Files

T

Paweł Domański 64944cf004 May 17, 2026, 11:40 PM

2026-05-18 06:40:19 +00:00

33 KiB

Raw Permalink Blame History

name, description

name	description
content-pipeline	A unified pipeline for content production and distribution. Material collection → Drafting → Layout → Cover → WeChat Moments copywriting → Multi-platform conversion → One-click distribution. Covers WeChat Official Account writing, Xiaohongshu carousel images, Jike copywriting, podcast audio, brand videos, and Chrome CDP automatic publishing.

Content Pipeline

This skill uses the 01fish brand colors as an example by default. You can change it to your own brand colors in the references/ directory, or override the path and personal settings in local/SKILL.local.md.

One-stop solution: Material collection → Article writing → Layout → Multi-platform content → One-click distribution.

Storage Location

[Output Directory]/
├── drafts/
│ └── current.json # Current material list
├── [Article Title].md # Output article
├── [Article Title]_preview.html # Layout preview
└── [Theme]-Xiaohongshu Version.html # Xiaohongshu carousel

The output directory is specified by the user, or a default path is configured in local/SKILL.local.md. If no directory is specified, multi-platform outputs are stored in /tmp/.

Triggering Keywords

Material Collection (Path A)

Triggering Keywords	Explanation
`/story`	View current material status
"Check out the materials"	View recorded materials
"Draft"	Generate article + layout + cover image
"Clear materials"	Clear current materials
"Note: xxx"	Manually add materials
"Material +1: xxx"	Manually add materials
"Write a Moments Post"	Generate Moments caption based on materials/articles

Content Generation (Path B)

Triggering Keywords	Explanation
`/xiaohongshu` + WeChat link	Convert WeChat article to Xiaohongshu carousel
"Convert to Xiaohongshu" + WeChat link	Same as above
"Make it into Xiaohongshu" + WeChat link	Same as above
"Convert to Jike" + WeChat link	Generate Instant Copy
"Share to Podcast" + WeChat Link	Generate Podcast Script + AI Voice
"Make Video" + WeChat Link	Trigger Brand Video Pipeline
"Create Video Canvas" + Any Material	Generate a recordable Prezi-style video canvas (WeChat link/md/pdf/html/text are all acceptable)
"Screen Recording Canvas" + Any Material	Same as above
"Journal Video" + Any Material	Same as above
"Multi-Platform Distribution" + WeChat Link	Generate content for all platforms at once
"Share to Xiaohongshu and Publish" + WeChat Link	Generate + Automatically Trigger Distribution

Article Reading

Trigger Keywords	Explanation
`/read-gzh` + WeChat Link	Crawle and summarize WeChat Official Account articles
"Read this WeChat Official Account for me"	Same as above
"Summarize this article"	Same as above

Layout and Images

Triggering Keywords	Description
"Layout"	Use 01fish theme to format Markdown → WeChat Official Account HTML
"Create Header Image" / "Cover Image"	Generate WeChat Official Account header image HTML (browser downloads PNG)
"Create Vertical Cover" / "Vertical Header Image"	Generate 3:4 vertical cover (1080×1440) from WeChat Official Account cover, suitable for Xiaohongshu/Video Account
"Create Images" / "Prepare Images"	Generate article images HTML (browser downloads PNG)
"Layout + Images" / "Complete Layout"	Layout + header image + images generated together

Distribution

Triggering Keywords	Description
`/distribute`	Read manifest for one-click publishing
"One-Click Publishing"	Publishing to all platforms
"Publish to all platforms"	Same as above
"Publish to Xiaohongshu"	Publish to a single platform
"Publish to Jike"	Publish to a single platform

Core Principle: Utter Honesty

**AI-generated content must be honestly labeled, without any human intervention, showcasing the authentic creative process. **

Honesty Labeling Guidelines

When AI participates in content creation, it must be clearly labeled in the article:


**Research & Writing**: AI (Claude)

**Leader & Proofreader**: [Username]

**Creation Time**: [Actual Time Spent] (Research X minutes + Writing Y minutes)

Prohibited False Statements

❌ Don't Write:

"We spent two weeks researching" (Actual time: a few minutes)
"Based on in-depth interviews" (No interviews)
"Discussed by the team" (No team)
"Author: XXX / Editor: AI Assistant" (AI wrote the entire article)

✅ Should Write:

"This article is based on N publicly available information sources, researched, analyzed, and written by AI"
"Materials collection time: X minutes, writing time: Y minutes"
"Human-led + AI collaboration"

Why Honesty is Important

Building Trust: Readers deserve to know how the content was generated.
Showcasing Value: AI's rapid, high-quality content creation is value in itself and needs no concealment.
Ethical Compliance: AI-generated content should be transparent.
Long-Term Vigilance: Honesty is the cornerstone of a long-term personal brand.

