wechat-creator

Public Account Content Workshop

1. Introduction

Public Account Content Workshop — one entry point covers the entire chain of public account creation and operations:10w+ viral article rankings / viral body content creation / title generation + scoring / copy rewriting / prohibited word detection / cover design / account diagnosis。

Based on the 1000+ 10w+ viral articlescontinuously collected daily by Honghu Data and public account viral data across the entire web, it polishes inspiration into a finished product ready for direct publishing:

#	Capability	One-sentence description
1	10w+ viral article rankings	Query the daily 10w+ reads TOP50 across all fields or 23 categories, with four-dimensional content analysis + HTML report
2	Viral body content creation	Keyword search for viral articles → extract traffic passwords → integrate style samples → generate a complete article ready for direct publishing
3	Title generation + scoring	Generate 10 viral titles, or perform a four-dimensional 100-point quantitative score on existing titles
4	Copy rewriting	One-click conversion of any material into a public account viral style, with automatic reporting of rewriting records (free, no points deducted)
5	Prohibited word detection	Four-end input for copy / files / images / web pages, with risk-level labeling + replacement suggestions + safe version
6	Cover design	AI analyzes the visual patterns of viral covers in the same track and outputs 3 sets of 2.35:1 horizontal design plans + HTML report
7	Account diagnosis	Four-dimensional 100-point scoring + S/A/B/C/D/E six-level rating + industry benchmarking + multi-account comparison

Target users: public account operators, content creators, self-media editors, brands, MCN institutions, viral article researchers.

---

2. Function routing (the Agent first identifies the intent before entering the corresponding module)

⚠️ Before entering specific execution, the Agent must first identify the user's intent according to the table below and route to the corresponding module,cross-module mixing is prohibited。

User intent (keywords/examples)	Enter module	Required reference
"10w+ viral article", "public account viral article rankings", "today's viral articles", "tech and digital track viral articles"	M1 10w+ viral article rankings	references/m1_category_mapping.md
"write a public account article", "generate public account copy", "an article about XX"	M2 Viral body content creation	references/m2_gzh_trend_data_format.md
"help me create a public account title", "generate viral titles", "a title about XX"	M3-A Title generation	This file §6.2
"score this title", "evaluate titles", "which title is better"	M3-B Title scoring	This file §6.3
"rewrite into public account style", "help me public-accountize this copy"	M4 Copy rewriting	assets/m4_platform_rules.md
"Are there any prohibited words", "detect sensitive words", "official account compliance", "throttling words"	M5 Prohibited Word Detection	references/m5_prohibited_word_core_workflow.md
"Design an official account cover", "make a tweet cover", "beauty/finance cover"	M6 Cover Design	references/m6_report_template.md
"Diagnose XX official account", "analyze XX account", "compare and diagnose A and B"	M7 Account Diagnosis	references/m7_core_workflow.md

Compound intent(e.g., "after writing the article, also detect prohibited words"): execute serially in module order, with each module outputting independently.

Feature inquiry only(e.g., "what can you do", "how to use"): only introduce the capability list, do not enter any module execution.

---

3. Global rules (shared by all modules)

3.1 Authentication (REDFOX_API_KEY)

All modules share the same API Key.

How to obtain：

1. Visit RedFox Hub to learn about the service
2. Go to the registration page to register (new users receive free credits)
3. Get the API Key in the personal center, format ak_xxxxxxxx

Configuration method(in descending order of priority):

Priority	Source	Description
1	Environment variable REDFOX_API_KEY	Directly read the current process environment variable
2	Shell configuration file	Automatically scan ~/.zshrc ~/.bashrc ~/.bash_profile ~/.profile ~/.zprofile
3	Prompt the user to configure	When none of the above are found, guide the user to configure

Configuration example：

• macOS/Linux（zsh）：echo 'export REDFOX_API_KEY=ak_xxx' >> ~/.zshrc && source ~/.zshrc
• macOS/Linux (bash): write to ~/.bashrc then source
• Windows：[Environment]::SetEnvironmentVariable("REDFOX_API_KEY", "<值>", "User"), restart the terminal

After configuration, verify:echo $REDFOX_API_KEY（macOS/Linux）/ echo %REDFOX_API_KEY%（Windows）。

3.2 Wording standards (mandatory)

❌ Prohibited	✅ Correct expression
Crawl / scrape	Retrieve / query
Explosive cover / explosive data	Viral cover / viral data
Crawl articles / scrape articles	Query viral articles / retrieve articles

All external output (dialogue, reports, prompts) must comply.

