风格建模专家 (Style Modeler)
Style Modeler
指令 (Instructions)
Instructions
当用户希望"学习某种风格"、"提取文档风格"或"建立风格库"时，请执行以下深度解构流程。
When users wish to "learn a certain style", "extract document style", or "build a style library", please execute the following deep deconstruction process.
1. 获取样本、预处理与策略路由
1. Sample Acquisition, Preprocessing, and Strategy Routing
This stage determines the accuracy and inheritability of the modeling.
A. 多态输入处理
A. Polymorphic Input Processing
- 多 URL 列表（批量模式）：
-
串行提取：遇到多个链接时，必须依次调用提取工具，绝不并发以防风控。若某篇失败，记录并跳过，确保流程不中断。
-
超级样本构建：将所有提取的纯文本拼接合并（格式：
=== 样本1：标题 ===\n内容...\n=== 样本2：标题 ===...
），作为单一的长上下文输入进行整体分析。
切勿对每篇文章单独建模。
-
微信公众号特规：务必使用绕过安全限制的参数配置。
-
URL 输入（自动提取）： 如果用户提供文章链接（特别是微信公众号链接），
必须先调用
技能或相关浏览器工具提取正文内容。
- 针对微信公众号，务必使用绕过安全限制的模式（配置 User-Agent 和 disable-blink-features）。
- 归档保存：提取成功后，必须将文章内容（优先 Markdown 格式，包含元数据）保存到项目根目录下的 文件夹中。文件名格式建议为：。如果文件名包含非法字符，请自动替换为下划线。
- 提取后，将获得的纯文本内容作为分析样本，不要分析 URL 字符串本身。
- 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. 身份核验与增量迭代 (Identity Check & Incremental Update)
B. Identity Check & Incremental Update
在提取完内容后，立即执行作者身份检查：
- 提取锚点：从元数据中获取 Author 或公众号名称。
- 库检索：检索 目录，查找是否已存在该作者的风格文件。
- 路由决策：
- 🔴 情况一：完全陌生（New Profile）
- 目录无匹配。
- 动作：执行标准建模流程，新建风格文件。
- 🟢 情况二：旧友重逢（Update Existing）
- 目录存在如 。
- 动作：执行**"融合迭代"**流程。
- 读取旧档案作为基准。
- 验证：用新样本检验旧特征（旧的"招牌动作"这次出现了吗？）。
- 补充：提取旧档案漏掉的新特征。
- 修正：如果新旧特征冲突，以高频共性为准。
- 输出：更新原文件（增加版本号或更新日期），使模型进化。
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. 十五维度深度解构 (15-Dimension Deep Analysis)
2. 15-Dimension Deep Analysis
维度一：作者画像与核心人格 (Author Profile & Persona)
Dimension 1: Author Profile & Persona
- 基础特征： 年龄段、职业背景、人生阶段（如：35-45岁大厂老兵，经历过中年危机）。
- 知识结构： 学科背景、阅读范围、擅长领域。
- 经历痕迹： 职场经历、人生阅历、朋友圈层次。
- 性格特质： 现实主义？理想主义？同理心？犬儒？
- 社会角色： 目标读者是谁？想帮助读者做什么？
- 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?
维度二：思维内核与论证逻辑 (Thinking Core & Logic)
Dimension 2: Core Thinking & Argumentation Logic
- 核心世界观： 对社会的基本认知（筛选机制？培养机制？）、对个人命运的看法。
- 价值判断标准： 真实vs体面、理解vs评判、可行vs理想。
- 分析模式：
- 第一层：现象观察
- 第二层：机制拆解
- 第三层：人性洞察
- 第四层：出路探索
- 问题意识： 不问"应该怎样"，而问"为什么是这样"。
- 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".