Two Input Paths

Path A: Daily Material Collection → Drafting

Applicable Scenarios: Automatically accumulate materials during daily collaboration with CC, and generate a draft with one click once enough is collected.

Path B: WeChat Link → Multi-Platform Content

Applicable Scenarios: Existing WeChat Official Account articles can be converted into multi-platform content and published with one click.

Path A Workflow: Material Collection → Drafting

Automatic Recording (Enabled by Default)

cc actively identifies and automatically records engaging moments in conversations, without requiring manual triggering.

Signal Recognition:

Type	Signal Recognition	Example
Pitfalls and Failures	Expectations ≠ Results, Errors, Half a Day of Frustration	"Tried three solutions and none worked"
Unexpected Discoveries	"Unexpectedly," "It works," Unexpectedly Effective	"Solved it like that!"
Iterative Refinement	Multiple Versions, From Complex to Simple	"200 lines reduced to 20 lines and it still runs"
Funny Moments	Hilarious Quotes, AI Glitch, Amazing Bugs	"It diligently wrote a bunch of wrong code for me"
Breakthrough Moments	Stuck for a long time but finally got it working	"Finally found the bug that bothered me for a week"
Method Consolidation	Reusable Techniques and Insights	"This is how I'll handle this situation in the future"

Automatic Recording: Does not interrupt dialogue, paragraph end marker (✓ Material +1)

Manual recording

Record when the user says "Record an entry: xxx" or "Material +1: xxx".

current.json Format


{
"topic": "Topic (optional, automatically extracted during drafting)",

"materials": [

{
"time": "2026-01-30 14:30",

"content": "Material content",

"type": "Funny Moments",

"context": "Optional context notes",

"auto": true

}

],

"created": "2026-01-30"

}

Drafting Steps

Read Materials — Read drafts/current.json
Analyze and Extract — Extract the theme and storyline
Determine Content Type → Select Writing Framework:

Content Type	Indicator Signals	Use Framework	References
Tutorials	Teach people how to install/use/configure tools and skills Introduction, Technical Practice, "How to Do xxx"	Six-Part Tutorial Framework	`references/tutorial-framework.md`
In-Depth Article	Industry Analysis, Personal Stories, Trend Judgment, Opinion Output, "Why xxx"	Four-Act In-Depth Framework	`references/writing-style.md`

Tutorial Article Framework (2000-4000 words):


→ I. What is the core concept? (Table + one-sentence definition)

→ II. How to install/use (step-by-step + code blocks + image annotations)

→ III. Practical demonstration (phased + table display + human-computer collaboration)

→ IV. Ready to use (quick installation command + usage method table)

→ Conclusion (elevation + CTA)

In-Depth Article Framework (8000-12000 words):


→ 01 Setting the Background

→ 02 Core Argument

→ 03 Transition/Case Study

→ 04 Elevation/Conclusion

Write the Article — Write the article according to the corresponding framework
Save — Save the Markdown file
Formatting — Use a formatting tool to generate an HTML preview (01fish theme)
Header Image + Accompanying Images — Generate a downloadable HTML file (→ Read references/cover-template.md)

Vertical Cover (Optional): When a user says "make a vertical cover," convert the generated WeChat Official Account header image HTML → Read references/cover-vertical-spec.md

WeChat Moments Copy — Generate WeChat Moments promotional copy (→ Read references/platform-copy.md)
Manifest — Generate manifest.json for use by /distribute. The WeChat section must include:

wechat.markdown: Article Markdown path
wechat.html: Path to the formatted _preview.html
wechat.cover_image: Cover PNG path (user needs to download from browser first)
wechat.title: Article title
wechat.author: Author name (default 01fish)
wechat.digest: Article summary (within 120 characters)
wechat.images: List of PNG image paths (if any)

Ask — Whether to clear current materials

Formatting Commands

cd "$MD_FORMATTER_DIR"
python3 md2wechat_formatter.py [article path] --theme [theme] --font-size [font size]

$MD_FORMATTER_DIR needs to be configured in local/.env or environment variables.

