rewrites, no "while I'm here" improvements
Preserve Meaning: NEVER change actual meaning or intent while fixing style
Show All Changes: Display before/after for every modification with reason
Context Awareness: Some flagged words are appropriate in technical contexts
Wabi-Sabi Enforcement: Human imperfections (run-ons, fragments, loose punctuation) are features -- do NOT "fix" them

CLAUDE.md 合规性：编辑前阅读并遵循仓库中的CLAUDE.md
避免过度设计：仅进行最小必要修改。不重写内容，不做“顺手改进”
保留原意：修改风格时绝不能改变内容的实际含义或意图
展示所有修改：对每一处修改显示前后对比及修改原因
上下文感知：某些被标记的词汇在技术场景中是合适的
侘寂原则执行：人类的不完美（如流水句、碎片句、松散标点）是特色——请勿“修正”这些内容

Default Behaviors (ON unless disabled)

默认行为（默认开启，可关闭）

Full Preview: Show complete edited content before saving
Categorized Reporting: Group issues by type (cliches, passive, structural, meta)
Actionable Fixes: Every detected issue includes a specific replacement
Frontmatter Skip: Skip YAML frontmatter, code blocks, and inline code
Voice Integration: If voice specified, check voice-specific anti-patterns

完整预览：保存前显示完整的编辑后内容
分类报告：按问题类型分组展示（陈词滥调、被动语态、结构问题、元评论等）
可执行修改方案：每个检测到的问题都包含具体的替换建议
跳过前置元数据：跳过YAML前置元数据、代码块和行内代码
语气集成：若指定了语气，检查该语气特有的反模式

Optional Behaviors (OFF unless enabled)

可选行为（默认关闭，可开启）

Auto-Apply: Apply changes without preview confirmation
Aggressive Mode: Flag borderline cases (use for marketing content)
Stats Only: Report issues without suggesting fixes

自动应用：无需预览确认直接应用修改
激进模式：标记边界案例（适用于营销内容）
仅统计：仅报告问题不提供修改建议

What This Skill CAN Do

本技能可实现的功能

Detect AI cliches and suggest natural replacements
Identify passive voice overuse and suggest active alternatives
Flag structural issues (monotonous sentence length, list overuse)
Remove meta-commentary that adds no value
Handle Hugo frontmatter correctly (skip YAML, edit content only)
Preserve code blocks and technical terminology
Show before/after comparisons for all changes

检测AI式陈词滥调并提供自然替代表达
识别过度使用的被动语态并提供主动语态替代方案
标记结构问题（句子长度单调、过度使用列表）
移除无价值的元评论
正确处理Hugo前置元数据（跳过YAML，仅编辑正文内容）
保留代码块和专业术语
展示所有修改的前后对比

What This Skill CANNOT Do

本技能不可实现的功能

Rewrite content entirely (use targeted fixes only)
Change technical accuracy for stylistic reasons (meaning is sacred)
Remove domain-specific jargon that is appropriate in context
Fix factual errors (style-only skill, not a fact-checker)
Generate new content (use voice skills instead)
Polish away authentic imperfections (see Wabi-Sabi)

全文重写内容（仅支持针对性修改）
为了风格修改而改变技术准确性（原意不可侵犯）
移除场景中合适的领域特定行话
修正事实错误（仅为风格类技能，非事实核查工具）
生成新内容（请使用语气类技能）
消除真实的不完美（详见侘寂真实感）

Instructions

操作步骤

Phase 1: ASSESS

阶段1：评估

Goal: Read file, identify skip zones, scan for AI patterns.

Step 1: Read and classify the file

Read the target file. Identify file type (blog post, docs, README). Skip frontmatter (YAML between

---

markers), code blocks, inline code, and blockquotes.