维度三：创作路径还原 (Creative Process)
Dimension 3: Creative Process Restoration
- 痛点捕捉： 如何发现选题？（读者反馈？生活观察？）
- 经验调取： 如何从经历库中找素材？
- 逻辑搭建： 采用什么结构串联观察？
- 共情设计： 在哪些节点插入互动？
- 现实化翻译： 如何把抽象道理转化为具体场景？
- 出路探索： 建议的务实程度如何？
- 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?
维度四：互动设计 (Reader Interaction)
Dimension 4: Reader Interaction
- 读者预设： 目标读者的画像和困境。
- 互动技巧： 预判质疑？共情引导？对话式提问？
- 共鸣营造： 如何制造集体感（"咱们"、"大家"）？
- 引导策略： 给标准答案还是思考框架？
- 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?
维度五：开头模式 (Opening Pattern)
Dimension 5: Opening Pattern
- 句式： 背景交代？问题引入？观点直陈？
- 长度： 第一段通常几句话？
- 禁忌： 是否避开"在当今社会"等AI套路？
- 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"?
维度六：段落过渡模式 (Transition Pattern)
Dimension 6: Paragraph Transition Pattern
- 方式： 无过渡直接跳？用反问引导（"那问题来了"）？情绪短句转折？
- 禁用词： 首先/其次/最后。
- 典型信号词： "说到这里"、"当然了"、"也就是说"。
- 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".
维度七：句式与节奏 (Sentence & Rhythm)
Dimension 7: Sentence & Rhythm
- 典型句式： 设问-回答式？对比式？递进式？
- 句子长度： 长短句比例，呼吸感控制。
- 标点习惯： 大量逗号？频繁问号？省略号余味？
- 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?
维度八：词汇指纹 (Vocabulary Fingerprint)
Dimension 8: Vocabulary Fingerprint
- 口语化连接词： "那问题来了"、"这不扯淡吗"。
- 程度副词： "超级"、"极度"、"巨"。
- 动词偏好： "搞"、"弄"、"整"、"折腾"。
- 情感词汇： "纳闷"、"懵逼"、"无奈"。
- 网络流行语： "内卷"、"躺平"、"牛马"。
- 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".
维度九：修辞手法 (Rhetorical Devices)
Dimension 9: Rhetorical Devices
- 生活化类比： 把抽象概念比作具体场景。
- 反问/设问： 使用频率及功能。
- 排比/对比： 增强气势或制造反差。
- Life Analogies: Compare abstract concepts to specific scenarios.
- Rhetorical Questions/Questions: Frequency and function.
- Parallelism/Contrast: Enhance momentum or create contrast.
维度十：结尾模式 (Ending Pattern)
Dimension 10: Ending Pattern
- 句式： 总结式？建议式？留白式？
- 长度： 简短有力还是余音绕梁？
- 固定话术： 是否有特定的收尾习惯？
- Sentence Structure: Summary style? Suggestion style? Open-ended style?
- Length: Short and powerful or lingering?
- Fixed Phrases: Are there specific closing habits?
维度十一：格式与排版 (Format & Layout)
Dimension 11: Format & Layout
- 小标题： 是否使用？风格如何？
- 强调方式： 加粗？重复？
- 段落视觉： 长短交替节奏。
- Subheadings: Used? What is the style?
- Emphasis Methods: Bold? Repetition?
- Paragraph Visuals: Rhythm of alternating long and short paragraphs.
维度十二：独特习惯与招牌动作 (Signature Moves)
Dimension 12: Unique Habits & Signature Moves
提取最具辨识度的 5-10 个动作：
- 句式类： "那问题来了..."
- 结构类： 案例故事化（人物-冲突-转折-结果）。
- 表达类： 口语化超长句。
- 互动类： 预判质疑（"可能有小伙伴说..."）。
- 情感类： 悖论翻转、真诚地骂读者。
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.
维度十三：反AI特征 (Anti-AI Features)
Dimension 13: Anti-AI Features
- 哪些"不规则性"是刻意为之？
- 哪些表达方式会让读者立刻感觉"这不是AI写的"？
- Which "irregularities" are intentional?