Recommended Themes: 01fish (01fish brand color, default), chinese (Chinese style), apple (minimalist and elegant)

Recommended Font Sizes: medium (15px, default), large (16px, recommended for long articles)

01fish Theme Description: Based on the chinese theme, using the 01fish brand colors (dark green #1A3328 + fish red #C44536 + rice paper background #F2EDE3)

Path B Process: WeChat Link → Multi-Platform Content

Step 1: Fetch Articles

Use a Python script to fetch the article (WeChat has anti-scraping verification; WebFetch will be blocked):

Timeout: 30 seconds. If it fails, prompt the user to manually copy the article text. If a user simply says "Read this article from a WeChat official account" (triggered by /read-gzh), the crawling process will directly generate a structured summary without proceeding to the subsequent content generation process. Summary format:


# Article Summary

## Basic Information (Title/Author/Type/Number of Images)

## Core Viewpoints (3 points)

## Key Information

## Quotes

## Image Content (Download and recognize text in images)

## Reflections/Iteration Points

Step 2: Analyze Article Structure

Extract: Title, Subtitle/Quotes, Core Concepts, Key Data, Steps/Processes, Highlights/Features, Methodology/Quotes, Call to Action.

Step 3: Break it down into cards

8-10 cards, following the Xiaohongshu reading rhythm (→ Read references/xiaohongshu-format.md):

Location	Card Type	Content
Card 1	Cover	Main Title + Hook + Mini Visual Elements
Card 2	See Results First	Finished Product Display + Core Data
Cards 3-4	Concept Explanation	Core Concept Breakdown
Cards 5-7	Process/Practical Application	Steps, Comparisons, Flowcharts
Card 8	Highlights/Features	Product/Work Highlight Card
Card 9	Methodology	One-Sentence Golden Sentence Extraction
Card 10	Call to Action	Links + Community Guidance

Step 4: Generate Image HTML

Output path: [Short Theme]-Xiaohongshu Version.html in the same directory as the article. If no directory is specified, place it in /tmp/. The browser will automatically open the preview.

The last call-to-action page must include: your WeChat ID [Your WeChat ID] (emphasized in large, highlighted text), keywords in the notes, and the core link.

Configure your actual WeChat ID in local/SKILL.local.md.

📚 Important: Read the Example Before Generating

Refer to the quality standards of references/xiaohongshu-examples/观鸟图鉴-范例.html:

✅ Card Design Requirements

Pure infographic design, no article screenshots
Pixel art/gamified interface (where applicable)
Rich visual elements such as flowcharts, card grids, and numbered lists
Restrained use of brand colors (85% dark green + 5% magenta)

✅ Copywriting Quality Requirements

Authentic sharing feel, with real-life scenarios and personal experiences
Conversational expressions: "Going crazy with fun," "Getting hooked," "And then I..."
298-350 words + 8-12 tags

The generated content should reach the professional level of the example.

Step 5: Generate Xiaohongshu Post Copy

Choose a Style Based on Content Type:

Personal IP Style (Real-person sharing, product development, pitfalls/trials)
→ Read the personal brand style file specified in local/SKILL.local.md (if any)
Authentic, chronological feel + specific time details + conversational expression
300-350 words, 8-12 tags
Persona: Customize in local/SKILL.local.md
01fish Style (Methodology summary, in-depth analysis)
→ Read the Xiaohongshu section of references/platform-copy.md
Structured breakdown + high density of practical information
Suitable for reposting articles from WeChat Official Accounts

Step 6: Generate Jike Post Copy

→ Read the Jike section of references/platform-copy.md

Step 7: Generate Podcast Script

→ Read the podcast section of references/platform-copy.md.

Step 8: AI Voice Generation

Use Fish Audio TTS to convert the podcast script to MP3 (→ Read references/tts-config.md).

