editor-review

---

name: editor-review description: [Workflow Sub-task] This is a sub-task Skill, usually invoked and scheduled by workflow-producer. Use only when workflow-producer is in the [Step 7: Review] stage or at the end of the collaboration process. Do not use it directly to respond to users' general "writing" requests.

Editor-in-Chief Review

Instructions

When the user requests a review, article optimization, or quality check, diagnose according to the following dimensions and output a report:

1. AI-flavor detection (highest priority!)

This is thefirst priorityof the review. If the article looks like it was written by AI at first glance, then everything else is meaningless.

Phase 1: Basic symptom detection (must be executed)

1.1 Paragraph length standard deviation detection ⭐⭐⭐⭐⭐

Detection method:

1. Count the word count of each paragraph in the full text (excluding metadata)
2. Calculate the standard deviation
3. Judgment:
   • Standard deviation < 50 words → Severe AI flavor
   • Standard deviation 50-80 words → Mild AI flavor
   • Standard deviation > 80 words → Normal

Example:

❌ AI写作：[180, 180, 200, 210, 150, 180, 200] → 标准差 22.4
✅ 人类写作：[250, 80, 180, 300, 50, 200, 120] → 标准差 85.6

Rewriting suggestions:

• Split some paragraphs into short sentences as separate paragraphs
• Merge some paragraphs into long paragraphs
• Goal: Make paragraph lengths appear "wavy" rather than "flat"

1.2 Emotional fluctuation detection ⭐⭐⭐⭐⭐

Detection indicators:

1. Changes in exclamation mark density (number of exclamation marks per paragraph)
2. Distribution of profanity/colloquial words ("damn", "bullshit", "forget it", etc.)
3. Distribution of short sentences (<10 words)

Judgment:

• If emotional intensity remains stable throughout the text (fluctuation < 30%) → AI flavor
• If there is no sudden emotional climax or trough → AI flavor

Rewriting suggestions:

• Suddenly insert a short sentence in some paragraphs: "Buy a lottery ticket."
• Suddenly erupt in some paragraphs: "This is fucking ridiculous!"
• Suddenly become calm in some paragraphs: "Forget it, let's not talk about this."

1.3 "Perfect ending" detection ⭐⭐⭐⭐

Typical AI ending patterns:

• Echo the theme + elevate the meaning + give an answer
• "So, the truth about XXX is... we should... only in this way can we..."
• "In summary...", "Let us all look forward to..."

Judgment: If the ending paragraph simultaneously satisfies the following 3 points → AI flavor

1. Clearly responds to the question raised at the beginning
2. Gives an "answer" or "insight"
3. Uses summary vocabulary

Rewriting suggestions:

• End abruptly, without explanation: "Time never owes anyone an explanation."
• Leave suspense: "Where did the time of those 4,000 children go?"
• End with a rhetorical question: "What do you think?"

1.4 Textbook-style expression detection ⭐⭐⭐⭐

Banned vocabulary list:

• "worthy of our deep reflection"
• "aroused widespread attention"
• "is of great significance"
• "it is not difficult to find"
• "from this it can be seen"
• "in today's society"
• "with the development of..."
• "As everyone knows"

Determination: If ≥2 forbidden words appear → AI flavor

Rewrite suggestion: Delete directly, or change it into plain human speech:

• "Worth pondering" → "Just thinking about it is fucking terrifying"
• "It is not hard to find" → "Look"
• "Thus it can be seen" → "So"

---

Phase two: deep symptom detection (recommended)

1.5 Sentence pattern repetition detection ⭐⭐⭐⭐

Detection method: Count the opening sentence patterns of each paragraph; if the repetition rate is > 30% → AI flavor

Common AI sentence patterns:

• "...right? But have you ever thought about it..." (rhetorical question + turn)
• "...sounds very X, but in fact..." (contrastive pattern)
• "The key is...", "The problem is..." (lead-in sentences)
• "First let's talk about...", "Then let's look at..." (sequence words)

