wechat-article

WeChat Official Account Article Creation (Stable Version)

Configuration File

Uniformly use workspace wechat-article.config.json：

{
  "appid": "公众号 AppID",
  "appsecret": "公众号 AppSecret",
  "author": "默认作者名",
  "account_name": "公众号名称",
  "slogan": "公众号 Slogan",
  "writing": {
    "perspective": "第一人称",
    "tone": "口语化",
    "length": "1500-2500字",
    "direction": "科技/AI/产品思考",
    "keywords_style": "短句为主，一行不超过30字",
    "article_style": "default",
    "style_profile": "auto",
    "reference_articles": []
  },
  "publish": {
    "need_open_comment": 1,
    "only_fans_can_comment": 0
  },
  "cover": {
    "html_template": null,
    "default_style": "minimal-grid",
    "palette": "auto",
    "rotate": "sequential",
    "seed": "title",
    "allowed_styles": ["minimal-grid", "card-editorial", "diagonal-motion", "soft-gradient"],
    "allowed_palettes": ["blue-tech", "purple-insight", "green-growth", "orange-energy", "rose-story", "slate-pro"]
  },
  "preview": {
    "send_cover_preview": 1,
    "require_confirm_before_publish": 1,
    "confirm_keyword": "确认发布"
  },
  "branding": {
    "header_html": "",
    "footer_html": ""
  },
  "article_reader": {
    "method": "auto",
    "methods_priority": ["camoufox", "jina_reader", "browser_snapshot"],
    "camoufox": {
      "enabled": true,
      "path": "~/.agent-reach/tools/wechat-article-for-ai/main.py"
    },
    "jina_reader": {
      "enabled": true
    },
    "browser_snapshot": {
      "enabled": true
    }
  }
}

First Use

If the configuration does not exist, ask all questions first and then write:

appid / appsecret
account_name / slogan (required)
Default author
Writing style (perspective, tone, length, direction)
Layout style (default default, optional gougestyle）
Comment switch (default: open comments=1, followers only=0)
Cover strategy (HTML template or PIL solution)
Writing Style DNA Initialization：
- Ask the user whether they have WeChat Official Account articles they like (3-5 articles is best)
- If yes, have the user provide links, and the system will automatically scrape and analyze them
- If no, use the default style and gradually learn during subsequent creation interactions
Article reading capability detection:
- Detect whether Camoufox is available (~/.agent-reach/tools/wechat-article-for-ai/main.py）
- If unavailable, prompt installation or fall back to another solution

If the built-in preview image does not exist or needs updating, then execute:

python3 scripts/create_cover_preview_grid.py

After sending the preview image to the user, you must useChinese + numberingAsk (do not expose English terminology to the user):

A. Default style (choose 1 of 4)
- A1 Minimal Grid (minimal-grid)
- A2 Editorial Card (card-editorial)
- A3 Diagonal Motion (diagonal-motion)
- A4 Soft Gradient (soft-gradient)
B. Color strategy (choose 1 of 2)
- B1 Automatic palette rotation (palette=auto, recommended)
- B2 Fixed single palette (choose 1 more set from section C)
C. Color scheme (choose 1 of 6, required only for B2)
- C1 Tech Blue (blue-tech)
- C2 Insight Purple (purple-insight)
- C3 Growth Green (green-growth)
- C4 Energy Orange (orange-energy)
- C5 Story Rose (rose-story)
- C6 Professional Slate (slate-pro)
D. Rotation method (choose 1 of 2, required only for B1)
- D1 Sequential rotation (sequential, recommended)
- D2 Random rotation (random)

User reply format:A2 B1 D1 or A1 B2 C3。

Then persist the selection result to wechat-article.config.json in the cover field.

Subsequent Use

When the configuration exists: the user only needs to provide a topic, and execute the process "creation → layout → cover → preview confirmation → draft publishing".

Writing Style DNA System

Core Concept

Each user's writing style comes from their own reference articles, not from a public template. This ensures that even if 100 people use the same skill, everyone's output is different.