File naming: [Podcast Title].mp3 + [Podcast Title]-Podcast Script.txt

Step 8.5: Output manifest.json

After all content is generated, manifest.json will be automatically output to the output directory. Format:


{
"version": "1.0",

"created": "<ISO timestamp>",

"source": "<WeChat link>",

"title": "<article title>",

"outputs": {

"xiaohongshu": { "html": "...", "copy": { "title": "...", "body": "...", "tags": [...] } },

"jike": { "copy": { "body": "...", "circles": [...] } },

"xiaoyuzhou": { "audio": "...", "script": "...", "copy": { "title": "...", "description": "...", "show_notes": "..." } },

"video_canvas": { "html": "...", "teleprompter_md": "...", "cover_html": "..." }

}
}

If the user says "share to Xiaohongshu and publish", /distribute will be executed automatically after the manifest is generated.

Step 9: Brand Video Generation (Optional)

Execute only if the user mentions "video", "TikTok", "WeChat Channel", or "brand video":

A. Remotion Brand Intro and Outro

cd "$REMOTION_DIR"

npx remotion render src/index.ts Intro --output /tmp/brand-intro.mp4

npx remotion render src/index.ts Outro --output /tmp/brand-outro.mp4

$REMOTION_DIR needs to be configured in local/.env or environment variables.

B. AI Video Prompt — Generates 4 video prompts for Seedance 2.0 or Google Veo

C. ffmpeg Stitching Command — Generates stitching commands for users to execute manually

Step 9B: Video Canvas Generation (Optional)

Executes only when the user says "Create Video Canvas", "Screen Recording Canvas", or "Journal Video". Accepts any input material: WeChat links, Markdown files, PDFs, HTML files, plain text, and user-generated content are all acceptable.

Content Acquisition — Automatic processing based on input type:

WeChat link: Uses scripts/fetch_wechat_article.py to fetch
File path (md/pdf/html/txt): Reads directly
User-pasted text: Uses directly

Structure Analysis — Extracts title, core data, pain points, steps, principles, comparisons, key phrases, and highlights
Splitting into 9 Cards — Reads references/video-canvas-template.md to obtain the complete CSS+JS template and card specifications
Generate 9 Teleprompter Scripts — Conversational style, 80-150 words per script, including [hint] cue tags
Output Teleprompter Script md — [Short Theme]-Teleprompter Script.md, user-editable
HTML Assembly — CSS framework + HTML skeleton + populated content + JS framework (SCRIPTS and md) (Consistent)
Output File — [Short Theme]-Video Canvas.html, save to the same directory as the article or /tmp/
Generate Cover Image — [Short Theme]-Cover.html, journal style + portrait circle frame, download PNG to the browser
Prompt the User — First check the teleprompter script's Markdown, then open the HTML recording in the browser. 16:9 fixed aspect ratio, direct upload to various platforms

Step 10: User Fine-tuning

Inform the user of all output paths, indicating adjustments are possible, and allowing one-click publishing by entering /distribute.

WeChat Official Account Synchronization Prompt: After downloading the cover PNG from the browser, simply use /distribute --platforms wechat to sync to the drafts (API mode, no need to open Chrome).

Produce five items at once, no additional requirements:

Xiaohongshu image HTML (including one-click download toolbar)
Xiaohongshu post text (title + body + tags)
Instant post text (body + circle tags)
Xiaoyuzhou Podcast (recording script + AI voice MP3)
manifest.json (for one-click publishing via /distribute)

Optional addition in step 9B (when saying "video canvas"):

Video canvas HTML (including recording + teleprompter + beauty filter, fixed 16:9 aspect ratio)
Teleprompter script md (editable; after modification, say "update teleprompter" to sync to HTML)
Cover image HTML (journal style + portrait circle frame, PNG downloadable in browser)

Distribution Process (/distribute)

Read manifest.json and automatically publish to various platforms via Chrome CDP (→ Read references/distribute-platforms.md).

