将 AI 变成自媒体内容策划与创作引擎——不是模板填空，而是基于 11 个内容洞见维度的系统性深度创作。
- 多平台适配创作：针对微信公众号（长文深度）、小红书（种草分享）、抖音（短快节奏）三种完全不同风格的写作
- 11 维度内容构思：核心观点、副观点、说服策略、情绪触发、金句、情感曲线、情感层次、论证多样性、视角转化、语言风格、互动钩子——在动笔前完成系统性思考
- 5 策略标题生成：好奇心缺口、数据冲击、痛点共鸣、反常识、社交货币——每种策略都有明确的心理学基础
- AI 生图 prompt 配套：封面图 + 正文配图的结构化 prompt，可直接喂给图片生成工具
- Markdown 结构输出：标准化文件格式，包含标题、正文、备选标题、配图指导
当用户提到”写文章”、“帮我创作”、“内容创作”、“公众号文章”、“小红书笔记”、“抖音文案”、“标题怎么起”、“爆款文案”等自媒体内容创作需求时自动触发。也适用于给出主题后要求生成完整内容的场景。
One-Line Summary
Section titled “One-Line Summary”Transform AI into a self-media content strategy and creation engine — not template filling, but systematic deep creation based on 11 content insight dimensions.
Core Capabilities
Section titled “Core Capabilities”- Multi-platform adaptive creation: Three completely different writing styles for WeChat Official Accounts (long-form depth), Xiaohongshu (lifestyle sharing), and Douyin (short, fast-paced)
- 11-dimension content ideation: Core thesis, supporting points, persuasion strategy, emotional triggers, quotable lines, emotional curve, emotional depth, argument diversity, perspective shifts, language style, interaction hooks — systematic thinking before writing begins
- 5-strategy title generation: Curiosity gap, data impact, pain-point resonance, counter-intuition, social currency — each with clear psychological grounding
- AI image prompt companion: Structured prompts for cover images and in-article illustrations, ready for image generation tools
- Markdown structured output: Standardized file format with title, body, alternative titles, and illustration guidance
Trigger Scenarios
Section titled “Trigger Scenarios”Automatically triggers when users mention “write an article”, “help me create”, “content creation”, “WeChat article”, “Xiaohongshu note”, “Douyin copy”, “help me with titles”, “viral content” or similar self-media content creation needs. Also applies when users provide a topic and request complete content generation.
File Inventory
Section titled “File Inventory”目录结构分析
Section titled “目录结构分析”Viral Writer 采用极简的”纯指令型”架构——整个 skill 只有两个 Markdown 文件，没有任何脚本、模板或外部依赖。SKILL.md（240 行）是核心指令文件，包含完整的创作方法论和工作流；README.md（187 行）是面向人类用户的说明文档，包含快速开始、测试数据和工作原理图。
这种”零依赖”设计是纯指令型 skill 的典范——当 skill 的目标是引导 AI 完成一个复杂的思考过程（而非执行代码）时，所有逻辑都可以编码在结构化的自然语言指令中。没有脚本意味着零维护成本、零环境依赖、以及最高的跨平台兼容性。
SKILL.md 结构解析
Section titled “SKILL.md 结构解析”SKILL.md 采用 YAML 前置元数据 + 六步工作流 + 原则约束 的三层架构：
第一层 — YAML Frontmatter：包含 name、description（中文描述 + 触发场景）、以及完整的触发关键词列表。description 字段不仅声明了 skill 的功能，还详细列举了触发短语（“写文章”、“公众号文章”、“小红书笔记”等），这是 skill 能被正确路由的关键。
第二层 — 六步创作流程：这是 skill 的核心引擎。从确认需求（主题、平台、受众、风格）到最终 Markdown 输出，每一步都有明确的标准和约束。关键在于第二步”内容构思”——11 个维度的分析不是输出给用户看的，而是 AI 的”内部思考过程”。这种”隐式推理”设计非常巧妙：它让 AI 在后台完成了高质量的分析，但用户看到的只是最终的创作成果。
第三层 — 重要原则：5 条原则作为对六步流程的补充和约束，防止 AI 在边界情况下的行为偏差。例如”平台适配不是缩写”告诫 AI 不要偷懒把同一篇文章简单缩短，而应该用完全不同的表达方式。
Directory Structure Analysis
Section titled “Directory Structure Analysis”Viral Writer adopts a minimalist “pure instruction” architecture — the entire skill consists of just two Markdown files with no scripts, templates, or external dependencies. SKILL.md (240 lines) is the core instruction file containing the complete creation methodology and workflow; README.md (187 lines) is human-facing documentation with quick start, test data, and workflow diagrams.
