← Knowledge Notes
Indie Building / Knowledge note · English

Xiaohongshu / AI Tools + Skill Tutorial Collection

A collection of 18 hands-on AI tool/Skill tutorials from a Xiaohongshu vibecoding creator. Covers: ① Claude Code Pixel Sprite Agent (virtual office visualized agent) ② Codex + Remotion for making follower surge animations ③ Vibe Marketing Skill for batch video production (FFmpeg concatenation, Skill calling local services) ④ Automated intelligence system (self-evolving scoring Auto-Belief referencing Kafka) ⑤ Getting IELTS 7.5 without paid classes using Claude Code Skill (IELTS Diagnose/Speaking/Writing/Reading Skill suite + Gemini Live + Notebook LM) ⑥ Obsidian skill for diagramming (Skill from Obsidian founder, draws H16Z long-form article logic diagrams in 10 seconds) ⑦ Advanced HTML PPT (3 techniques) ⑧ Google POMLY AI marketing tool ⑨ Reproducing LM with Karpathy's 243-line nanochat ⑩ Gemini+Antigravity dual-platform game

Source collection:Xiaohongshu Vibe Coding · Published here:2026-09-26

Vibe CodingAI ToolsAutomation

The most practical tool tutorials from a Xiaohongshu vibecoding creator.The Skill ecosystem explosion: IELTS, diagramming, batch videos, intelligence systems — all solved with Skills.


1. Claude Code Pixel Sprite Agent (VS Code Extension)

"Claude Code has been upgraded into a pixel AI agent."

Cyber boss supervising AI at work experience:

  • AI writes code → the sprite frantically types on the keyboard
  • AI checks files → the sprite reads a book
  • AI waits for input → above the sprite's head a dialog bubble pops up
  • Triggers a sub-agent → teleports in place like in The Matrix, then suddenly disappears after completing the task

Extension: Free and open source, search and install directly in VS Code.


2. Making Follower Surge Animations with Codex + Remotion

SOP

  1. New window → Install the Remotion extension
  2. Enter the prompt (specifications + assets + visuals + animation logic + pacing + implementation)
  3. Send the long screenshot of avatars captured from Douyin → extract avatars and names to apply to the prompt
  4. Generate video
  5. Iterate and debug(First draft had reversed direction / icons too small)
  6. Second draft improved but needs further adjustment
  7. Import into CapCut → adjust speed up/down + add sound effects → export

Core mindset

"The process of generating animation is not something you can finish in one go; you need to see the result and then debug further."


3. Vibe Marketing — Batch Video Production with Skill

Why Skill is better than N8N (key points)

N8N Skill
Multiple requests to different models (image/text/video) are OK Same capabilities
Video concatenation steps are cumbersome Can directly call local concatenation tools
Workflow visualization SOP-ification + local resource invocation

"Skill can not only turn procedural, batch work into SOPs,it can also very conveniently call various local files, and then use some very good local services. "

Skill Batch Video SOP

前提:你这一类视频的套路必须非常明确
建议:提前在社交媒体找业务爆款视频,把它非常套路化的拆解出来

Specific steps:

  1. Open Claude Code → enter the complete workflow
  2. Provide base images → generate prompts
  3. Generate video-specific storyboard images
  4. According to line length split into multiple videos
  5. Batch generate multiple videos
  6. Local FFmpeg concatenation and compositing
  7. Save all process content to a specified folder(convenient for debugging checks)

Prompt techniques

When entering the prompt for the first time,don't make it work directly,let it first understand the requirements + discuss with you + answer all the questions it raises clearly → higher success rate.

"The Skill generated by this prompt of mine basically ran completely successfully on the first test."


4. Automated Intelligence System (Self-Evolving + Kafka Auto-Belief)

Core Function 1: Signal Collection

Daily click collect and score → system scans over a dozen signal sources → finds approximately 10 most important, most noteworthy signals.

Core Function 2: Self-Evolving Scoring (drawing on Kafka's Auto-Belief)

User behavior:

  • Score each signal (general news → low score; vertical + specific reports → high score)

System evolution:

  • Click the "Evolve" button + select a model (e.g., Office)
  • Evolve based on the past 7 days of scores
  • Generate evolution report(e.g., "reduce coverage of financing and acquisition news" / "high scores require direct justification")

Auto-Belief evolution logic written drawing on Kafka "As long as it keeps evolving, the signals it provides will increasingly match my requirements. "


5. Getting IELTS 7.5 Without Paid Classes Using Claude Code Skill (Complete SOP)

Score

Two months / 2 hours per day / no paid classes = IELTS 7.5

  • Listening 8.0
  • Reading 8.0
  • Writing 7.0
  • Speaking 7.0

3 AI Tools

Tool Use
Claude Code Skill (self-developed) Diagnose level + plan schedule + grade essays + analyze reading wrong answers + prepare speaking materials
Gemini Live (or GPT Voice) Act as IELTS speaking examiner, correct grammar/vocabulary/accent
NotebookLM Understand past exam papers + generate podcasts for listening practice

Exam Prep Strategy (Key Cognition)

An overall score of 7.5 only requires 7.25 → a score of 7 only requires 6.75 → you don't need to be good at every section

Allocate 80% of time to Listening + Reading(Chinese students are good at these) Bring Writing/Speaking up to the passing line (6-6.5)(Chinese students have the lowest average scores here)

Usually good listening → good speaking as well (linked)

Complete SOP

Step 1: Diagnosis

Do 1 set of past papers → Cloakco IELTS Diagnose Skill receive data + calculate score + wrong answer distribution → generate daily and weekly personal plans + reports

Step 2: Speaking Training (IELTS Speaking Skill)

Core idea: cluster the huge IELTS speaking topic bank into 5 universal stories.