DNA Files

Template：references/style-dna-template.md(blank template, copied on first use)
User DNA：references/style-dna.md(system-generated + continuously updated)
Reference articles：references/my-articles/(reference articles provided by the user)

DNA Initialization Process

用户提供 3-5 篇喜欢的公众号文章链接
    ↓
系统通过 article_reader 抓取文章内容
    ↓
保存到 references/my-articles/
    ↓
分析所有参考文章，提取风格特征：
- 句子平均长度
- 常用句式（反问？排比？短句连击？）
- 金句结构（比喻型？数字型？对比型？）
- 开头偏好（故事？数据？场景？）
- 结尾偏好（号召？金句？悬念？）
- 高频词汇和禁用词
- 段落结构和节奏
    ↓
生成 references/style-dna.md
    ↓
展示给用户确认/调整

Continuous DNA Evolution

In each creation interaction, the system automatically updates the DNA based on user feedback:

User behavior	DNA update
"This opening is too flat"	Record: avoid flat, straightforward openings
"This punchy line is good"	Record: preferred this sentence pattern
User manually revised a certain paragraph	Compare before and after the revision, extract preference changes
User chose title B instead of A	Record title preference
"Don't use this expression"	Add to avoid_phrases
"I like this feeling"	Extract features and add to preferred

Update rules (confirmation mode recommended):

After each creation session, check whether there are new preference signals
If there are,first show the user the suggested DNA update content, and write it in only after the user confirms
By default, do not overwrite automatically, to prevent "DNA drift" (mistaking an incidental preference for a fixed style)
Fields that the user has manually adjusted have the highest priority and will not be overwritten by automatic suggestions

Anti-templating detection

After generating an article, automatically check:

❌ Whether any expression from the avoid_phrases list in the DNA appears
❌ Whether the repetition rate with the punchy lines in recent_golden_phrases is > 30%
❌ Whether the same opening/ending structure as the last 3 articles is used
❌ Whether the paragraph structure is completely identical to the most recent article

If detected, automatically regenerate that part using a different expression.

WeChat Official Account article reading

Reading scheme (multi-layer fallback)

When the user provides a WeChat Official Account article link, read it in the following priority order:

1. Camoufox (first choice, most stable)

cd ~/.agent-reach/tools/wechat-article-for-ai && python3 main.py "https://mp.weixin.qq.com/s/xxx"

Can bypass WeChat's anti-scraping mechanisms
Outputs markdown format, including title, author, date, and body text
Requires pre-installation

2. Jina Reader (alternative)

curl -s "https://r.jina.ai/https://mp.weixin.qq.com/s/xxx"

Some articles may be blocked

3. Browser Snapshot (last resort)

Use OpenClaw's browser tool to open the link
Extract the page text content

4. Manual paste (final fallback)

Prompt the user to manually copy the article content

Reading capability detection

Automatically detect on first use:

# 检测 Camoufox
python3 ~/.agent-reach/tools/wechat-article-for-ai/main.py --help 2>/dev/null

If unavailable, prompt the user:

Install the agent-reach skill to get the best article reading capability
Or degrade to using other schemes

Workflow (must be followed in order)

Copy and check off:

WeChat Article Progress:
- [ ] Step 0: 读取/初始化配置 + 加载风格DNA
- [ ] Step 1: 生成文章内容（基于风格DNA）
- [ ] Step 2: 产出 HTML（内联样式 + 页头页脚）
- [ ] Step 3: 校验元数据（标题/摘要/作者）
- [ ] Step 4: 生成封面图
- [ ] Step 4.5: 生成正文配图（可选）
- [ ] Step 5: 发送预览（文本 + 封面图）并等待确认
- [ ] Step 6: 发布前预检（凭证/依赖/文件）
- [ ] Step 7: 推送草稿
- [ ] Step 8: 返回结果 + 更新风格DNA

Step 0: Read/initialize configuration + load style DNA

Read wechat-article.config.json
If it does not exist, enter first-time configuration and write it
If the configuration exists but is missing fields: only fill in the missing fields, do not overwrite the user's existing preferences
Read references/style-dna.md
- If it does not exist and the user provided reference articles, first execute DNA initialization
- If it does not exist and there are no reference articles, use the writing.* configuration as the base style

Step 1: Generate article content (based on style DNA)

1a. Choose the article structure

Do not default to "opening → 3-5 sections → ending". According to references/structures.md selection rules, choose the structural archetype by topic type:

Money-making/monetization → timeline narrative
Technology/tools → standard sections or listicle
Opinions/insights → Q&A dialogue format
Experience/retrospective → timeline narrative
Recommendations/roundups → listicle

Anti-repetition rule: Check the most recent 2 structures of recent_articles in the DNA; if they are the same, force a switch.