3.3 Keyword type determination (shared by M2/M3/M6)

• Niche term/vertical track(with scenario/attribute modifiers, such as "workplace job-hopping", "fat-loss meals") → query directly
• Specific product term / specific entity(such as "sunscreen", a brand name, a certain tool) → query directly,do not expand to other forms
• General term/broad category(pure broad-category term without modifiers, such as "emotion", "finance", "technology") → execute the general term expansion strategy

General term expansion: first query recent web information, generate 10 niche directions (2-3 each of trend terms, audience terms, scenario terms, intent terms), and after outputting,stopand wait for the user to reply "expand" or "do not expand".

3.4 Data time rules

• Official account data source coverageyesterday to 30 days ago, today's data is not yet in the database
• M1 special update time: daily 18:30 syncs the previous day's data; when querying "today/yesterday", automatically convert according to the table below:

Current time	User inquiry	Actual query interval	Reason
< 18:30	Today / Yesterday	T-2 ~ T-1	Yesterday's data not updated, query the day before yesterday's interval
≥ 18:30	Today / Yesterday	T-1 ~ T	Yesterday's data already updated

• User says "today/this day" but actually the data queried is yesterday/the day before yesterday: must clearly prompt, e.g., "Today's data will only update tomorrow/after 18:30 today; the most recent available data has been queried for you"
• Exceeds 30 days: prompt "Only data from the last 30 days is supported"
• Insufficient data automatically expands(Do not change the wording): last 1 day → last 3 days → last 7 days → last 30 days

3.5 Iron rules for data output (highest priority)

All viral data must first be fully output, then analyzed. If data is not output, analysis does not begin.

Data source	Output timing	Format requirements
M1 10w+ viral articles	Before outputting viral pattern analysis	Complete article table (serial number/title with link/author/read count/publish time)
M2 Viral article content query	Before analyzing viral patterns	Complete viral article table (rank/title/official account/reads/Wow/Likes/Comments)
M3 Title generation	Before generating new titles	Complete viral title table (rank/title/official account/reads/Wow/Likes)
M6 Cover design	Before outputting cover style analysis	Viral cover data list (including imageUrl/title/read count)
M7 Account diagnosis	Before outputting diagnosis conclusions	Complete recent work data table (title with link/reads/likes/comments/Wow/publish time)

Execution rules：

1. Data takes priority over analysis: first output the complete raw data → then begin data analysis/conclusions
2. Must not omit: even if there are more than 20 entries, all must be output; do not display only TOP N (except when the user actively requests a preview)
3. Must not skip: under no circumstances may the data output step be skipped to directly provide analysis conclusions
4. Violation means correction: if any module process contains instructions to "not output" or "execute internally," this iron rule shall prevail, and output must be provided

---

4. M1 10w+ viral article ranking

4.1 Trigger

"Check out today's 10w+ viral articles," "viral articles in the technology and digital sector over the last 7 days," "viral articles across all fields."

4.2 Core capabilities

• Daily collection of 1000+ entries 10w+ read articles, covering 23 standard categories(Humanities information, knowledge encyclopedia, health and wellness, fashion trends, food and dining, lifestyle, travel, humor, emotion and psychology, sports and entertainment, beauty and body care, digest selections, livelihood information, wealth and finance, technology and digital, venture capital and business, automotive and transportation, real estate and housing market, career development, education and exams, academic research, corporate brands, overall ranking)
• Four-dimensional content analysis (content overview / hot topic utilization / dissemination role / achieved results)
• Viral pattern analysis (title characteristics / content themes / publish time / official account characteristics)
• WeChat green-themed HTML visualization report (supports PDF export)

4.3 Execution flow

Step 1: Intent recognition and category matching

• Overall ranking request / user did not specify a category → --type "总排名"
• Clear domain keywords (such as "AI" "weight loss") → by references/m1_category_mapping.md mapped to one of the 23 standard categories
• Ambiguous input → prompt the user to choose or go to "overall ranking"

Step 2: Time conversion

Follow the M1 special update timetable in global rule 3.4, automatically calculate --start_date and --end_date. By default, query the past 7 days.

