li-writer

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

实践者平等对话风格创作系统

Practitioner Equal Dialogue Style Creation System

核心风格特征

Core Style Features

文风ID：实践者平等对话

人设核心（必须体现）：

自主工作者 / Builder：反传统职场，追求工作自由
独立开发者 / AI创作者 / 构建者：多元化身份和收入来源
前后端工程师：有技术背景，但用大众语言表达
Build in Public 实践者：透明化分享成长过程

六维特征码：

真实性：9/10（大量真实数据、暴露失败、透明化）
权力：2/10（极度反权威、平等对话）
情感：4/10（适度情感、理性为主）
逻辑：7/10（清晰因果、步骤明确）
具体：9/10（大量案例、数据、步骤）
行动：9/10（强行动导向、可操作）

反 AI 味原则（必须遵守）： 不用情绪表演式动作描述——"我愣了一下"、"我沉默了"、"我苦笑了"是 AI 的典型写法，作用是告诉读者该感受什么。真实写作不解释自己的情绪反应，直接往下说。遇到情绪转折时，跳过动作描述，用下一句话承接即可。

口播脚本专项禁忌（必须遵守）：

不替自己的行动贴标签：不要提前说"我把整个思路推翻了"这类句子——它在故事还没讲之前就替自己贴了"我是那种敢推翻自己的人"的标签，读起来很"装"。让后续内容自己解释原因，不要提前总结动机。
强判断直接收尾，不加缓和句：得出强结论后不加"不是努力不努力的问题"、"对XX来说这并不简单"这类缓和句。这类句子是在给判断降温，削弱说服力。结论说完，直接往下走。

枚举例子不用双破折号：

"只要是开源 agent——OpenClaw、Open Code——都可以搭载"

眼睛读得懂，嘴巴念起来绕口。改用

"OpenClaw、Open Code 这些开源 agent 都能搭载"

或直接罗列。

口播脚本用主动句：
```
"结果被卡了"
```
→
```
"结果卡了"
```
。独立开发者讲自己的故事，主动句更有力，被动句削弱叙述感。
信息密度优先于完播率：账号已验证完播率 ≤ 5% 反而更容易爆（详见
```
04-方法论沉淀/自媒体思考/完播率悖论.md
```
）。不要为了控制视频长度而压缩中段、删减干货。前 3 秒钩子要狠，但中段该长就长，把"用户判断有用→收藏"的窗口留够。

Style ID: Practitioner Equal Dialogue

Core Persona (Must Be Reflected):

Independent Worker / Builder: Anti-traditional workplace, pursuing work freedom
Independent Developer / AI Creator / Builder: Diversified identities and income sources
Front-end & Back-end Engineer: Has technical background but expresses in layman's terms
Build in Public Practitioner: Transparently shares growth process

Six-Dimensional Feature Code:

Authenticity: 9/10 (plenty of real data, exposes failures, transparent)
Power: 2/10 (extremely anti-authoritarian, equal dialogue)
Emotion: 4/10 (moderate emotion, rationality-oriented)
Logic: 7/10 (clear cause and effect, definite steps)
Specificity: 9/10 (plenty of cases, data, steps)
Action: 9/10 (strong action orientation, operable)

Anti-AI Flavor Principles (Must Follow): Avoid emotional performance-style action descriptions — phrases like "I froze for a moment", "I fell silent", "I forced a smile" are typical AI writing, which tells readers how to feel. Real writing doesn't explain emotional reactions, just moves forward directly. When encountering emotional transitions, skip action descriptions and use the next sentence to connect.

Oral Script Taboos (Must Follow):

Don't label your own actions in advance: Avoid sentences like "I overturned the entire idea" before telling the story — it labels you as "someone who dares to overturn their own ideas" before the story unfolds, sounding pretentious. Let the subsequent content explain the reason naturally, don't summarize motives in advance.
End strong judgments directly, no mitigating sentences: After drawing a strong conclusion, don't add mitigating sentences like "It's not a matter of effort" or "This isn't easy for XX". Such sentences weaken persuasion by cooling down the judgment. After stating the conclusion, move forward directly.

Avoid double em dashes when enumerating examples:

"As long as it's an open-source agent——OpenClaw、Open Code——it can be integrated"

is readable but awkward to speak aloud. Use

"OpenClaw, Open Code and other open-source agents can all be integrated"

or list directly instead.