1b. Load style DNA and personal material

If references/style-dna.md：

Generate content strictly according to the style characteristics defined in the DNA
Imitate the sentence patterns, rhythm, and expression habits recorded in the DNA
Avoid the expressions listed in avoid_phrases and avoid_patterns in the DNA
Check recent_golden_phrasesto ensure punchy lines are not reused
According to the topic type, refer to 主题风格映射 in the DNA to adjust the language style

If references/story-bank.md：

Prioritize selecting material related to the topic from the user's real experience library
Replace AI-fabricated opinions with real opinions
Use the user's catchphrases and language habits
Refer to the user's emotional palette to adjust the article's emotional tone

If no DNA exists:

Produce the body text according to writing.* in the configuration

1c. Content safety awareness

During generation, proactively comply with the rules in references/content-safety.md :

Identify whether the current topic is a high/medium risk topic
High-risk topics are automatically switched to a safe writing approach ("information organization" mode, no subjective advice)
When specific data is involved, distinguish between "hard facts" and "soft opinions":
- Hard facts (specific numbers, rankings, statistics): sources must be cited; if there is no source, change to vague wording
- Soft opinions (personal judgments, experience sharing): can be expressed directly, but qualified with "personal experience"
Prohibited use of content-safety.md prohibited statements listed in

1d. Reduce template feel

Paragraph length should vary: short paragraphs (1-2 sentences) and long paragraphs (5-8 sentences) should alternate
Do not "deliver value" in every paragraph; leave appropriate blank space and transitions
Set a "small hook" (rhetorical question, suspense, twist) every 400-600 characters to maintain reading momentum
If there is a personal material library, insert real details in appropriate places to add a "human touch"

Constraints:

Title suggestion ≤ 20 Chinese characters (spread-friendly)
Body text suggestion ≤ 1000~2500 characters (according to user configuration)

Step 2: Produce HTML (inline styles + header and footer)

Strictly follow the typography specifications defined in references/styles/{article_style}.md .

Outer <section>
Body <p>
Emphasis <strong>
Between sections <hr>
Do not output markdown, do not rely on external CSS
If branding.header_html or branding.footer_htmlis configured, insert it into the corresponding position in the article content HTML.

Step 3: Validate metadata

Before publishing, there must be:

title(cannot be empty)
digest(suggested ≤ 120 characters)
author(prefer configuring author)

Backfill order:

Explicitly given by the user
Default configuration values
Auto-generated (title from main title, summary from first paragraph compressed)

Step 4: Generate cover image

Core principle: Each article's cover must be designed based on the topic; the same visual style must not be reused.

Priority:

User-provided cover_path
Use the HTML template specified by cover.html_template combined with Playwright screenshot (recommended)
- Available templates:akai-cover (A Kai cover template), gouge-cover (Gou Ge cover template)
Fall back to the PIL approach, using cover.default_style and cover.palette

HTML + Playwright screenshot approach (recommended)

According to cover.html_template configuration, render /templates/covers/{template_name}.html, and take a screenshot using Playwright.

Parameters:--html-template, --title, --subtitle, --account-name, --slogan, --tag

PIL fallback approach

If cover.html_templateis not specified, use scripts/create_cover.py for PIL image generation.

Parameters:--title, --subtitle, --style, --palette, --rotate, --seed

Command parameter priority:

Explicitly specified by the user this time
Configuration cover.*
Script default values

Step 4.5: Generate body illustrations (optional)

Based on the article content, illustrations can be generated to enhance readability.

Use the scripts in the scripts/illustrations/ directory to generate images (for example:generate_concept.py, generate_flow.py, generate_table.py）。
Generated images must be uploaded via publish_draft.py 's --content-img KEY=PATH parameter.
In article.html , use the  placeholder; the script will automatically replace it with the uploaded image URL.