Step 3: Obtain viral article data

python scripts/fetch_hot_articles.py \
  --type "<分类>" \
  --start_date <yyyy-MM-dd> \
  --end_date <yyyy-MM-dd> \
  --mode preview --limit 10 \
  --source "公众号10w+阅读文章推荐-GitHub"

• --mode preview --limit 10: first display a preview of the first 10 items
• User replies "show all" or "yes" → rerun --mode full to display all TOP50
• When the data does not exceed 10 items, display all directly without asking
• For complete parameters and API specifications, see references/m1_api_spec.md

It is forbidden to throw raw JSON to the user; it is forbidden to send only a file path or brief prompts such as "saved".

Step 4: Four-dimensional content analysis (each article)

• Content overview: concisely summarize the core content
• Hot topic utilization: which hot topics or trends were used
• Dissemination role: satisfy user needs / trigger emotional resonance / provide practical value
• Achieved effect: increase brand awareness / complete user conversion / promote information dissemination

Step 5: Analysis of viral patterns

Based on actual data, analyze from four aspects: title characteristics, content themes, publishing time, and official account characteristics.

Step 6: Ask about subscription

After the pattern analysis, proactively ask:

📬 订阅服务
是否需要订阅具体赛道的每日爆文推送？支持 23 个标准分类。
- 订阅推送 - 每天 19:30 推送最新 10w+ 公众号爆文
- 暂不需要 - 仅本次查询

Step 7: Generate HTML report (optional)

python scripts/generate_hot_html.py \
  --temp_file temp_articles.json \
  --output "10w+爆文_榜单.html" \
  --display_count <对话中实际展示数量>

HTML visual specifications: primary color rgb(0, 179, 84)(WeChat green), use 🥇🥈🥉 for rankings 1-3, author names link to the official account card, PDF export pagebreak: { mode: 'none' }。

4.4 Output format

| 序号 | 标题 | 作者 | 阅读数 |
|------|------|------|--------|
| 1 | [标题](文章链接) | [作者名称](公众号名片链接) | 5w+ |

**序号、[标题](文章链接)**
📄 作者：[作者名称](公众号名片链接)
👀 阅读数：5w+
⏰ 发布时间：yyyy-MM-dd HH:mm
🔍 内容分析：[四维度分析]
---

4.5 Exception handling

• Fewer than 10 articles with 100k+ reads on a certain day in a certain track: "💡 There are relatively few 100k+ articles in xx track on xx day. You can expand to the past 30 days or check the comprehensive 100k+ articles~"
• Query returns empty: recommend viewing the comprehensive ranking or other categories
• User asks about time and it is inconsistent with the actual query: inform according to the prompt in the §3.4 table

---

5. M2 Viral article body creation

5.1 Trigger

"Help me write a WeChat official account article about XX", "Generate WeChat official account copy on the theme of XX".

⚠️ This module is executed only in the main Agent,not dispatched to sub-Agents(long context chain; sub-Agents will lose viral patterns + style samples).

5.2 Execution flow

Step 0: Intent recognition and keyword extraction

• Understand the core demand, target audience, and creation purpose
• Determine the type of writing object:
  • General topics(workplace/emotions/health/finance, etc.) → core keywords + scenario keywords + extended keywords
  • Specific entities(products/tools/brands/accounts, etc.) → intent keywords + subject keywords (category words / positioning words / user terms), with at least 2 subject keywords

Step 0.5: Differentiation advantage breakdown (specific entities only)

Collect information (user description → structured document → web search), break it down along four dimensions: input differences / process differences / output differences / positioning differences. Output 3-5 differentiated advantages, each containing "advantage description + comparison difference with competitors," and mandatorily reference them in Step 2/4.

Step 1: Query viral articles

python3 scripts/fetch_gzh_trends.py \
  --keyword "<关键词1>,<关键词2>" \
  --max-items 10 \
  [--start-date <yyyy-MM-dd>]

• Separate multiple keywords with commas, maximum 5, total length ≤ 200 characters
• Default to the last 7 days; if data is insufficient, automatically expand to the last 30 days
• The complete viral data must first be displayed in a table, then proceed to Step 2 analysis (follow the iron rule in §3.5)

Data output format:

## 📊 获取到的爆款数据

| 排名 | 标题 | 公众号 | 阅读数 | 在看 | 点赞 | 评论 | 链接 |
|------|------|--------|--------|------|------|------|------|
| 1 | [标题] | [公众号名] | [阅读数] | [在看数] | [点赞数] | [评论数] | [链接] |

> 数据来源：红狐数据 API，统计截止 [日期]

Step 2: Analyze viral patterns

Based on the data, extract:

• Title patterns: characteristics of high-engagement titles (numbers, emotional words, audience labels), length and structure
• Content patterns: topic direction, structure and paragraph distribution, interaction guidance
• Data analysis dimensions: reads (reach) / Wow (depth) / likes (resonance) / comments (topicality)
• Differentiated advantage comparison(if Step 0.5 was executed): find points of alignment

Step 3: Guide the user to upload personal style references (mandatory question)

"Want me to understand your style better? Just send me a few of your usual articles, notes, or essays"

• Provide samples → analyze style and integrate it
• Do not provide → generate directly based on viral patterns

Step 4: Generate copy

Preparation:

1. Determine the core viewpoint (mandatory): the entire article should revolve around this; it must be novel, controversial, or counterintuitive; if Step 0.5 was executed, the core viewpoint should revolve around the differentiated advantages
2. Review viral patterns: high-frequency keywords must be incorporated, popular title patterns must be followed, viral structural characteristics must be referenced
3. Optional: use web_search to search for the latest information as supplementary material (supplementary only; it cannot replace viral patterns)

Writing:

1. Title: 3-5 candidate titles, following viral patterns, ≤ 30 characters
2. Body: viral opening "hook" + solid main body (bullet points/steps/comparison structure) + powerful ending + interaction guidance, suitable for mobile reading
3. Summary: 1-2 sentences summarizing the core viewpoint, suitable for sharing
4. Tags: 5-10 related topic tags

Step 5: Self-check before output

• Recommended titles: 3-5, including highlight explanations, avoid clickbait
• Complete publish-ready body (engaging opening + solid main body + powerful ending)
• Article summary
• Recommended tags: 5-10
• Viral formula sources - referenced viral patterns
• Viral formula sources - referenced viral articles: 2-3, including title / official account / link / complete interaction data

5.3 Output format (strict)

### 推荐标题
1. [标题1] —— [亮点说明]
2. [标题2] —— [亮点说明]
3. [标题3] —— [亮点说明]

### 文章摘要
[1-2 句核心观点]

### 正文内容
[完整可发布的正文]

### 推荐标签
#标签1 #标签2 #标签3 #标签4 #标签5

### 爆款公式来源
**参考的爆款规律**：[简述爆款公式]

**参考的爆款文章**（2-3 篇核心参考，优先互动数据最高）：
1. [文章标题](文章链接) - [公众号名](公众号名片链接) - 阅读 X / 在看 X / 点赞 X / 评论 X
2. ...

Prohibited output: script commands, file paths/download links, displaying data only without generating copy.

---

6. M3 Title generation and scoring

6.1 Trigger

• M3-A Title generation: "Help me come up with an official account title," "Generate a viral title," "A title about XX"
• M3-B Title scoring: "rating", "scoring", "how is this title", "evaluate title"

6.2 M3-A Title Generation Process

Step 1: Extract keywords(follow global rule 3.3)

Step 2: Identify generic words and prompt expansion

If it is a generic word, first output:

我识别到「XX」是较大的分类，已查询近期热门趋势，推荐以下细分方向：
[10 个细分词]
回复「拓展」将同时搜索这 10 个词，回复「不拓展」将继续搜索「XX」

Step 3: Time range and data query

• Default last 7 days (--days 7), maximum 30
• Command:

python3 scripts/fetch_official_account_trends.py --keyword "<关键词>" --days <天数>
# 无关键词（全站热门）
python3 scripts/fetch_official_account_trends.py --keyword "" --days <天数>

Step 4: Read and output viral data

Script generates 关键词_爆款数据.md, after readingfull table display(follow §3.5 iron rule):

## 📊 获取到的爆款数据

| 排名 | 标题 | 公众号 | 阅读数 | 在看 | 点赞 | 链接 |
|------|------|--------|--------|------|------|------|
| 1 | [标题] | [公众号名] | [阅读数] | [在看数] | [点赞数] | [链接] |

> 数据来源：红狐数据 API，统计截止 [日期]

When data is fewer than 5 items, automatically extend the time per §3.4, but all data already obtained must be displayed in full.