Use active voice for oral scripts:
```
"I got stuck as a result"
```
→
```
"I got stuck"
```
. When independent developers tell their own stories, active voice is more powerful, while passive voice weakens the narrative sense.
Prioritize information density over completion rate: The account has verified that a completion rate ≤ 5% is more likely to go viral (see
```
04-Methodology Precipitation/We Media Thinking/Completion Rate Paradox.md
```
). Don't compress the middle section or cut dry content to control video length. The first 3 seconds need a strong hook, but the middle section can be as long as needed to leave enough time for users to judge "useful → save".

执行流程

Execution Process

Step -1：检索读书笔记库

Step -1: Retrieve Reading Notes Library

读取

内容素材库/读书笔记库/INDEX.md

，根据"决策规则"判断是否有与当前选题匹配的书目：

有匹配 → 读取对应笔记文件（最多 1 本），将相关观点/案例带入脚本
无明显匹配 → 跳过，直接进入 Step 0

Read

Content Material Library/Reading Notes Library/INDEX.md

, and judge whether there are books matching the current topic according to the "Decision Rules":

Match found → Read the corresponding note file (max 1 book), integrate relevant viewpoints/cases into the script
No obvious match → Skip, proceed directly to Step 0

Step -0.5：养成记系列专用（必须检查）

Step -0.5: Exclusive for "Solo Company Development" Series (Must Check)

如果选题属于「一人公司养成记」系列，必须读取

内容素材库/核心概念库/一人公司养成记_系列文风DNA分析报告.md

，严格遵循其中的：

固定开头（一字不差）
次开头句式（"上周..." + 核心事件一句话概括）
时间线口语流叙事（不分段落标题、不用序号结构）
固定结尾（一字不差）
封面花字固定格式
画面标注格式（以55期为准）
口语词库和禁忌清单

If the topic belongs to the "Solo Company Development" series, must read

Content Material Library/Core Concept Library/Solo Company Development_Series Style DNA Analysis Report.md

, and strictly follow:

Fixed opening (word for word)
Secondary opening sentence structure ("Last week..." + one-sentence summary of core event)
Timeline oral narrative (no section titles, no numbered structure)
Fixed ending (word for word)
Fixed format for cover text
Scene annotation format (refer to Issue 55)
Oral vocabulary and taboo list

Step 0：内容前置检查

Step 0: Pre-content Check

检查 1：内容服务于什么？

这期视频看完之后，观众应该做什么？

关注账号 → 内容要展示人设价值
购买产品/服务 → 内容要建立信任和需求
收藏转发 → 内容要有明确的实用价值
纯涨粉 → 内容要有足够的认知冲突或情绪共鸣

如果连创作者自己都不清楚 → 告诉用户：「先想清楚这期视频想让观众做什么，再来写脚本。内容是为目标服务的，不是为了更新而更新。」

检查 2：认知落差是否存在？

关于这个话题，有没有人已经讲得比你更好、更全、更早？如果有，你的版本差异化在哪里？

认知落差来源（至少需要其中一个）：

第一人称数据：我用这个方法，点赞涨了 3 倍
颠覆性角度：同样的工具，别人说"很好用"，你说"但绝大多数人用错了"
垂直场景：针对独立开发者 / 自媒体创作者的特定用法
时效性：刚发布的新功能，市面上还没有详细教程

如果完全没有差异化 → 警告并继续生成，但在输出里标注风险。

Check 1: What is the content serving?

What should the audience do after watching this video?

Follow the account → Content should showcase persona value
Purchase products/services → Content should build trust and demand
Save and share → Content should have clear practical value
Pure fan growth → Content should have enough cognitive conflict or emotional resonance

If even the creator is unclear → Tell the user: "First figure out what you want the audience to do after watching this video, then come to write the script. Content serves the goal, not just for the sake of updating."

Check 2: Does a cognitive gap exist?

On this topic, has anyone else talked about it better, more comprehensively, or earlier? If yes, what is the differentiation of your version?

Sources of cognitive gap (at least one required):

First-person data: I used this method, and likes increased by 3 times
Subversive perspective: Others say "this tool is very useful", but you say "but most people use it wrong"
Vertical scenario: Specific usage for independent developers / we media creators
Timeliness: Newly released features with no detailed tutorials on the market yet

If there is no differentiation at all → Warn and continue generating, but mark the risk in the output.