Rewrite suggestion: Shuffle the sentence patterns and add variation:

• Some paragraphs start with declarative sentences
• Some paragraphs start with rhetorical questions
• Some paragraphs start with data
• Some paragraphs start with profanity

1.6 Logic chain too neat ⭐⭐⭐⭐

Typical AI logic chain:

引入话题 → 正面案例 → 反面案例 → 深层原因 → 总结升华

Determination: If the overall logical progression shows an obvious "general-specific-general" or "progressive" structure, and every paragraph strictly serves the theme → AI flavor

Rewrite suggestion:

• Allow "going off topic": while writing, suddenly rant about something else
• Allow "pulling it back": "I've gone off track, back to XXX"
• Allow "repetition": repeat a certain point, saying it from different angles

1.7 Data piling detection ⭐⭐⭐

Determination: If every paragraph has data/cases, and they are evenly distributed → AI flavor

AI writing:

段1：数据A → 段2：案例B → 段3：数据C → 段4：案例D

Human writing:

段1-3：连续吐槽，没数据 → 段4：突然甩出3个数据 → 段5-6：又开始吐槽

Rewrite suggestion:

• Concentrate the data in a few paragraphs
• Make other paragraphs pure opinion/pure ranting

1.8 Transition naturalness detection ⭐⭐⭐

AI transition words:

• "Speaking of this, we cannot fail to mention..."
• "But the problem goes far beyond this..."
• "The deeper reason lies in..."

Human transition words:

• "I've gone off track."
• "Forget it, let's not talk about this."
• "By the way, there's one more thing..."

Determination: If the transitions between paragraphs are all very "smooth", with no abruptness → AI flavor

1.9 Paragraph function singularity detection ⭐⭐⭐

AI paragraph functions:

段1：引入 → 段2：论证 → 段3：举例 → 段4：反驳 → 段5：总结

Human paragraph functions:

段1：引入+吐槽+举例 → 段2：论证+跑题+拉回来 → 段3：纯吐槽 → 段4：突然总结

Determination: If each paragraph does only one thing, and the functions are evenly distributed → AI flavor

1.10 Punctuation monotony detection ⭐⭐⭐

Count the usage rate of punctuation in the full text:

• Exclamation marks: should account for 10-15%
• Ellipses: should account for 5-8%
• Dashes: should account for 3-5%
• Question marks: should account for 8-12%

Determination: If the full text uses only periods, commas, and question marks, and the usage rate of exclamation marks/ellipses/dashes is < 5% → AI flavor

---

1.11 Fake Celebrity Case Detection ⭐⭐⭐⭐

Detection method: Scan the full text, look for the case pattern of "name + age + occupation"

Typical AI case patterns:

• "Zhang San, 35 years old, resigned from the position of department manager..."
• "Li Si, 40 years old, worked within the system for 15 years..."
• "Wang Wu, 28 years old, employee at a big internet company..."
• "Xue Binglei, 35 years old, department manager..."

Judgment: If a case of "unfamiliar name + specific age + occupational background" appears → AI flavor

Problem:

1. Readers will question: "Who is this person?"
2. Weak sense of realism (seems made up)
3. Interrupts the reading rhythm
4. Not a celebrity but uses a real name, seems very fake

Rewriting suggestions:

Option 1: Replace with "my friend"

❌ "薛冰雷，35岁，部门经理..."
✅ "我有个朋友，35岁，部门经理..."

Option 2: Replace with an identity label

❌ "程亮，体制内辞职的..."
✅ "一个体制内的朋友，辞职后..."

Option 3: Use generic forms of address such as "Old Wang" or "Xiao Li"

❌ "薛冰雷，35岁..."
✅ "老王，35岁..."

Option 4: Use "I've seen" to increase the sense of realism

❌ "薛冰雷辞职创业了"
✅ "我见过一个35岁的部门经理，辞职创业..."