This “zero-dependency” design is exemplary of pure instruction skills — when a skill’s goal is to guide AI through a complex thinking process (rather than executing code), all logic can be encoded in structured natural language instructions. No scripts means zero maintenance cost, zero environment dependencies, and maximum cross-platform compatibility.
SKILL.md Structure Analysis
Section titled “SKILL.md Structure Analysis”SKILL.md employs a three-layer architecture: YAML frontmatter + six-step workflow + guiding principles.
Layer 1 — YAML Frontmatter: Contains name, description (Chinese description + trigger scenarios), and a comprehensive list of trigger keywords. The description field not only declares the skill’s capabilities but also enumerates trigger phrases in detail — this is key to correct skill routing.
Layer 2 — Six-Step Creation Workflow: This is the skill’s core engine. From confirming requirements (topic, platform, audience, style) to final Markdown output, each step has clear standards and constraints. The key is Step 2 “Content Ideation” — the 11-dimension analysis is not displayed to the user but serves as the AI’s “internal thinking process.” This “implicit reasoning” design is ingenious: it lets AI complete high-quality analysis in the background while users only see the final creative output.
Layer 3 — Guiding Principles: Five principles serve as guardrails for the six-step workflow, preventing AI behavior drift in edge cases. For example, “platform adaptation is not abbreviation” warns AI not to lazily shorten the same article but to use completely different expression styles.
Viral Writer 是典型的纯指令型单体 skill——没有脚本模块，所有逻辑依赖自然语言指令的层次化组织。SKILL.md 作为唯一的执行单元，内部通过六步工作流形成”管道式”处理链条。README.md 是独立的文档层，与 skill 执行逻辑无耦合——它只面向人类开发者，不影响 AI 的行为。
Module Relationships
Section titled “Module Relationships”Viral Writer is a classic pure-instruction monolithic skill — no script modules; all logic relies on hierarchically organized natural language instructions. SKILL.md serves as the sole execution unit, internally forming a “pipeline” processing chain through the six-step workflow. README.md is an independent documentation layer decoupled from skill execution logic — it faces human developers only and does not affect AI behavior.
viral-writer 模块关系图
graph TD A[SKILL.md
核心指令 240行] --> B[第一步: 确认需求] B --> C[第二步: 11维度内容构思
隐式推理 不展示] C --> D[第三步: 按平台规范创作] D --> E[第四步: 5策略标题生成] E --> F[第五步: 配图与封面指导] F --> G[第六步: Markdown输出] H[README.md
人类文档 187行] -.->|文档参考| A I[YAML Frontmatter
触发路由] -->|激活| A J[5条原则约束] -->|规范| B J -->|规范| C J -->|规范| D
Viral Writer 不包含任何脚本文件——这是 纯指令型 skill 的标志特征。所有功能通过自然语言指令中的结构化工作流实现。这种设计选择反映了内容创作领域的特点：创作过程本身是一个认知和判断密集型任务，不适合用确定性的代码逻辑来规范；AI 需要的是思考框架和方法论指引，而非自动化脚本。
指令结构深度分析
Section titled “指令结构深度分析”虽然没有脚本，但 SKILL.md 的 240 行指令本身就是一个高度结构化的”程序”。以下是其核心”算法”的逐层解析：
Script-Free Design
Section titled “Script-Free Design”Viral Writer contains no script files — this is the hallmark of a pure instruction skill. All functionality is realized through structured workflows in natural language instructions. This design choice reflects the nature of content creation: the creative process itself is a cognition and judgment intensive task unsuitable for deterministic code logic; AI needs thinking frameworks and methodological guidance, not automation scripts.
