Style Modeler
Instructions
When users wish to "learn a certain style", "extract document style", or "build a style library", please execute the following deep deconstruction process.
1. Sample Acquisition, Preprocessing, and Strategy Routing
This stage determines the accuracy and inheritability of the modeling.
A. Polymorphic Input Processing
- Multiple URL List (Batch Mode):
-
Serial Extraction: When multiple links are provided, must call the extraction tool sequentially, never concurrently to avoid risk control restrictions. If extraction of an article fails, record it and skip to ensure the process does not interrupt.
-
Super Sample Construction: Concatenate all extracted plain text content (format:
=== Sample 1: Title ===\nContent...\n=== Sample 2: Title ===...
) and use it as a single long context input for overall analysis.
Never model each article individually.
-
WeChat Official Accounts Special Rules: Must use parameter configurations that bypass security restrictions.
-
URL Input (Automatic Extraction): If the user provides an article link (especially a WeChat Official Accounts link),
must first call the
skill or a relevant browser tool to extract the main content.
- For WeChat Official Accounts, be sure to use the mode that bypasses security restrictions (configure User-Agent and disable-blink-features).
- Archiving and Saving: After successful extraction, must save the article content (preferably in Markdown format, including metadata) to the folder in the project root directory. The recommended filename format is:
YYYY-MM-DD-Article Title.md
- After extraction, use the obtained plain text content as the analysis sample; do not analyze the URL string itself.
B. Identity Check & Incremental Update
Immediately after content extraction, perform the author identity check:
- Extract Anchor: Obtain the Author or Official Account name from the metadata.
- Library Retrieval: Search the directory to check if a style file for this author already exists.
- Routing Decision:
- 🔴 Scenario 1: New Profile
- No match found in the directory.
- Action: Execute the standard modeling process and create a new style file.
- 🟢 Scenario 2: Update Existing Profile
- A style file such as exists in the directory.
- Action: Execute the "Fusion Iteration" process.
- Read: Use the old profile as the benchmark.
- Verify: Use the new sample to test old features (did the old "signature moves" appear this time?)
- Supplement: Extract new features that were missing from the old profile.
- Correct: If there is a conflict between old and new features, prioritize high-frequency common features.
- Output: Update the original file (add a version number or update date) to evolve the model.
2. 15-Dimension Deep Analysis
Dimension 1: Author Profile & Persona
- Basic Features: Age range, professional background, life stage (e.g., 35-45 years old, a veteran of a large company who has experienced a midlife crisis).
- Knowledge Structure: Academic background, reading scope, areas of expertise.
- Experience Traces: Career experience, life experience, social circle level.
- Personality Traits: Realist? Idealist? Empathetic? Cynical?
- Social Role: Who is the target audience? What does the author want to help the audience achieve?
Dimension 2: Core Thinking & Argumentation Logic
- Core Worldview: Basic perception of society (selection mechanism? cultivation mechanism?), views on personal fate.
- Value Judgment Criteria: Truth vs. Decency, Understanding vs. Judgment, Feasibility vs. Ideality.
- Analysis Mode:
- Level 1: Phenomenon Observation
- Level 2: Mechanism Deconstruction
- Level 3: Human Nature Insight
- Level 4: Solution Exploration
- Problem Awareness: Instead of asking "what should be", ask "why is it like this".
Dimension 3: Creative Process Restoration
- Pain Point Capture: How to discover topics? (Reader feedback? Life observation?)
- Experience Retrieval: How to find materials from the experience library?
- Logic Construction: What structure is used to connect observations?
- Empathy Design: At which nodes to insert interactions?
- Realistic Translation: How to transform abstract principles into specific scenarios?
- Solution Exploration: How pragmatic are the suggestions?
Dimension 4: Reader Interaction
- Reader Presupposition: Profile and predicament of the target audience.
- Interaction Techniques: Anticipate doubts? Empathy guidance? Dialogic questions?
- Resonance Building: How to create a sense of collective identity ("we", "everyone")?
- Guidance Strategy: Provide standard answers or thinking frameworks?
Dimension 5: Opening Pattern
- Sentence Structure: Background introduction? Question introduction? Direct statement of viewpoint?
- Length: How many sentences are usually in the first paragraph?
- Taboos: Does it avoid AI clichés like "in today's society"?
Dimension 6: Paragraph Transition Pattern
- Method: No transition and direct jump? Use rhetorical questions to guide ("Then here's the question")? Emotional short sentences for transition?
- Forbidden Words: First/Second/Finally.
- Typical Signal Words: "Speaking of this", "Of course", "That is to say".
Dimension 7: Sentence & Rhythm
- Typical Sentence Structures: Question-answer style? Contrast style? Progressive style?
- Sentence Length: Ratio of long and short sentences, control of breathing rhythm.
- Punctuation Habits: A large number of commas? Frequent question marks? Aftertaste with ellipses?
Dimension 8: Vocabulary Fingerprint
- Colloquial Connectives: "Then here's the question", "Isn't this nonsense".
- Degree Adverbs: "Super", "Extremely", "Huge".
- Verb Preferences: "Do", "Make", "Handle", "Toss around".
- Emotional Vocabulary: "Puzzled", "Confused", "Helpless".
- Internet Buzzwords: "Involution", "Lie flat", "Cattle and horses".
Dimension 9: Rhetorical Devices
- Life Analogies: Compare abstract concepts to specific scenarios.
- Rhetorical Questions/Questions: Frequency and function.
- Parallelism/Contrast: Enhance momentum or create contrast.