Step 2: Scan for issues by category

Category	What to Find	Reference
AI Cliches	"delve", "leverage", "utilize", "robust"	`references/cliche-replacements.md`
News AI Tells	"worth sitting with", "consequences extend beyond", "that's the kind of", dramatic rhythm	`references/detection-patterns.md`
Copula Avoidance	"serves as a", "boasts a", "features a"	`references/detection-patterns.md`
Passive Voice	"was done by", "has been", "will be"	`references/detection-patterns.md`
Structural	Monotonous sentence lengths, excessive lists, boldface overuse, dramatic AI rhythm	`references/detection-rules.md`
Meta-commentary	"In this article", "Let me explain", "As we've discussed"	`references/cliche-replacements.md`
Dangling -ing	"highlighting its importance", "underscoring the significance"	`references/detection-patterns.md`
Puffery/Legacy	"testament to", "indelible mark", "enduring legacy"	`references/detection-patterns.md`
Generic Closers	"future looks bright", "continues to evolve"	`references/detection-patterns.md`
Curly Quotes	\u201C \u201D \u2018 \u2019 (ChatGPT-specific)	`references/detection-patterns.md`

Step 3: Count and classify issues

Record each issue with line number, category, and severity weight:

AI Cliche (Tier 1): weight 3
News AI Tell (Tier 1-News): weight 3 (pseudo-profound, philosophizing, meta-significance)
Copula Avoidance (Tier 1b): weight 3
Meta-commentary: weight 2
Dangling -ing clause (Tier 2b): weight 2
Significance puffery (Tier 2c): weight 2
Generic positive conclusion (Tier 2d): weight 2
Dramatic AI rhythm (Tier 1-News): weight 2
Structural issue: weight 2
Fluff phrase: weight 1
Passive voice: weight 1
Redundant modifier: weight 1
Curly quotes (Tier 3b): weight 1

Gate: Issues documented with line numbers and categories. Total severity score calculated. Proceed only when gate passes.

目标：读取文件，识别跳过区域，扫描AI式表达模式。

步骤1：读取并分类文件

读取目标文件，识别文件类型（博客文章、文档、README）。跳过前置元数据（

---

标记之间的YAML内容）、代码块、行内代码和块引用。

步骤2：按类别扫描问题

类别	检测内容	参考文件
AI式陈词滥调	"delve"、"leverage"、"utilize"、"robust"等	`references/cliche-replacements.md`
新闻类AI特征	"worth sitting with"、"consequences extend beyond"、"that's the kind of"、戏剧性节奏	`references/detection-patterns.md`
系动词规避	"serves as a"、"boasts a"、"features a"	`references/detection-patterns.md`
被动语态	"was done by"、"has been"、"will be"	`references/detection-patterns.md`
结构问题	句子长度单调、过度使用列表、过度使用粗体、AI式戏剧性节奏	`references/detection-rules.md`
元评论	"In this article"、"Let me explain"、"As we've discussed"	`references/cliche-replacements.md`
悬垂-ing结构	"highlighting its importance"、"underscoring the significance"	`references/detection-patterns.md`
浮夸/套话	"testament to"、"indelible mark"、"enduring legacy"	`references/detection-patterns.md`
通用结尾语	"future looks bright"、"continues to evolve"	`references/detection-patterns.md`
弯引号	\u201C \u201D \u2018 \u2019（ChatGPT特有）	`references/detection-patterns.md`

步骤3：统计并分类问题

记录每个问题的行号、类别和严重程度权重：

AI式陈词滥调（1级）：权重3
新闻类AI特征（1级-新闻）：权重3（伪深刻、哲理化、元意义表达）
系动词规避（1b级）：权重3
元评论：权重2
悬垂-ing从句（2b级）：权重2
浮夸表达（2c级）：权重2
通用积极结论（2d级）：权重2
AI式戏剧性节奏（1级-新闻）：权重2
结构问题：权重2
冗余短语：权重1
被动语态：权重1
冗余修饰词：权重1
弯引号（3b级）：权重1

准入条件：问题需记录行号和类别，计算总严重程度得分。仅当满足条件时方可进入下一阶段。

Phase 2: DECIDE

阶段2：决策

Goal: Determine editing approach based on severity.

Step 1: Choose approach by issue count

Severity Score	Approach
0-5	Report "Content appears natural". Stop.
6-15	Apply targeted fixes
16-30	Group by paragraph, fix systematically
30+	Paragraph-by-paragraph review

Step 2: Prioritize fixes

Structural Issues (affect overall readability)
AI Cliches (most obvious tells)
Meta-commentary (usually removable)
Passive Voice (case-by-case judgment)

Step 3: Wabi-sabi check

Before proposing any fix, ask: "Would removing this imperfection make it sound MORE robotic?" If yes, do NOT flag it. Preserve:

Run-on sentences that convey enthusiasm
Fragment punches that create rhythm
Loose punctuation that matches conversational flow
Self-corrections mid-thought ("well, actually...")