Instruction Structure Deep Dive
Section titled “Instruction Structure Deep Dive”While script-free, the 240-line SKILL.md itself is a highly structured “program.” Below is a layer-by-layer analysis of its core “algorithm”:
-
隐式推理架构：11 个维度的分析过程不展示给用户——AI 在”后台”完成深度思考，用户只看到最终的创作成果。这是一种精妙的 UX 设计：用户不需要知道创作方法论也能享受到方法论带来的质量提升。
-
示例驱动的指令设计：情绪触发点用内心独白示例（“这说的不就是我吗”）、互动钩子用完整问句示例（“你遇到过类似的情况吗？”）——这种”show, don’t tell”的指令写法比纯描述性的指令有效得多。
-
智能省略与默认值：第一步需求确认中明确要求”如果用户已说明则直接使用”并提供合理的默认值（平台默认微信公众号、风格根据主题推断），避免了死板的表单式交互。
-
平台适配范式转换：不是”同一篇文章的不同长度版本”而是”完全不同的表达方式”——这个根本性的范式声明防止了 AI 偷懒做成简单的文本压缩/扩展。
-
从方法论到元数据的完整链路：description 中的触发关键词列表使得 skill 能被自动路由匹配，而 skill 内部的 11 维度框架又确保触发后的执行质量——形成了”发现→执行→交付”的闭环。
| 模式 | 描述 | 应用场景 |
|---|---|---|
| 隐式推理 | AI 内部完成复杂分析但不展示过程，用户只看到最终输出 | 需要深度思考但输出应简洁的任务 |
| 示例驱动指令 | 用具象化的示例而非抽象描述来定义行为标准 | 需要 AI 把握语气、风格、调性的场景 |
| 智能省略机制 | 根据已有信息自动补全参数，避免重复提问 | 多参数输入的任务型 skill |
| 平台规范矩阵 | 将多目标平台的差异化为结构化的约束参数表 | 需适配多端/多渠道的 skill |
| 原则约束层 | 在工作流之外增加独立的原则层，防止边界行为偏差 | 复杂工作流需要防抖和容错的场景 |
| Pattern | Description | Use Case |
|---|---|---|
| Implicit Reasoning | AI completes complex analysis internally without displaying the process | Tasks needing deep thought but concise output |
| Example-Driven Instructions | Use concrete examples rather than abstract descriptions to define behavior standards | Scenes requiring AI to grasp tone, style, and nuance |
| Smart Omission | Auto-fill parameters from context, avoiding repetitive questioning | Multi-parameter task-oriented skills |
| Platform Spec Matrix | Structure multi-platform differences as a constraint parameter table | Skills needing multi-endpoint/channel adaptation |
| Principle Guard Layer | Add independent principles beyond the workflow to prevent edge-case drift | Complex workflows needing debouncing and fault tolerance |
Viral Writer 的核心方法论（11 维度内容构思 + 平台规范矩阵）高度可移植。要适配其他语言或平台：换平台规范矩阵（如 Twitter/Instagram/LinkedIn），保持 11 维度分析框架不变；换语言风格参数（修改第 10 维度中的具体参数）；换触发关键词（修改 YAML frontmatter）；保留核心框架（11 维度体系是跨平台、跨语言的通用方法论）。
⚠️ 描述过于具体导致匹配范围过窄：YAML description 中的触发关键词虽然全面，但如果用户使用了列表之外的表达方式（如”生成一篇推文”），可能无法触发 skill。建议定期从实际使用日志中补充高频触发词。
⚠️ 隐式推理无法验证：11 维度的内部思考既不可见也不可验证——如果 AI 跳过了某个维度（如忘记设计互动钩子），用户无法从输出中直接发现。可以在第六步输出中增加一个可选的”创作说明”折叠区域供有需要的用户展开查看。
⚠️ 平台规范需要持续更新：社交媒体平台的规则和用户习惯变化很快（如小红书字数限制、抖音推荐算法偏好），skill 中的平台规范参数需要定期维护。
Design Highlights
Section titled “Design Highlights”-
Implicit Reasoning Architecture: The 11-dimension analysis process is hidden from users — AI completes deep thinking in the “background” while users only see the final creative output. This is an elegant UX design: users benefit from methodology-driven quality without needing to understand the methodology itself.