Exceptions: If it is a real celebrity (such as Liu Run, Ma Yun, Lei Jun), the name can be used directly

---

1.12 Flatness Detection (Lack of Spreadability) ⭐⭐⭐⭐⭐

Core problem: The article is correct, complete, and logically clear, but has no spreadability at all

Detection dimensions:

1. Emotional outburst point detection

• Count the number of exclamation marks in the full text
• Count the number of swear words/colloquial words ("damn", "bullshit", "forget it")
• Count the number of short sentences (<10 characters)

Judgment:

• Fewer than 3 exclamation marks in the full text → emotionally flat
• No swear words/colloquial words in the full text → too mild
• Fewer than 5 short sentences in the full text → lacks emotional outburst

2. Golden sentence density detection

• Scan the full text, look for sentences that can be screenshotted and shared
• Golden sentence characteristics: parallelism, reversal, extreme expression, hitting the nail on the head

Judgment:

• Fewer than 2 golden sentences in the full text → no sharing value

Golden sentence examples:

✅ "27岁是企业的燃料，40岁是家庭的支柱。"
✅ "船票有限，但你的命只有一条。"
✅ "欲望是别人的，但命是你自己的。"
❌ "这就是问题的核心。"（太平淡）

3. Case visual imagery detection

• Check whether the cases have specific details (dialogue, actions, scenes)

AI cases (no visual imagery):

薛冰雷，35岁，从部门经理位置辞职创业。

Human cases (with visual imagery):

老王，35岁，部门经理。辞职那天，老板问他想清楚了吗，他说想清楚了。
两年后，他在朋友圈发了条动态："创业不易，且行且珍惜。"
配图是一张空荡荡的办公室。你猜他想清楚了吗？屁。

Judgment: If all cases are "stating facts," with no dialogue/actions/scenes → lacks visual imagery

4. Transition word mildness detection

• Count the frequency of use of "argumentative transition words"

AI transition words (mild):

• "Then here's the question"
• "This is the core of the problem"
• "So, back to..."
• "Where's the difference?"

Human transition words (direct):

• "I've gone off track."
• "Forget it, let's not talk about this."
• "By the way, there's one more thing..."
• "Stop pretending, damn it."

Judgment: If the full text is all "argumentative transition words," with no "colloquial transition words" → it's like giving a lecture

5. Ending detection (most fatal)

• Check whether there is a "thought question" ending

AI-ending (biggest failure):

思考题：你现在追求的东西，如果没人知道，你还追求吗？

Human ending (powerful):

方案1（反问）：你说呢？
方案2（金句）：欲望是别人的，但命是你自己的。
方案3（戛然而止）：算了，不说了。

Verdict: If the ending is a "thought question" → it directly exposes the AI identity and must be deleted

---

Rewrite suggestions:

Problem 1: Lacks emotional outburst

❌ "心理学家研究了几十年，结论是：绝大多数人一辈子都想不清楚。"

✅ "心理学家研究了几十年，结论他妈的扎心：绝大多数人一辈子都想不清楚。
你以为你40岁了就该想明白了？做梦。"

Problem 2: Lacks a punchline

❌ "这就是问题的核心。"

✅ "27岁是企业的燃料，40岁是家庭的支柱。"

Problem 3: The case lacks vivid imagery

❌ "薛冰雷，35岁，从部门经理位置辞职创业。"

✅ "老王，35岁，部门经理。辞职那天，老板问他想清楚了吗，他说想清楚了。
两年后，他在朋友圈发了条动态：'创业不易，且行且珍惜。'
配图是一张空荡荡的办公室。你猜他想清楚了吗？屁。"

Problem 4: Delete the "thought question"

❌ "思考题：你现在追求的东西，如果没人知道，你还追求吗？"

✅ "下次当你说'我想要'的时候，停下来问自己——这个'想要'，是谁的？

你说呢？"

---

After diagnosis, you must provide:

1. Specifically point out which paragraphs/sentences have an AI flavor (mark the paragraph numbers)
2. What reaction this AI flavor will cause in readers (e.g., "the paragraph lengths are too uniform; readers will feel this is template writing")
3. Directly give rewrite examples, rather than a vague "suggested revision"
4. Priority ranking: which must be changed, which are suggested to change