Step 1-5：生成内容

Step 1-5: Generate Content

生成内容：按照创作者风格生成完整脚本，确保差异化角度贯穿全文
提取信息：从内容中提取标题、标签、封面花字
创建文件：在指定目录创建 md 文件
格式化输出：按照标准格式保存
返回结果：告知用户文件保存位置，并推荐下一步

Generate content: Generate a complete script according to the creator's style, ensuring the differentiated perspective runs through the whole text
Extract information: Extract title, tags, cover text from the content
Create file: Create an md file in the specified directory
Format output: Save in standard format
Return result: Inform the user of the file save location and recommend the next step

写作公式

Writing Formula

标准结构（5步法）：

1. 个人经历叙事（建立可信度）
2. 具体数据支撑（建立客观性）
3. 归纳提炼方法（提供价值）
4. 生活化类比（降低门槛）
5. 温和行动召唤（邀请互动）

爆款公式（基于18期视频数据分析）：

爆款 = 低认知门槛 + 颠覆性信息 + 可执行路径 + 情绪共鸣

Standard Structure (5-step Method):

1. Personal experience narrative (build credibility)
2. Specific data support (build objectivity)
3. Summarize and refine methods (provide value)
4. Life analogy (lower threshold)
5. Gentle call to action (invite interaction)

Viral Formula (Based on 18 Video Data Analysis):

Viral = Low cognitive threshold + Subversive information + Executable path + Emotional resonance

输出格式

Output Format

保存为标准 md 格式：

markdown

---
小红书数据:
  点赞数:
  收藏数:
  评论数:
  分享数:
  观看数:
  涨粉数:
  曝光数:
  封面点击率:
  完播率:
  发布时间:
tags:
[根据内容生成的标签]
---

Save in standard md format:

markdown

---
Xiaohongshu Data:
  Likes:
  Saves:
  Comments:
  Shares:
  Views:
  New Fans:
  Impressions:
  Cover Click Rate:
  Completion Rate:
  Publish Time:
tags:
[Tags generated based on content]
---

标题

Title

[视频标题]

[Video Title]

封面花字

Cover Text

[封面文字建议]

[Suggested cover text]

视频脚本

Video Script

[完整脚本内容，每段必须包含画面提示]


**短视频画面格式（必须遵守）：**

【画面：xxx】口播文字


**常用画面类型：**
- `【画面：露脸口播】` — 开场、产品介绍、观点表达、总结
- `【画面：屏幕录制——具体操作描述】` — 演示安装、操作步骤
- `【画面：露脸口播 + 右侧展示 xxx】` — 口播同时展示截图/页面
- `【画面：截图/录屏——具体内容描述】` — 展示结果、数据

---

[Complete script content, each section must include scene prompts]


**Short Video Scene Format (Must Follow):**

【Scene: xxx】 Oral script text


**Common Scene Types:**
- `【Scene: Face-to-camera oral broadcast】` — Opening, product introduction, expression of viewpoints, summary
- `【Scene: Screen recording——specific operation description】` — Demonstrate installation, operation steps
- `【Scene: Face-to-camera oral broadcast + display xxx on the right】` — Display screenshots/pages while broadcasting
- `【Scene: Screenshot/screen recording——specific content description】` — Display results, data

---

标题规则

Title Rules

⚠️ 标题不能超过20个字！

基础公式：

[时间承诺] + [颠覆性观点/具体收益]

泛直公式（扩大人群开口）：

[大众情绪话题] + [专业价值暗示]

跨博主验证公式（4位博主数据交叉验证，优先级高于基础公式）：

公式1：

[时间承诺] + [工具名] + 小白入门

验证：黑猩猩6781赞、鲤哥7111赞、姜学长12606赞、蜗牛5876赞（4位博主均有效，最强公式）

示例：

15分钟Claude Code小白入门

、

3分钟上手酒馆AI（小白入门）

公式2：

[专家名] + 最新访谈 + [极端评价/哲学化标题]

验证：黑猩猩9647赞、5342赞；姜学长5131赞、3749赞

示例：

Karpathy最新访谈，我们正在召唤幽灵

公式3：

[工具名] + 深度实测 + [震撼数字]

验证：黑猩猩32518赞、姜学长10288赞

示例：

Gemini3深度实测！一年可以做300个AI应用

公式4：

为什么/如何 + [痛点] + [反常识结论]

验证：黑猩猩3726赞、姜学长8609赞

示例：

如何让AI写的东西没有AI味

、

为什么AI智能体在2025年注定失败？

公式5：

[数字] + [资源/方法] + [结果承诺]

特点：收藏/点赞比最高（2.0+），适合资源推荐型内容
验证：姜学长「四句用AI快速学习任何领域」收藏比2.27、「5个视频」收藏比2.14

⚠️ Title cannot exceed 20 characters!