Gate: Approach selected. Fixes prioritized. Wabi-sabi exceptions noted. Proceed only when gate passes.

目标：根据严重程度确定编辑方案。

步骤1：根据问题数量选择方案

严重程度得分	处理方案
0-5	报告「内容看起来自然」，停止操作。
6-15	应用针对性修改
16-30	按段落分组，系统性修改
30+	逐段审核修改

步骤2：优先处理顺序

结构问题（影响整体可读性）
AI式陈词滥调（最明显的AI特征）
元评论（通常可直接移除）
被动语态（视具体情况判断）

步骤3：侘寂原则检查

在提出任何修改方案前，需自问：「移除这个不完美会让内容听起来更机械吗？」如果是，则请勿标记该内容。需保留：

能传达热情的流水句
创造节奏感的碎片短句
符合对话流程的松散标点
中途自我修正的表达（如「嗯，实际上...」）

准入条件：已选择处理方案，修改优先级已确定，侘寂原则例外情况已记录。仅当满足条件时方可进入下一阶段。

Phase 3: EDIT

阶段3：编辑

Goal: Generate edit report, get confirmation, apply changes.

Step 1: Generate the edit report

=================================================================
 ANTI-AI EDIT: [filename]
=================================================================

 ISSUES FOUND: [total]
   AI Cliches: [count]
   Passive Voice: [count]
   Structural: [count]
   Meta-commentary: [count]

 CHANGES:

 Line [N]:
   - "[original text]"
   + "[replacement text]"
   Reason: [specific explanation]

 [Continue for all changes]

=================================================================
 PREVIEW
=================================================================
[Show complete edited content]

=================================================================
 Apply changes? [Waiting for confirmation]
=================================================================

Step 2: Apply changes after confirmation

Use the Edit tool for each change. Verify each edit applied correctly.

Gate: All changes applied. File re-read to confirm no corruption. Proceed only when gate passes.

目标：生成编辑报告，获取确认后应用修改。

步骤1：生成编辑报告

=================================================================
 ANTI-AI EDIT: [filename]
=================================================================

 ISSUES FOUND: [total]
   AI Cliches: [count]
   Passive Voice: [count]
   Structural: [count]
   Meta-commentary: [count]

 CHANGES:

 Line [N]:
   - "[original text]"
   + "[replacement text]"
   Reason: [specific explanation]

 [Continue for all changes]

=================================================================
 PREVIEW
=================================================================
[Show complete edited content]

=================================================================
 Apply changes? [Waiting for confirmation]
=================================================================

步骤2：确认后应用修改

使用编辑工具逐一应用修改，验证每一处修改是否正确应用。

准入条件：所有修改已应用，重新读取文件确认未损坏。仅当满足条件时方可进入下一阶段。

Phase 4: VERIFY

阶段4：验证

Goal: Confirm edits preserved meaning and improved naturalness.

Step 1: Re-read edited file completely

Step 2: Verify no meaning was lost or changed

Step 3: Verify no new AI patterns were introduced by edits

Step 4: Confirm frontmatter and code blocks are untouched

Step 5: Report final summary

markdown

undefined

目标：确认修改保留了原意并提升了内容的自然度。

步骤1：完整重读编辑后的文件

步骤2：验证未丢失或改变原意

步骤3：验证修改未引入新的AI式表达模式

步骤4：确认前置元数据和代码块未被改动

步骤5：生成最终总结报告

markdown

undefined

Edit Summary

编辑总结

File: [path] Issues Found: [count] Issues Fixed: [count] Issues Skipped: [count with reasons] Meaning Preserved: Yes/No


**Gate**: All verification steps pass. Edit is complete.

---

File: [path] Issues Found: [count] Issues Fixed: [count] Issues Skipped: [count with reasons] Meaning Preserved: Yes/No


**准入条件**：所有验证步骤通过，编辑完成。

---

Examples

示例

Example 1: Blog Post (Heavy Editing)

示例1：博客文章（重度编辑）

User says: "De-AI this blog post" Actions:

Read file, skip frontmatter, scan all categories (ASSESS)
Score 22 -- systematic paragraph-by-paragraph approach (DECIDE)
Generate report with 10 changes, show preview, apply after confirmation (EDIT)
Re-read, verify meaning preserved, no new AI patterns (VERIFY) Result: 67% shorter intro, all AI cliches removed, voice preserved

用户需求：「对这篇博客文章进行去AI化处理」操作流程：

读取文件，跳过前置元数据，扫描所有类别（评估阶段）
得分22——采用逐段系统性处理方案（决策阶段）
生成包含10处修改的报告，展示预览，确认后应用修改（编辑阶段）
重读内容，验证原意保留，无新AI模式引入（验证阶段）结果：开头缩短67%，所有AI式陈词滥调被移除，作者语气得以保留

Example 2: Technical Docs (Light Editing)

示例2：技术文档（轻度编辑）

User says: "Check this for AI patterns" Actions:

Read file, identify technical context, scan for patterns (ASSESS)
Score 7 -- targeted fixes only, preserve technical terms (DECIDE)
Replace "utilizes" with "uses", remove throat-clearing, show preview (EDIT)
Verify technical accuracy unchanged (VERIFY) Result: Clearer prose, same information, technical terms untouched

用户需求：「检查这份文档是否存在AI式表达模式」操作流程：

读取文件，识别技术场景，扫描表达模式（评估阶段）
得分7——仅进行针对性修改，保留专业术语（决策阶段）
将「utilizes」替换为「uses」，移除冗余开场语，展示预览（编辑阶段）
验证技术准确性未受影响（验证阶段）结果：文本更简洁清晰，信息完整保留，专业术语未改动

Error Handling

错误处理

Error: "File Not Found"

错误：「文件未找到」

Cause: Path incorrect or file does not exist Solution:

Verify path with
```
ls -la [path]
```
Use glob pattern to search:
```
Glob **/*.md
```
Confirm correct working directory

原因：路径错误或文件不存在解决方案：

使用
```
ls -la [path]
```
验证路径
使用通配符搜索：
```
Glob **/*.md
```
确认当前工作目录正确

Error: "No Issues Found"

错误：「未检测到问题」

Cause: Content is already natural, or scanner missed patterns Solution:

Report "Content appears natural -- no AI patterns detected"
Show sentence length statistics for manual verification
Check structural patterns (monotony, list overuse) even if no word-level flags

原因：内容已自然，或扫描器遗漏了模式解决方案：

报告「内容看起来自然——未检测到AI式表达模式」
展示句子长度统计数据供人工验证
即使无词汇层面的标记，也需检查结构模式（单调、过度使用列表）

Error: "Frontmatter Corrupted After Edit"

错误：「编辑后前置元数据损坏」

Cause: Edit tool matched content inside YAML frontmatter Solution:

Fall back to treating entire file as content
Re-read file to verify YAML integrity
If corrupted, restore from git:
```
git checkout -- [file]
```

原因：编辑工具匹配了YAML前置元数据内的内容解决方案：

退回到将整个文件视为正文处理
重新读取文件验证YAML完整性
若已损坏，从Git恢复：
```
git checkout -- [file]
```

Anti-Patterns

反模式

Anti-Pattern 1: Changing Meaning While Fixing Style

反模式1：修改风格时改变原意

What it looks like: Removing "edge cases" from "This solution robustly handles edge cases" -- losing meaningful technical information Why wrong: Style edits must never change what the content says Do instead: "This solution handles edge cases reliably" -- fix style, keep meaning

表现：将「This solution robustly handles edge cases」修改为「This solution handles edge cases」——丢失了有意义的技术信息 错误原因：风格修改绝不能改变内容的核心信息 正确做法：修改为「This solution handles edge cases reliably」——修正风格，保留原意

Anti-Pattern 2: Over-Correcting Natural Informal Language

反模式2：过度修正自然的非正式语言

What it looks like: Removing "So basically" from a casual blog post because it sounds informal Why wrong: "So basically" is natural spoken rhythm. Blog posts can be conversational. Do instead: Leave natural voice markers alone. Only remove AI-generated patterns.