2. In-depth diagnosis dimensions

• Length redundancy: Identify off-topic content, repeated arguments, or overly long paragraphs (it is recommended that a single paragraph not exceed 200 characters).
• Rhythm and cadence: Check the variation in sentence patterns and paragraph lengths. Mix long and short sentences, with long and short paragraphs interspersed.
• Content repetition: Avoid the same viewpoint or case appearing repeatedly in different chapters.
• Eliminating the "textbook flavor": Replace abstract concepts (such as "cognitive upgrade") with specific behavioral descriptions or concrete scenes.

3. Build an improvement path

• Core problem description: Clearly point out which paragraph and which sentence has the problem.
• Consequence analysis: Explain why the problem affects the reader's experience (e.g., "information overload will cause readers to skim").
• Actionable suggestions: Provide clear actions (delete, merge, rewrite as a rhetorical question, etc.).

4. Output the review report format

The report must include:

• Overall assessment: Give comprehensive feedback on length, rhythm, and expression.
• Revision list: List suggestions by priority (must revise / suggested revision / optional optimization).
• Re-review invitation: Inform the user that after revising, they can send it again for verification.

Examples

User: Help me review this article (excerpt):

The harm of content scraping

First, content scraping seriously harms the interests of original creators.Second, it damages the healthy development of the content ecosystem.Finally, this behavior will lead to a trust crisis across the entire industry.

How to identify content scraping?

Identifying content scraping requires mastering the following key techniques: Technique 1: Reverse search... Technique 2: Comparative reading...

AI：

AI-flavor diagnosis (fatal problems):

1. Overuse of subheadings — a "##" appears every few paragraphs; this is a typical AI output format, and humans writing WeChat public account articles would not do this
2. "First... Second... Finally..." three-part structure — readers know immediately upon seeing this format that it was written by AI, and skip it
3. "Technique 1:"This kind of numbered bolding — too neat, like a PPT outline rather than an article

Rewrite example: Cut that "first/secondly/finally" stuff and just write:

What's the most disgusting part about content spinning? Something an original creator spent three months writing gets "spun" by someone in half an hour, and it gets ten times more reads than yours. Aren't you pissed? What's even more infuriating is that if you want to sue them, the lawyer fees might cost more than the compensation.

This is what real human speech looks like. No need for subheadings, no need for "first, second," just say it all in one breath.”

Best Practices

• Removing AI flavor is the top priority: No problem is more serious than "this article looks like it was written by AI." Every other problem can be tolerated, but not this one.
• Basic detection must be performed: Paragraph length standard deviation, emotional fluctuation, perfect ending, textbook-style expression — these 4 items must be checked.
• Quantitative detection takes priority: Problems that can be explained with numbers (such as a standard deviation < 50 characters) are more convincing than subjective judgments.
• No soft evaluations: Avoid vague wording like "it's just so-so"; directly say "cut half of the case study in paragraph 3."
• Distinguish primary from secondary: AI flavor > structural problems > logical problems > rhetorical elegance.
• Provide rewrite examples: For every problem pointed out, a "human-speak version" rewrite must be attached; you cannot just say "suggested revision."

Version History

• v2.2.0 (2026-01-19): Added "blandness detection" to detect dissemination problems such as lack of emotion, memorable lines, and visual imagery in an article.
• v2.1.0 (2026-01-19): Added "fake celebrity case detection" to solve the problem of AI using cases with stranger names + ages + occupations.
• v2.0.0 (2025-12-26): Major upgrade! Added 10 quantitative AI detection rules, including paragraph length standard deviation, emotional fluctuation, sentence pattern repetition, and more.
• v1.1.0 (2025-12-20): Added "AI flavor detection" as the highest-priority diagnostic dimension.
• v1.0.0 (2025-12-20): Initial version, optimized based on official Skill specifications.