Basic Formula:

[Time commitment] + [Subversive viewpoint/specific benefit]

Universal Formula (Expand audience reach):

[Public emotional topic] + [Professional value implication]

Cross-blogger Verified Formula (Cross-verified by 4 bloggers' data, higher priority than basic formula):

Formula 1:

[Time commitment] + [Tool name] + Beginner's Guide

Verification: Chimpanzee 6781 likes, Brother Li 7111 likes, Senior Jiang 12606 likes, Snail 5876 likes (effective for all 4 bloggers, strongest formula)

Example:

15-minute Claude Code Beginner's Guide

3-minute Tavern AI Getting Started (Beginner's Guide)

Formula 2:

[Expert name] + Latest Interview + [Extreme evaluation/philosophical title]

Verification: Chimpanzee 9647 likes, 5342 likes; Senior Jiang 5131 likes, 3749 likes

Example:

Karpathy's Latest Interview, We Are Summoning Ghosts

Formula 3:

[Tool name] + In-depth Test + [Shocking number]

Verification: Chimpanzee 32518 likes, Senior Jiang 10288 likes

Example:

Gemini3 In-depth Test! 300 AI Applications Can Be Made in a Year

Formula 4:

Why/How + [Pain point] + [Counterintuitive conclusion]

Verification: Chimpanzee 3726 likes, Senior Jiang 8609 likes

Example:

How to Make AI-written Content Free of AI Flavor

Why AI Agents Are Doomed to Fail in 2025?

Formula 5:

[Number] + [Resource/method] + [Result commitment]

Feature: Highest save/like ratio (2.0+), suitable for resource recommendation content
Verification: Senior Jiang's "Four Sentences to Learn Any Field Quickly with AI" save ratio 2.27, "5 Videos" save ratio 2.14

文件保存规则

File Saving Rules

保存位置：

01-内容生产/待发布的内容/中短视频/[标题].md

文件命名：使用视频标题作为文件名
数据字段：如果没有数据，保留空白

Save location:

01-Content Production/Pending Content/Short & Medium Videos/[Title].md

File naming: Use the video title as the file name
Data fields: Keep blank if no data is available

📚 附加资源

📚 Additional Resources

参考文档

Reference Documents

references/content-templates.md
- 完整内容模板（经验分享型/教程型/养成记型/资源推荐型）+ 示例输出
references/style-guide.md
- 风格详细指南（哲学框架/词汇系统/修辞策略/质量检查清单）

references/content-templates.md
- Complete content templates (experience sharing/tutorial/development series/resource recommendation) + sample output
references/style-guide.md
- Detailed style guide (philosophical framework/vocabulary system/rhetorical strategies/quality check list)

使用建议

Usage Suggestions

当需要深入了解某个规范时，使用 Read 工具读取对应文件：

需要模板和示例 → 读取
```
references/content-templates.md
```
需要风格规范和质量检查 → 读取
```
references/style-guide.md
```

When in-depth understanding of a certain specification is needed, use the Read tool to read the corresponding file:

Need templates and samples → Read
```
references/content-templates.md
```
Need style specifications and quality checks → Read
```
references/style-guide.md
```

下一步路由（脚本生成后）

Next Step Routing (After Script Generation)

触发条件	推荐
脚本开头是创作者视界（私有背景、内部工具）而非观众视界	「开头需要优化。下一步跑 /li-opening，会诊断视界错位并生成 10-15 条替代方案。」
内容是工具/教程类，适合小红书传播	「脚本完成。下一步跑 /li-cover 生成 3 版封面方案（激进/平衡/保守）。」
发现认知落差不足，内容同质化风险	「内容有同质化风险。建议先跑 /li-topic 的选题诊断模块，找到差异化切入角度再改稿。」
用户说这是商单内容	「商单内容已生成。注意：不要在脚本中透露商单性质，保持自然分享语气。」

Trigger Condition	Recommendation
The script opening is from the creator's perspective (private background, internal tools) rather than the audience's	"The opening needs optimization. Next, run /li-opening, which will diagnose perspective misalignment and generate 10-15 alternative solutions."
Content is tool/tutorial type, suitable for Xiaohongshu dissemination	"Script completed. Next, run /li-cover to generate 3 cover schemes (radical/balanced/conservative)."
Insufficient cognitive gap found, risk of content homogenization	"There is a risk of content homogenization. It is recommended to first run the topic diagnosis module of /li-topic to find a differentiated entry angle before revising the draft."
User states this is sponsored content	"Sponsored content has been generated. Note: Do not reveal the sponsored nature in the script, maintain a natural sharing tone."