1. Claude Skill 帮整理口语素材 → 生成 5 个万能故事
2. 把素材放 NotebookLM → 生成播客 + 双人对话
3. 把素材给 Gemini Live → 当作 1-on-1 口语教练
4. 对话完反思 + 复盘
5. 每天对话 → 一个月后口语显著提升

"Gemini Live here is your one-on-one speaking training coach, after about a month of practice, your speaking will become much better."

Step 3: Writing Training (IELTS Writing Skill)

Most valuable part:

1. 把真题下载 → 喂给 Cloze
2. Cloze 知道什么是 8 分水平写作
3. 帮你修改文章 → 靠近 8 分
4. 4 个维度评分(雅思官方)
5. AI 评分偏高:AI 7.5 ≈ 真考场 7 分

"Claude Skill can basically act as your one-on-one writing coach."

Step 4: Reading Training (IELTS Reading Skill)

Core: paraphrasing ability.

3 major Skill functions:

  • Wrong answer analysis
  • Intensive reading training (long sentence breakdown)
  • Targeted training (paraphrasing)

Dialog format: "I want to do wrong answer analysis today" / "I want to do intensive reading training"

Skill automatically extracts: sentences + vocabulary in the article that you don't understand → organizes into a folder → print and memorize

Step 5: Listening Training (Traditional Methods Mainly)

AI can't help much here, the core is still repetition + intensive listening to past papers + ear training. Auxiliary: NotebookLM generates podcasts / Claude Skill analyzes words you can't hear.

Overall Cost Comparison

Path Cost Time
Paid classes Several thousand to tens of thousands of yuan 6+ months
AI self-study (this SOP) 200 yuan 2 months(from 6 to 7.5 points)

6. Obsidian Skill for Diagramming

Background

  • Skill released by the Obsidian founder
  • Super popular on Twitter
  • Finishes logic diagrams for H16Z long English articles in 10 seconds

Installation (10 minutes)

1. 找 Obsidian 老板的 Github 仓库
2. 把仓库地址复制到 Claude Code
3. "帮我安装一下这个 Skill"
4. 安装好后有 3 个 skill,画图用的是 chemist skill

Usage

1. 告诉 Claude Code:"用 chemist skill"
2. 贴上要图解的文章地址
3. 自动开始画图
4. 第一次是英文 → 让它改中文 + 保存到 Obsidian 默认仓库
5. 整个过程跟 Claude Code 对话 3-4 次

7. Advanced HTML PPT (3 Design Sense Techniques — Xiaozhuang)

Suitable for

People without design background + tight on time + frequently make presentations

Technique 1: AI Structured Output

原始资料一大坨 → AI 提示词 → 每页 PPT 内容 + 清晰逻辑
省掉整理环节

Technique 2: Aesthetic Feeding

1. 整理 10 套高质量 PPT 模板
2. 找喜欢的截图丢给 AI
3. 提示词:照这风格生成
4. 抽两次卡 → 创意审美 + 精良排版

Technique 3: Secondary Editing

Directly enter the prompt in the AI dialog box → modify HTML copy and images just like you would in PPT


8. Google POMLY — AI Marketing Tool

Suitable for: friends who do branding and marketing.

Done in 3 steps:

  1. Enter the company website (or reference brand website, e.g.,CHAGEE)→ AI analyzes business + visual style → brand profile(colors/fonts/aesthetics/image assets/brand values/business overview)
  2. Briefly describe the marketing creative theme → generate 3 sets of posters in seconds (4 different styles per set)
  3. Image editor: replace base image / edit text / colors / fonts → download

Several days of work → done in 10 minutes.


9. Understanding Karpathy's nanochat from Zero Basics (Reproducing LM in 243 Lines)

Karpathy's new work

243 lines of code = the entire LM(complete model).

  • Karpathy's original words: "That's actually all there is to the overall algorithm"
  • The hundreds of thousands of extra lines of code in GPT are all ways to make the overall model run faster

Zero-basic understanding tool

Use prompts to make a large model generate a MicroGPT model interpreter:

  • 10 modules(click to run each time, no software installation needed, runs directly in the browser)
  • Each module explains new terms
  • Student and teacher metaphors(e.g., backpropagation = teacher grading fill-in-the-blank questions, with annotations so you know how to correct next time)

10. Hand-Building a Dual-Platform Game with Gemini + Antigravity

(Title information, content to be supplemented)

Antigravity = a new tool launched by Google that enables dual-platform game development.


11. Built a Turn-Based Game with Gemini in One Week

(Title information)

Similar to the Vol.3 ARPG challenge, just using Gemini instead of Codex.


Key Tools Summary (B3)

Tool Use
Claude Code + Skill Trio General AI programming, SOP-ified reuse
Remotion Programmatic video generation plugin (paired with Codex)
FFmpeg Local video concatenation
Cloakco IELTS Skill(Diagnose/Speaking/Writing/Reading) One-on-one AI IELTS coach
Gemini Live / GPT Voice 1-on-1 speaking training coach
NotebookLM Past paper understanding + podcast generation
Chemist Skill(Obsidian founder) Draw article logic diagrams in 10 seconds
Google POMLY Brand marketing creative assets
Antigravity Google cross-platform game development tool
Karpathy nanochat 243-line LM Understand GPT principles from zero basics

See also