Step 5: Analyze first, then generate (order cannot be reversed)

1. Analyze viral titles: extract structure/keywords/emotion/audience characteristics, and show the conclusions to the user
2. Generate 10 new titles: innovate based on patterns (not reuse), each including: title content (≤ 30 characters), match index (8.0-10.0, one decimal place, same value ≤ 2 times), reference viral hits (including link and read count, separate multiple with a Chinese enumeration comma), recommendation reason (natural and colloquial, AI-style phrasing such as "adopts XX structure" or "target audience is" is prohibited)

6.3 M3-A Output Format

基于本次分析，为您生成以下 10 个爆款标题：

**标题1：[标题内容]**
📈匹配指数：[8.0-10.0]
🔥参考的爆款：[爆款A](链接)（阅读：XXX）、[爆款B](链接)（阅读：YYY）
👍推荐理由：[自然口语化表达]

---
（共 10 个，格式同上）

Usage suggestions: first test 2-3 with higher match indexes → adjust based on data feedback → personalize modifications based on the content.

6.4 M3-B Title Scoring Process

Step 1: Understand the title, extract keywords/topic/audience

Step 2: Query data(default last 30 days)

python3 scripts/fetch_official_account_trends.py --keyword "<关键词>" --days 30

Step 3: Four-dimensional 100-point scoring(total score 100, all scoring must be based on real API data)

Dimension	Full score	Scoring points
Track match	15	Hits ≥3 core words→15; 1-2→8; 0→0. If unmatched, mark "positioning deviation, adjustment recommended"
Click incentive strength	35	Clear benefit/pain point (18) + emotion/curiosity hook (12) + scarcity/exclusivity (5)
Structural compliance	15	Word count fits 15-25 characters (8) + structure matches 100k+ preferences (7); violations deduct 3-8 points
Viral potential match	35	Last 7 days hotspot match (20) + 100k+ semantic similarity (15)

Grade：S(90+) / A(75-89) / B(60-74) / C(<60)

Step 4: Output scoring results(no dimension may be omitted)

## 标题评分结果

**原标题**：[用户输入的标题]

**综合评分**：[X] 分/100 分
**等级**：[S/A/B/C]

**各维度评分明细**：

**a. 赛道匹配度（15 分）**
- 所属赛道：[自动识别]
- 命中核心词：[词1]、[词2]、[词3]
- 得分：[X] 分

**b. 点击诱因强度（35 分）**
- 明确利益/痛点：[X] 分
- 情绪/好奇心钩子：[X] 分
- 稀缺/专属感：[X] 分
- 得分：[X] 分

**c. 结构合规性（15 分）**
- 字数适配：[X] 分
- 结构符合 10w+ 偏好：[X] 分
- 无违规/负向感知：[扣/不扣]
- 得分：[X] 分

**d. 爆文潜质匹配度（35 分）**
- 近期热点匹配：[X] 分
- 10w+ 语义相似度：[X] 分
- 得分：[X] 分

**优势分析**：[自然语言描述]
**优化建议**：[自然语言描述]

S grade: explain advantages based on the four dimensions + 3-5 highly liked titles on the same topic and their similarities A/B/C grade: core problem list + 3-5 reference viral hits + 2-3 revised versions (practical style / emotional style / curiosity style)

---

7. M4 Copy Rewriting

7.1 Trigger

"Rewrite into WeChat official account style", "help me make this copy WeChat official account style", "turn this passage into a WeChat official account article".

7.2 Execution Process

Step 1: Read rewriting rules

Read assets/m4_platform_rules.md, and strictly follow the 10 writing principles, 10 high-frequency keywords, and Workflow in it.

Step 2: Report rewriting record (non-blocking)

python scripts/rewrite.py "<原始文案>"

• API:https://redfox.hk/story/api/skill/record/save
• source Fixed as "公众号文案改写-GitHub"
• Requires REDFOX_API_KEY(free, no points deducted), passed in via X-API-Key header
• Reporting failure does not affect the main flow, only prints a warning
• View rule prompt:python scripts/rewrite.py prompt

Step 3: Rewrite and output directly

Output must include:

• An attractive WeChat Official Account-style title (≤ 30 characters)
• Use bullet points / step-by-step tutorial / comparison display in the body, and guide interaction at the end
• Choose 2-3 high-frequency keywords to naturally integrate into the body
• Colloquial, friendly and natural, suitable for WeChat Official Account reading scenarios

After output is complete, do not ask the user whether adjustments are needed. Rewrite record reporting is an automatic backend behavior and does not need to be shown in the body result.

---

8. M5 Prohibited Word Detection

8.1 Trigger

"Are there any prohibited words", "detect sensitive words", "WeChat Official Account compliance", "throttling words", "post compliance".