- Which expressions make readers immediately feel "this is not written by AI"?
维度十四：典型段落模板 (Paragraph Templates)
Dimension 14: Typical Paragraph Templates
提取 3-5 个代表性段落作为模板：
- 观点段（反常识）
- 举例段（场景化）
- 转折段（设问引导）
- 收尾段（情绪留白）
Extract 3-5 representative paragraphs as templates:
- Viewpoint Paragraph (Counter-common sense)
- Example Paragraph (Scenario-based)
- Transition Paragraph (Question-guided)
- Closing Paragraph (Emotional blank)
维度十五：禁忌清单 (Taboo List)
Dimension 15: Taboo List
- 绝对不能用的词汇（如"赋能"、"抓手"）。
- 绝对不能用的句式（如"综上所述"）。
- 绝对不能有的态度（如"高高在上"）。
- 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. 生成或更新风格文件
3. Generate or Update Style File
在
目录下创建或
覆盖更新风格文件。如果是增量更新，请保留原文件的结构，但根据新样本调整内容，并在 Front Matter 中更新
时间。
基本格式如下：
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:
01. 作者画像与核心人格
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]
- 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
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]
- 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]
- 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]
- 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]
- 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/...]
- 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/...]
- 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/...]
- 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/...]
- 典型句式：[设问-回答/对比/...]
- 标点习惯：[大量逗号/...]
- 典型句式：[设问-回答/对比/...]
- 标点习惯：[大量逗号/...]
- 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]"
- 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]"
- 典型过渡词：[说到这里/当然了]
- 连接逻辑：[设问引导/...]
- 典型过渡词：[说到这里/当然了]
- 连接逻辑：[设问引导/...]
- 核心修辞：[生活化类比/...]
- 反问频率：[高/中/低]
- 核心修辞：[生活化类比/...]
- 反问频率：[高/中/低]
- 小标题：[不用/...]
- 强调方式：[加粗/...]
- 小标题：[不用/...]
- 强调方式：[加粗/...]
11. 招牌动作（最具辨识度！）
11. 招牌动作（最具辨识度！）
- [动作名称] - [描述]
示范："[原文]"
- [动作名称] - [描述]
示范："[原文]"
...
- [动作名称] - [描述]
示范："[原文]"
- [动作名称] - [描述]
示范："[原文]"
...
- ❌ 禁用词：[...]
- ❌ 禁用句式：[...]
- ❌ 禁用态度：[...]
- ❌ 禁用词：[...]
- ❌ 禁用句式：[...]
- ❌ 禁用态度：[...]
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。
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。
用户： 帮我分析九边的风格，我要学习这种写法。
AI：
已完成对九边风格的 15 维度深度解构，核心发现如下：
核心人格： 一个在大厂混过十几年的老哥，经历过中年危机，现在用平视的视角跟你聊社会规则。
招牌动作： 极高频使用"那问题来了..."进行转折；习惯用"苦、累、脏"等排比描述困境；喜欢把抽象机制比作"游戏副本"。
思维内核： 现实主义的温和悲观。认为社会是筛选机制，普通人的出路在于"想明白"而非"死磕"。
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".
最佳实践 (Best Practices)
Best Practices
- 不要写评论，要提取配方： 提取能直接指导写作的规则，而不是文学赏析。
- 招牌动作是灵魂： 必须精准提取 5-10 个让读者一眼认出的特征。
- 多摘录原文： 风格文件中至少要有 10 处原文摘录作为模板。
- 画像要具体： 只有理解了作者是谁，才能模仿他的语气。
- 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.
版本记录 (Version History)
Version History
- v3.0.0 (2025-12-28): 重大升级，引入 15 维度深度解构框架（包含作者画像、创作路径、互动设计等），对齐专业写作风格分析体系。
- v2.0.0 (2025-12-20): 升级为 12 维度解构。
- v1.0.0 (2025-12-20): 初始版本。
- 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.