Usage


# Publish to all platforms

npx -y bun "${SKILL_DIR}/scripts/distribute/distribute.ts" --manifest /path/to/manifest.json

# Select platform

npx -y bun "${SKILL_DIR}/scripts/distribute/distribute.ts" --manifest /path/to/manifest.json --platforms xhs,jike

# Preview mode (no submission, only pre-filled content)

npx -y bun "${SKILL_DIR}/scripts/distribute/distribute.ts" --manifest /path/to/manifest.json --platforms xhs --preview

Platform Abbreviations

Abbreviation	Platform	Status
`wechat`	Official Account	Available
`xhs`	Xiaohongshu	Available
`jike`	Jike	Available
`xiaoyuzhou`	Xiaoyuzhou	Available
`douyin`	Douyin	Experimental
`shipinhao`	Video Account	Under Development

Execution Order

Official Account → Xiaohongshu → Jike → Xiaoyuzhou → Douyin → Video Account (Execute in sequence to avoid Chrome port conflicts)

Four-Level Degradation

Level	Mode	Trigger Condition
L0	API Direct Push	Official Account API directly pushes drafts to the draft box, no Chrome required
L1	Automatic Publishing	CDP Fully Automated
L2	Auxiliary Publishing	Login State Invalid/Selector Invalid/`--preview`
L3	Manual Mode	CDP Connection Failure

Official Account takes priority L0 (API), automatically degrading when credentials are missing or fail. L1 (CDP).

Brand Design Guidelines

Two Brand Color Systems:

01fish: Professional content brand (WeChat Official Account, in-depth articles, methodologies)
Fish Head: Real-person IP brand (Xiaohongshu, Jike, daily sharing)

Single Source of Truth: Specify your brand color document path in local/SKILL.local.md.

If color values conflict, the brand document prevails. The following color palette is used as a default example.

01fish Color Palette (Dark Green System)

Proportion Rule: Dark Green 85% : Fish Red 5% : Others 10%

Name	Color Value	Use
Dark Green Main Color	`#1A3328`	Dark Card Background
Rice Paper Background	`#F2EDE3`	Light Card Background
Fish Red	`#C44536`	Accent Color, Numbers, Labels (For Highlighting Only)
Semi-Transparent White	`rgba(255,255,255,0.5)`	Brand Name on Dark Background
Semi-Transparent Dark Green	`rgba(26,51,40,0.4)`	Brand Name on Light Background
Moss Gray	`#7A8C80`	Secondary Text
Dark Ink	`#0F1F18`	Darker Background
Light Cyan	`#D4DDD7`	Divider, Border

Fish Head Color Palette (Peach Pink System)

Proportion Rule: Peach Pink 15% : Creamy Yellow 40% : Twilight Purple 10% : Gray 35%

Name	Color Value	Use
Peach Pink	`#FF6B9D`	Main Emphasis Color, Title, Key Data
Creamy Yellow Background	`#FFF9E6`	Light Background, Card Background
Twilight Purple	`#9D7BA8`	Secondary Color, Secondary Information, Brand Name
Warm Gray	`#6B6B6B`	Body Text
Light Gray Background	`#F5F5F5`	Modern Background
Midnight Blue	`#2D3047`	Dark background (rarely used)

Brand Selection Rules

Content Type	Brand Used	Reason
In-depth articles on WeChat Official Accounts	01fish	Professional, authoritative, content brand
Industry analysis reports	01fish	Calm and objective
Real-person sharing on Xiaohongshu	Yutoutou	Warm, authentic, real-person IP
Instant daily updates	Yutoutou	Lively and approachable
Product development records	Yutoutou	Real process demonstration
B-end product introduction	01fish	Professional and trustworthy

Dual-brand collaboration: The same content uses 01fish color on WeChat Official Accounts and Yutoutou color when reposted on Xiaohongshu.

Font

font-family: -apple-system, 'PingFang SC', 'Microsoft YaHei', sans-serif;

### Brand Corner

Top left corner of each page: `01fish` logo + text. Use `.light` for dark backgrounds and `.dark` for light backgrounds.

### Page Number

Bottom right corner: `1/N`. Use `rgba(255,255,255,0.2)` for dark backgrounds and `rgba(26,51,40,0.2)` for light backgrounds.

---

## Content Rewriting Principles

WeChat → Xiaohongshu is not a direct copy; adaptation is required:

| Dimensions | WeChat          | Xiaohongshu                       |
| ---------- | --------------- | --------------------------------- |
| Length     | 2000-3000 words | 50-80 words per page              |
| Structure  | Linear reading  | Card-style skipping               |
| Tone       | Technical, in-depth | Concise, intuitive, impactful     |
| Visuals    | Text-centric    | Visual-centric, text-embellished |

Rewriting Key Points: Headlines should be impactful, numbers should be large, one key point per page, visuals replace text, retain core links.