8.2 Input Methods

Input Form	Execution Action
Plain text	Count characters → character count gate → --content="..." → three-section output
Upload TXT/DOC/DOCX	--file=path --extract-only Check character count → detect → three-section output
Upload image	Extract text from image → count characters → --content="提取文字" → three-section output
Paste https://...	--url=... --extract-only Check character count → detect → three-section output
Extra-long copy (>3000 characters)	Ask first, continue only after receiving a clear reply

--content、--file、--url The three are mutually exclusive.

⚠️ PDF is not supported; it must be converted to an image or text before uploading.

8.3 Execution Command

python scripts/check_sensitive_words.py --content "<文案>"
python scripts/check_sensitive_words.py --file "<路径>" [--extract-only]
python scripts/check_sensitive_words.py --url "<网址>" [--extract-only]

Script API Key priority: environment variable → shell configuration file → prompt. For the complete execution procedure (output template, format iron rules, examples), see references/m5_prohibited_word_core_workflow.md。

8.4 Three-Section Output

1. 🔍 Prohibited word detection result: hit words in the original text are bolded + risk level
2. 💡 Revision suggestions: each prohibited word is paired with a contextual replacement word + reason
3. ✏️ Optimized copy: the publishable version after replacement + plain text file card

8.5 Advanced

• --extract-only: only extract text preview, no detection
• Long copy in batches: 3001-10000 characters ask (1=only first 3000 / 2=batch all / 3=cancel), >10000 characters not currently supported

---

9. M6 Cover Design

9.1 Trigger

"Design a WeChat Official Account cover", "make a post cover", "how to design a beauty/finance cover".

Does not trigger: only asks "what can you do", "how to use it".

9.2 Execution Flow (7-step standard flow, cannot be skipped)

For complete specifications, see references/m6_report_template.md。

Step	Name	Description
1	User intent parsing	Structured parsing of theme/type/style/keywords
2	Viral data query	Call script to obtain viral cover data
3	Intelligent matching and filtering	Comprehensive scoring to filter the 20 most relevant
4	AI image analysis	Identify cover visual elements one by one (text/scene/color tone/composition), filter out blank and small-size images (width/height <10px)
5	Style summary	Output a hit cover style summary table (including the real imageUrl)
6	Solution generation	Output 3 differentiated design solutions (including image generation prompts) + HTML report
7	Output self-check	Check completeness item by item

9.3 Data query commands

python scripts/fetch_explosive_covers.py --keyword "<关键词1>,<关键词2>"
# 全站热门
python scripts/fetch_explosive_covers.py --keyword ""

• Multiple keywords separated by commas, maximum 5, total length ≤ 200 characters
• By default, do not pass time parameters (the API decides on its own); when the user specifies "last N days" startDate = 今天 - (N-1) 天
• For the data format, see references/m6_gzh_trend_data_format.md

9.4 HTML report generation

Use references/m6_report_template.html(JS-rendered template),only inject JSON data; generating a complete HTML on the fly is prohibited：

1. Copy the template:cp references/m6_report_template.html ./爆款封面分析报告_{关键词}.html
2. Inject JSON data into the template placeholders
3. Open in the browser:open ./爆款封面分析报告_{关键词}.html

JSON structure：

{
  "keyword": "XX",
  "summary": "风格总结",
  "cases": [{ "imageUrl": "", "url": "", "title": "", "author": "", "reads": "" }],
  "plans": [{
    "name": "",
    "coreVisual": "",
    "case": { "imageUrl": "", "url": "", "title": "", "author": "", "reads": "" },
    "prompt": ""
  }]
}

9.5 Output requirements

• All cover solutions are adapted to WeChat Official Account 2.35:1 horizontal format (900×383 px), use object-fit: cover + object-position: center
• For cover image links, use the original value of the imageUrl field (make no modifications）
• Add anti-hotlinking to HTML:<meta name="referrer" content="no-referrer">
• Each solution includes: visual style description, case references (imageUrl + title link + author + read count), actionable image generation prompts (following a 2.35:1 ratio)
• When the article link cannot be opened, prompt the user to copy the title and search within WeChat

---

10. M7 Account Diagnosis

10.1 Trigger

"Diagnose XX Official Account", "Analyze XX account", "Compare and diagnose A and B", "Official account health check".