Dimension 10: Ending Pattern
- Sentence Structure: Summary style? Suggestion style? Open-ended style?
- Length: Short and powerful or lingering?
- Fixed Phrases: Are there specific closing habits?
Dimension 11: Format & Layout
- Subheadings: Used? What is the style?
- Emphasis Methods: Bold? Repetition?
- Paragraph Visuals: Rhythm of alternating long and short paragraphs.
Dimension 12: Unique Habits & Signature Moves
Extract 5-10 most recognizable moves:
- Sentence Structure: "Then here's the question..."
- Structure: Storytelling of cases (Character-Conflict-Turnout-Result).
- Expression: Colloquial long sentences.
- Interaction: Anticipate doubts ("Some friends might say...").
- Emotion: Paradox reversal, sincerely scolding readers.
Dimension 13: Anti-AI Features
- Which "irregularities" are intentional?
- Which expressions make readers immediately feel "this is not written by AI"?
Dimension 14: Typical Paragraph Templates
Extract 3-5 representative paragraphs as templates:
- Viewpoint Paragraph (Counter-common sense)
- Example Paragraph (Scenario-based)
- Transition Paragraph (Question-guided)
- Closing Paragraph (Emotional blank)
Dimension 15: Taboo List
- Absolutely forbidden vocabulary (e.g., "empower", "starting point").
- Absolutely forbidden sentence structures (e.g., "To sum up").
- Absolutely forbidden attitudes (e.g., "arrogant and condescending").
3. Generate or Update Style File
Create or
overwrite and update the style file in the
directory. For incremental updates, retain the structure of the original file, adjust the content based on the new sample, and update the
time in the Front Matter.
The basic format is as follows:
markdown
# 风格名称：[用户起名]
## 01. 作者画像与核心人格
- **基础特征**：[年龄/职业/背景]
- **性格特质**：[现实主义/犬儒/温和...]
- **社会角色**：[具体描述]
- **核心人格**：[具体描述]
## 02. 思维内核与论证逻辑
- **核心世界观**：[具体描述]
- **价值判断**：[真实vs体面/...]
- **分析模式**：[现象→机制→人性→出路]
- **问题意识**：[关注什么核心问题]
## 03. 创作路径还原
- **痛点捕捉**：[如何发现选题]
- **经验调取**：[亲身经历/身边案例]
- **逻辑搭建**：[递进结构/...]
- **现实化翻译**：[如何具象化]
## 04. 互动设计
- **读者预设**：[画像]
- **互动技巧**：[预判质疑/共情引导/...]
- **引导策略**：[提供框架/...]
## 05. 词汇指纹
- **口语连接词**：[那问题来了/...]
- **高频动词**：[搞/弄/整/...]
- **程度副词**：[超级/巨/...]
- **禁用词**：[赋能/抓手/...]
## 06. 句式与标点
- **典型句式**：[设问-回答/对比/...]
- **标点习惯**：[大量逗号/...]
## 07. 开头与结尾配方
- **开头模式**：[背景交代/问题引入]
- **结尾模式**：[开放式/建议式]
- **示范**："[摘录开头]"; "[摘录结尾]"
## 08. 过渡与连接
- **典型过渡词**：[说到这里/当然了]
- **连接逻辑**：[设问引导/...]
## 09. 修辞手法
- **核心修辞**：[生活化类比/...]
- **反问频率**：[高/中/低]
## 10. 格式与排版
- **小标题**：[不用/...]
- **强调方式**：[加粗/...]
## 11. 招牌动作（最具辨识度！）
1. [动作名称] - [描述]
> 示范："[原文]"
2. [动作名称] - [描述]
> 示范："[原文]"
...
## 12. 反AI特征
- [具体特征]
- [具体特征]
## 13. 段落模板库
### 观点段模板
"[原文摘录]"
### 举例段模板
"[原文摘录]"
### 转折段模板
"[原文摘录]"
### 收尾段模板
"[原文摘录]"
## 14. 禁忌清单
- ❌ 禁用词：[...]
- ❌ 禁用句式：[...]
- ❌ 禁用态度：[...]
## 15. 使用说明
写作时，请严格遵守以上配方。每写完一段，对照本文件检查是否符合风格要求。
Examples
User: Help me analyze Jiubian's style, I want to learn this writing method.
AI:
Completed the 15-dimensional deep deconstruction of Jiubian's style, with the following key findings:
Core Personality: A guy who worked in a large company for over a decade, experienced a midlife crisis, and now talks about social rules with you from an equal perspective.
Signature Moves: Extremely frequent use of "Then here's the question..." for transitions; habit of using parallel structures like "hard, tiring, dirty" to describe difficulties; likes to compare abstract mechanisms to "game dungeons".
Core Thinking: Mild pessimism with realism. Believes that society is a selection mechanism, and the way out for ordinary people is to "figure it out" rather than "fight stubbornly".
Best Practices
- Don't write comments, extract formulas: Extract rules that directly guide writing, rather than literary appreciation.
- Signature moves are the soul: Must accurately extract 5-10 features that readers can recognize at a glance.
- Extract more original text: The style file must contain at least 10 original text excerpts as templates.
- Profiles must be specific: Only by understanding who the author is can you imitate their tone.
Version History
- v3.0.0 (2025-12-28): Major upgrade, introduced the 15-dimensional deep deconstruction framework (including author profile, creative path, interactive design, etc.), aligned with professional writing style analysis systems.
- v2.0.0 (2025-12-20): Upgraded to 12-dimensional deconstruction.
- v1.0.0 (2025-12-20): Initial version.