---

## Complete Template Reference

When generating Xiaohongshu images for the first time, refer to the example files in the `references/xiaohongshu-examples/` directory for complete CSS + JS.

> If you have additional local template references, specify the path in `local/SKILL.local.md`.

When generating new content, reuse the CSS + JS parts of the example file, replacing only the card content.

---

## Script Directory

**Agent Execution**: Determine this SKILL.md directory as `SKILL_DIR`, then use `${SKILL_DIR}/scripts/<name>`.

| Script                             | Purpose                                            |
| ---------------------------------- | -------------------------------------------------- |
| `scripts/fetch_wechat_article.py`  | WeChat article scraping (Python, simulating WeChat UA) |
| `scripts/distribute/distribute.ts` | Distributor editor                                 |
| `scripts/distribute/cdp-utils.ts`  | Shared CDP tools                                   |
| `scripts/distribute/platforms/.ts`| Platform publishing modules        |

## Reference File Index

cc reads on demand, do not load all references at once.

| Scenario                            | Reading Files                                                                                                                     |
| ----------------------------------- | --------------------------------------------------------------------------------------------------------------------------------- |
| Writing In-Depth Articles           | `references/writing-style.md` — Character Design + Writing Guidelines + Formatting (Four-act structure, 8000-12000 words)           |
| Writing Tutorial Articles           | `references/tutorial-framework.md` — Six-part tutorial framework (Results → Concepts → Operations → Practice → Ready to Use, 2000-4000 words) |
| Generating Cover Images/Images      | `references/cover-template.md` — 01fish style layout guidelines (cover image + images + visual components)                         |
| Horizontal to Vertical Covers       | `references/cover-vertical-spec.md` — CSS conversion guidelines for converting WeChat Official Account covers to vertical format   |
| Generating Xiaohongshu Carousel Images | `references/xiaohongshu-format.md` — HTML templates + visual component library                                                  |
| Generate copy for various platforms   | `references/platform-copy.md` — Xiaohongshu/Jike/Podcast/WeChat Moments copywriting guidelines                                    |
| Generate podcast audio              | `references/tts-config.md` — Fish Audio configuration + generation script                                                         |
| Distribute to various platforms     | `references/distribute-platforms.md` — Platform configuration + manifest format + fallback strategy                               |
| Generate video canvas               | `references/video-canvas-template.md` — Journal collage video canvas template (CSS+JS+card guidelines)                              |

---

## Troubleshooting

| Issue                           | Solution                                              |
| ------------------------------- | ----------------------------------------------------- |
| WeChat scraping failed          | Prompt user to manually copy article text             |
| Article too short (<500 words)    | Compress into 5-6 cards                               |
| Article too long (>5000 words)    | Select core content, keep to 10 cards or less         |
| Exported image is blurry        | Check SCALE=2, browser zoom 100%                      |
| Manifest does not exist         | Prompt to run content generation first                |
| Chrome startup failed           | Downgrade to L3 (manual mode)                         |
| TTS 402 Insufficient balance    | Prompt to recharge at fish.audio                      |
| TTS generation failed           | Only script text is output, prompts for manual recording |

---

## Diagram Potoku


``` mermaid
graph TD
    subgraph "Ścieżka A: Zbieranie materiałów"
        A1[Zbieranie materiałów automatyczne/ręczne] --> A2{"Użytkownik mówi 'Szkic'"};
        A2 --> A3[Generuj artykuł];
        A3 --> A4[Zastosuj układ];
        A4 --> A5[Generuj okładkę];
        A5 --> A6[Generuj podpis do WeChat Moments];
        A6 --> A7[Generuj manifest.json];
    end

    subgraph "Ścieżka B: Link WeChat"
        B1[Link WeChat] --> B2[Pobierz artykuł];
        B2 --> B3[Analizuj strukturę];
        B3 --> B4[Generuj treści na wiele platform];
        B4 --> B5[Generuj manifest.json];
    end

    subgraph "Dystrybucja"
        C1[manifest.json] --> C2{"Użytkownik mówi '/distribute'"};
        C2 --> C3[Publikuj na platformach];
    end

    A7 --> C1;
    B5 --> C1;

33 KiB Raw Permalink Blame History Unescape Escape