-
Example-Driven Instruction Design: Emotional triggers use internal monologue examples (“This is exactly me!”), interaction hooks use complete question examples (“Have you encountered similar situations?”) — this “show, don’t tell” instruction style is far more effective than purely descriptive instructions.
-
Smart Omission & Defaults: Step 1 explicitly requires “use directly if already stated by user” and provides reasonable defaults (platform defaults to WeChat, style inferred from topic), avoiding rigid form-like interactions.
-
Platform Adaptation Paradigm Shift: Not “different length versions of the same article” but “completely different expression styles” — this fundamental paradigm declaration prevents AI from lazily doing simple text compression/expansion.
-
Complete Chain from Methodology to Metadata: The trigger keyword list in the description enables automatic skill routing, while the internal 11-dimension framework ensures execution quality after triggering — forming a complete “discovery → execution → delivery” loop.
Reusable Patterns
Section titled “Reusable Patterns”| 模式 | 描述 | 应用场景 |
|---|---|---|
| 隐式推理 | AI 内部完成复杂分析但不展示过程，用户只看到最终输出 | 需要深度思考但输出应简洁的任务 |
| 示例驱动指令 | 用具象化的示例而非抽象描述来定义行为标准 | 需要 AI 把握语气、风格、调性的场景 |
| 智能省略机制 | 根据已有信息自动补全参数，避免重复提问 | 多参数输入的任务型 skill |
| 平台规范矩阵 | 将多目标平台的差异化为结构化的约束参数表 | 需适配多端/多渠道的 skill |
| 原则约束层 | 在工作流之外增加独立的原则层，防止边界行为偏差 | 复杂工作流需要防抖和容错的场景 |
| Pattern | Description | Use Case |
|---|---|---|
| Implicit Reasoning | AI completes complex analysis internally without displaying the process | Tasks needing deep thought but concise output |
| Example-Driven Instructions | Use concrete examples rather than abstract descriptions to define behavior standards | Scenes requiring AI to grasp tone, style, and nuance |
| Smart Omission | Auto-fill parameters from context, avoiding repetitive questioning | Multi-parameter task-oriented skills |
| Platform Spec Matrix | Structure multi-platform differences as a constraint parameter table | Skills needing multi-endpoint/channel adaptation |
| Principle Guard Layer | Add independent principles beyond the workflow to prevent edge-case drift | Complex workflows needing debouncing and fault tolerance |
Porting Guide
Section titled “Porting Guide”Viral Writer’s core methodology (11-dimension content ideation + platform spec matrix) is highly portable. To adapt for other languages or platforms: swap the platform spec matrix (e.g., Twitter/Instagram/LinkedIn), keeping the 11-dimension framework unchanged; swap language style parameters (modify Dimension 10); swap trigger keywords (modify YAML frontmatter); keep the core framework (the 11-dimension system is universal across platforms and languages).
Common Pitfalls
Section titled “Common Pitfalls”⚠️ Over-specific description narrowing match scope: While the YAML description’s trigger keyword list is comprehensive, users using expressions outside the list (like “generate a tweet”) may not trigger the skill. Regularly supplement high-frequency trigger words from actual usage logs.
⚠️ Implicit reasoning cannot be verified: The 11-dimension internal thinking is neither visible nor verifiable — if AI skips a dimension (e.g., forgets to design interaction hooks), users cannot directly detect it from output. Consider adding an optional collapsed “Creation Notes” section in Step 6 output for users who want to expand and review.
⚠️ Platform specs need continuous updates: Social media platform rules and user habits change rapidly (e.g., Xiaohongshu word limits, Douyin algorithm preferences), so platform spec parameters in the skill need regular maintenance.