Step 5: Send preview (text + cover image) and wait for confirmation

Before pushing the draft, you must first show the user a preview:

The preview content must include at least:

Title
Summary
Author
Cover image (the cover file generated this time)
Body preview (first 2-3 paragraphs or first 200-300 characters)

Sending rules:

If the current channel supports images, send "text + cover image"
The preview text must be in Chinese and include clear action prompts:
- 确认发布(Continue)
- 修改封面(Redo cover only)
- 修改正文(Return to body editing)

Confirmation policy (default):

preview.require_confirm_before_publish = 1 When preview.confirm_keyword(default 确认发布) is not received, publishing must not be executed
If the user replies 修改封面, keep the body and re-execute Step 4, then preview again
If the user replies 修改正文, return to Step 1/2, adjust, and preview again

Step 6: Pre-publish check

Before publishing, you must check:

python3 available
Pillow installed (if a cover or illustrations need to be generated)
Playwright and its browser driver installed (if using an HTML template cover)
appid/appsecret non-empty
article.html and the cover file exist
Content safety validation(must be executed):

python3 scripts/validate_article.py --content-file article.html

Validation result handling:

risk_score = 0，can_publish = true → go directly to Step 7
needs_review = true(has warnings) → show the validation report to the user, mark the data to be verified and medium-risk content, and continue only after the user confirms
can_publish = false(has errors) → Block publishing, show the specific issue and require the user to re-validate after making changes

If items are missing, fix them first; do not publish directly.

Step 7: Push draft

Use scripts/publish_draft.py。

python3 scripts/publish_draft.py \
  --title "文章标题" \
  --author "作者名" \
  --digest "摘要（120字内）" \
  --content-file article.html \
  --cover cover.jpg \
  --appid <appid> \
  --appsecret <appsecret> \
  --need-open-comment 1 \
  --only-fans-can-comment 0 \
  --header-img <页眉图片路径> \
  --content-img <KEY1=PATH1> --content-img <KEY2=PATH2>

Comment parameter priority:

The user explicitly requested this time
Configure publish.*
Default value (1 / 0)

Step 8: Return results + update style DNA

Fixed return:

Title, summary, author
Cover file + style/color scheme used
Comment toggle status
Draft media_id
List of uploaded body image URLs
Next step: Go to the WeChat Official Account backend "Content Management → Draft Box" to preview and publish

DNA update (if style-dna.md exists):

Add this article's title to recent_articles
Add the opening/closing type used this time to the history
Add the golden quote used this time to recent_golden_phrases
If the user provides feedback during the creation process (modifications, selections, evaluations), update the corresponding preference fields
Record this change in the update log

Security and boundaries

Only push to the draft box, do not mass-send directly
Credentials are stored only in local configuration, not written into skill files
Any outbound action (automatic publishing/mass sending) must separately seek user confirmation

Project structure

wechat-article-skill/
├── SKILL.md
├── README.md
├── assets/
│   ├── NotoSansCJKsc-Bold.otf
│   └── cover-style-palette-preview-grid.jpg
├── references/
│   ├── article-style.md              (排版风格索引)
│   ├── structures.md                 (文章结构原型库)
│   ├── content-safety.md             (内容安全与合规指南)
│   ├── style-dna-template.md         (风格DNA空白模板)
│   ├── story-bank-template.md        (个人素材库模板)
│   ├── style-dna.md                  (用户风格DNA，系统生成)
│   ├── story-bank.md                 (用户个人素材库)
│   ├── my-articles/                  (用户参考文章)
│   │   └── README.md
│   └── styles/
│       ├── default.md                (默认排版风格)
│       └── gougestyle.md             (狗哥排版风格)
├── templates/
│   └── covers/
│       ├── akai-cover.html           (阿凯封面模板)
│       └── gouge-cover.html          (狗哥封面模板)
└── scripts/
    ├── font_utils.py                 (跨平台字体工具)
    ├── create_cover.py               (封面生成脚本)
    ├── create_cover_preview_grid.py
    ├── publish_draft.py              (草稿发布脚本)
    ├── validate_article.py           (发布前内容校验)
    └── illustrations/
        ├── generate_concept.py       (概念示意图生成)
        ├── generate_flow.py          (流程图生成)
        ├── generate_header.py        (页眉图片生成)
        └── generate_table.py         (数据表格图生成)