10.2 Core capabilities

• Four-dimensional scoring: content health / user activity / core data / operational standardization (each dimension on a 100-point scale)
• Intelligent rating: S/A/B/C/D/E six levels (S ≥ 85 points) + industry benchmarking
• Multi-account comparison: supports diagnosing multiple accounts at the same time + horizontal comparison summary
• 5-chapter fixed template: chapters/order/format are strictly locked and must not deviate

10.3 Execution flow

Step 1: Parse input

• Single account: Official Account name or ID
• Multiple accounts: extract multiple accounts from natural language ("Compare and diagnose A and B and C")

Step 2: Execute the diagnosis script

# 单账号/多账号通用查询
python scripts/wechat_analyzer.py query --account_names "账号A,账号B"
# 或按 ID 查询
python scripts/wechat_analyzer.py query --account_ids "id1,id2"
# 强制执行（即使无作品数据）
python scripts/wechat_analyzer.py query --account_names "账号A" --force_analyze

Step 3: Single-account HTML report(optional)

python scripts/wechat_analyzer.py generate_html

Step 4: Multi-account comparison HTML report(optional)

python scripts/wechat_analyzer.py generate_multi_html

10.4 Scoring algorithm

• Main formula(for accounts from the Honghu data source): comprehensive score = health 30% + activity 30% + core data 30% + operational standardization 10%
• Alternative formula(for accounts not from the Honghu data source): comprehensive score = core data 40% + average of two-dimensional diagnosis 35% + operational standardization 25%
• Levels：S(≥85) / A(75-84) / B(65-74) / C(55-64) / D(45-54) / E(<45)

For complete scoring rules, field mapping, and output templates, see references/m7_core_workflow.md and references/m7_workflow_guide.md。

10.5 Output format (5-chapter fixed template, must not deviate)

一、账号信息
  账号名 / ID / 赛道 / 粉丝数 / 作品数 / 数据来源

二、综合评分
  综合评分 / 等级 / 四维度雷达图 / 行业对标

三、近期作品数据
  完整近期作品表格（标题带可点击链接 / 阅读 / 点赞 / 评论 / 在看 / 发布时间）
  —— 遵循 §3.5 铁律：必须先完整输出，不得省略

四、优化建议
  基于短板的可落地建议

五、行业对标分析
  与同赛道账号的横向对比

Multi-account comparison: output a 5-chapter report for each account separately + a horizontal comparison summary table + differentiation suggestions.

10.6 Mandatory Constraints

• Data must come from the Redfox API,estimating or using external data sources is strictly prohibited
• When no account is foundgenerating an estimated report is strictly prohibited, the user must be prompted to verify the account name
• Recent work titles must include complete clickable links and must not be omitted

---

11. One-Click Installation

11.1 Prerequisites

• Python 3.8+
• Redfox Hub API Key (ak_xxxxxxxx）

11.2 Dependency Installation

pip install requests>=2.28.0 python-docx==1.1.0 beautifulsoup4==4.12.3 playwright==1.58.0
playwright install chromium

M1 scripts use the Python standard library (urllib.request) and require no additional installation; M5 prohibited word detection requires python-docx / beautifulsoup4 / playwright; M6/M7 require only requests.

11.3 Configure API Key

See global rule 3.1.

---

12. Project Architecture

12.1 Directory Structure

wechat-creator/
├── SKILL.md                                        # 本文件（入口）
├── scripts/
│   ├── fetch_hot_articles.py                       # M1 10w+ 爆文数据获取
│   ├── generate_hot_html.py                      # M1 HTML 报告生成
│   ├── fetch_gzh_trends.py                       # M2 爆款正文数据查询
│   ├── fetch_official_account_trends.py          # M3 标题生成/评分数据查询
│   ├── rewrite.py                                # M4 改写记录上报 + 规则提取
│   ├── check_sensitive_words.py                  # M5 违禁词检测
│   ├── fetch_explosive_covers.py                 # M6 封面数据查询
│   └── wechat_analyzer.py                        # M7 账号诊断引擎
├── references/
│   ├── m1_api_spec.md                            # M1 API 接口规范
│   ├── m1_category_mapping.md                    # M1 分类泛化映射
│   ├── m2_gzh_trend_data_format.md               # M2 爆款数据格式
│   ├── m5_prohibited_word_core_workflow.md       # M5 违禁词输出模板
│   ├── m6_gzh_trend_data_format.md               # M6 封面数据格式
│   ├── m6_report_template.html                   # M6 HTML 报告模板
│   ├── m6_report_template.md                     # M6 封面 7 步执行规范
│   ├── m7_core_workflow.md                       # M7 核心工作流
│   └── m7_workflow_guide.md                      # M7 评分细则
└── assets/
    ├── m4_platform_rules.md                      # M4 公众号改写规则
    └── m7_report_template.html                   # M7 单账号 HTML 报告模板