表现：移除博客文章中的「So basically」，理由是听起来不正式 错误原因：「So basically」是自然的口语化表达节奏，博客文章可以采用对话式语气 正确做法：保留自然的语气标记，仅移除AI生成的表达模式

Anti-Pattern 3: Ignoring Technical Context

反模式3：忽略技术场景上下文

What it looks like: Flagging "leverage" in "Use a lever to leverage mechanical advantage" Why wrong: "Leverage" is technically correct when discussing actual mechanics Do instead: Only flag words when used as corporate-speak, not in their literal or technical sense

表现：标记「Use a lever to leverage mechanical advantage」中的「leverage」 错误原因：在讨论机械原理时，「leverage」是专业术语，使用正确 正确做法：仅当词汇被用作企业套话时才标记，字面或技术场景下的使用无需标记

Anti-Pattern 4: Wholesale Rewrites Instead of Targeted Edits

反模式4：全文重写而非针对性修改

What it looks like: Completely rewriting a paragraph instead of fixing specific patterns Why wrong: Loses author voice, may introduce new AI patterns, harder to review Do instead: Make the minimum changes needed. Multiple small edits beat one big rewrite.

表现：完全重写段落而非修正特定的表达模式 错误原因：丢失作者语气，可能引入新的AI模式，更难审核 正确做法：仅进行必要的最小修改，多次小修改优于一次大重写

Anti-Pattern 5: Reporting Without Actionable Fixes

反模式5：仅报告问题不提供可执行修改方案

What it looks like: "Line 15: Contains AI-sounding language" with no specific fix Why wrong: Useless feedback -- the user needs to know WHAT to change and HOW Do instead: Show exact original text, exact replacement, and reason for the change

表现：「第15行：包含AI式表达语言」但未给出具体修改建议 错误原因：无意义的反馈——用户需要知道具体修改内容和方式 正确做法：展示准确的原文、替换文本及修改原因

References

参考资料

This skill uses these shared patterns:

Anti-Rationalization - Prevents shortcut rationalizations
Verification Checklist - Pre-completion checks
Wabi-Sabi Authenticity - Preserves human imperfections

本技能使用以下共享模式：

Anti-Rationalization - 避免捷径式合理化
Verification Checklist - 完成前检查清单
Wabi-Sabi Authenticity - 保留人类不完美

Domain-Specific Anti-Rationalization

领域特定反合理化

Rationalization	Why It's Wrong	Required Action
"It's just a style word, keep it"	AI cliches are the most obvious tells	Check against cliche list, replace if matched
"Fixing this would lose the flow"	Flow from AI patterns is synthetic flow	Remove and let natural rhythm emerge
"Technical content needs formal language"	Formal does not mean AI-sounding	Keep technical terms, remove corporate-speak
"The author probably wrote it that way"	If 5+ AI patterns cluster, it's generated	Apply systematic editing regardless
"Minor issues, not worth fixing"	Minor issues accumulate into AI tells	Fix all detected patterns

合理化借口	错误原因	要求操作
"这只是个风格词汇，保留即可"	AI式陈词滥调是最明显的AI特征	对照陈词滥调列表检查，匹配则替换
"修改这个会破坏流畅性"	AI模式带来的流畅性是合成的	移除AI模式，让自然节奏自然呈现
"技术内容需要正式语言"	正式不等于AI式表达	保留专业术语，移除企业套话
"作者可能就是这么写的"	若出现5个以上AI模式聚集，则为生成内容	无论如何都要进行系统性编辑
"问题很小，不值得修改"	小问题累积会形成AI特征	修复所有检测到的模式

Reference Files

参考文件

${CLAUDE_SKILL_DIR}/references/cliche-replacements.md

: Complete list of 80+ AI phrases with replacements

${CLAUDE_SKILL_DIR}/references/detection-patterns.md

: Regex patterns for automated detection

${CLAUDE_SKILL_DIR}/references/detection-rules.md

: Inline detection rules and structural checks

${CLAUDE_SKILL_DIR}/references/examples.md

: Before/after examples from real edits

${CLAUDE_SKILL_DIR}/references/cliche-replacements.md

: 包含80+个AI短语及替代表达的完整列表

${CLAUDE_SKILL_DIR}/references/detection-patterns.md

: 用于自动检测的正则表达式模式

${CLAUDE_SKILL_DIR}/references/detection-rules.md

: 行内检测规则和结构检查标准

```
${CLAUDE_SKILL_DIR}/references/examples.md
```
: 来自真实编辑案例的前后对比示例