12.2 Technology Stack

Component	Technology	Description
Runtime Environment	Python 3.8+	Standard Python environment
HTTP Requests	requests / urllib	Redfox API calls, record reporting (SSL certificate verification enabled)
Document Parsing	python-docx	Word text extraction (M5)
Web Page Parsing	beautifulsoup4 + playwright	Dual engine for static HTML + JS dynamic rendering (M5)
Image Analysis	Agent native image understanding	M6 cover visual element recognition
Report Rendering	HTML/CSS/JS + html2pdf.js	M1/M6/M7 HTML report + PDF export
Data Source	Redfox Data API (redfox.hk)	Daily updates of 1000+ 100k+ viral articles + viral articles from public accounts across the web

12.3 Script Quick Reference

Module	Script	Core Parameters
M1	scripts/fetch_hot_articles.py	--type --start_date --end_date --mode --limit --source
M1	scripts/generate_hot_html.py	--temp_file --output --display_count
M2	scripts/fetch_gzh_trends.py	--keyword --max-items --start-date
M3	scripts/fetch_official_account_trends.py	--keyword --days
M4	scripts/rewrite.py	<文案> or prompt
M5	scripts/check_sensitive_words.py	--content / --file / --url / --extract-only
M6	scripts/fetch_explosive_covers.py	--keyword --start-date
M7	scripts/wechat_analyzer.py	query --account_ids/--account_names / generate_html / generate_multi_html

---

13. FAQ

Q1: Do all seven capabilities need to be installed? A: Install once and you have all 7 capabilities, routed on demand. If you only want to use some of them, the dependencies of other modules (such as playwright) can be left uninstalled, and a missing dependency prompt will appear when the corresponding module is called.

Q2: Is the API Key charged? A: M4 rewriting record reporting is completely free and does not deduct points; the data queries and prohibited word detection of M1/M2/M3/M5/M6/M7 consume points according to Redfox Hub standards, and newly registered users receive free points that can be used directly.

Q3: Why can't M2 be dispatched to a sub-Agent? A: Body content creation relies on complete context (viral patterns + style samples + differentiation advantages + latest information) for long-chain creation, and a sub-Agent will lose context, affecting output completeness.

Q4: Why does M1 return yesterday's/the day before yesterday's data when querying "today"? A: 10w+ data syncs the previous day's data every day at 18:30. Before 18:30, querying "today" is actually T-2 ~ T-1; after 18:30 it is T-1 ~ T. The system automatically converts and prompts.

Q5: What is the data time range? A: The database covers yesterday to 30 days ago. M1 defaults to the last 7 days, M2 defaults to the last 7 days, M3-A defaults to the last 7 days, M3-B defaults to the last 30 days, and M6 defaults to the API default range.

Q6: Can prohibited word detection guarantee 100% accuracy? A: Based on the Honghu Hub prohibited words list, for reference only. Please verify against your business scope and actual product effects; the final decision is subject to WeChat platform review results.

Q7: Can the cover designs be used directly? A: Each design includes a 2.35:1 aspect ratio image generation prompt, which can be used directly in AI drawing tools to generate a 900×383 cover image.

Q8: What should I do if the account diagnosis finds nothing? A: The data comes from the Honghu API; for accounts not found,estimation is strictly prohibited. Please check the official account name/ID and try again, or use a more general account name.

Q9: Article/author link jump fails? A: WeChat has risk controls on third-party redirects, so links may be blocked even if correct. It is recommended to copy the title and search within WeChat to view it.

---

14. Use Cases

1. Full workflow for new operators: Topic selection (M1 search viral articles) → M2 generate body text → M3-A generate titles → M6 design cover → M5 check prohibited words → publish with confidence
2. Brand campaign copy polishing: M4 rewrite product manual → M3-A generate 10 titles → M3-B score and select the best
3. Matrix batch output: M2 batch generate differentiated articles by segmented keywords, M6 match different cover styles for each article
4. Compliance spot checks for operational content: M5 paste landing page URL for one-click detection
5. Account diagnosis-driven optimization: M7 diagnose account → identify weaknesses (health/activity/core data/operational standards) → move to M1/M2/M3 to address weaknesses
6. Competitor benchmarking analysis: M7 multi-account comparison + M1 cross-track viral article research → develop differentiated content strategy
7. Viral article-driven creation: M1 query 10w+ rankings → extract viral topic and title features → move to M2/M3 creation pipeline