SuperAPI AI Directory
Partner中文AI newsFavorites
SkillsGeek-skills-wechat-article-writer

Geek-skills-wechat-article-writer

Writing & research

View repository

Updated 10-16 06:52

About this skill

WeChat Official Account Article Creation Assistant

Overview

Help users create high-quality WeChat Official Account articles. The core capability isflexible style adaptation—the same skill can write both formal copy for corporate official accounts and technical blogs for personal IP. Automatically matches the most suitable style and structure based on user needs.

Step 1: Style Identification and Material Analysis

Input Processing

  • Website link → Use web_fetch to retrieve the full content
  • Text material → Analyze directly
  • Image material → Identify content and extract key information

Automatic Style Identification

Automatically determine the style based on the user's description, with no need to manually select each time:

Signal Matching Style
"company account", "official", "corporate promotion", "brand" 🏢 Official Copy
"my Official Account", "technical blog", "personal sharing", "hands-on", mentions "Kazike" 🧑‍💻 Technical Blog
"event recap", "sharing session", "community", "meetup" 🎤 Event Recap
"review", "comparison", "experience report" 🔍 Review Comparison
Unable to determine → Ask the user: "Is this article for a corporate official account or a personal account? Should the style be more formal or more relaxed?"

Quick Requirements Confirmation

Ask only when information is severely insufficient (no more than 2 questions):

  • Target readers? (general users / technical developers / industry professionals / parents and students)
  • Article length? (short article 800-1500 words / medium 1500-3000 words / long article 3000+ words)

When information is sufficient, start creating directly.

Step 2: Content Research

Use web search to supplement content(applicable to all styles):

Choose a search strategy based on the article type:

Technical Articles：

  • Search technical documentation, GitHub repositories, developer blogs, official changelogs
  • Search technical community discussions (StackOverflow, Zhihu technical topics, V2EX)
  • Find comparison information for competing/similar tools
  • Example queries: "Claude Code Skills 2.0 update eval benchmark", "OpenClaw deployment architecture"

Corporate/Product Articles：

  • Search industry reports, market data, authoritative analysis
  • Search competitor features and user reviews
  • Find real cases and success stories
  • Example queries: "AI language learning industry report 2025 2026", "speaking assessment competitors Liulishuo ELSA"

Event/Community Articles：

  • Search for technical background related to the event
  • Find the history and positioning of the community/organization
  • Understand the technical details of the speaking topics

Step 3: Create by Style

🏢 Style A: Corporate official copywriting

Applicable to: product launches, corporate updates, brand promotion, industry white papers

Tone: Professional and trustworthy, warm but with a sense of distance. Reference references/style-guide.md。

Core features：

  • Use "we" to represent the brand, not "I"
  • Formal written language, avoid colloquialisms and internet slang
  • Data and cases support every point
  • Positive and forward-looking, highlight solutions
  • Follow sensitive word avoidance (references/style-guide.md Section 7)

Structure：

  1. Engaging opening (100-200 words) → pain point/trend/story hook
  2. Core content (3-5 subheadings expanded) → each paragraph 300-500 words, interspersed with data and cases
  3. Strong ending (100-150 words) → summary + CTA call to action

Title style: Reference references/title-formulas.md, leaning toward authoritative, numeric, and benefit-oriented types.

Example tone：

In the wave of digital transformation, language learning is undergoing a profound change. OneOneTalk, based on cutting-edge AI speech technology, brings learners a brand-new speaking assessment experience—real-time feedback, multi-dimensional scoring, and personalized learning suggestions, so every time you speak you gain something.

🧑‍💻 Style B: Personal tech blog

Applicable to: technical tutorials, tool reviews, hands-on sharing, opinion commentary, in-depth analysis of AI topics

This is the mode with the greatest difference from the official style. The core is the feeling of "a tech person chatting about tech with friends".

Tone: Personal, opinionated, down-to-earth but with depth.

Core features：

  • Use "I" as the narrative perspective—"I gave it a try", "my experience is", "to be honest"
  • Can include subjective judgments and personal leanings—"I think this update is pretty explosive", "frankly, this feature still has room for improvement"
  • Colloquial expressions are OK—"get it done", "so good", "hit a pitfall", but don't overdo it
  • Technical content must be precise—concepts, parameters, and code must not contain errors
  • Include a hands-on process—not secondhand information retelling, but with the texture of "after I actually used it, I found..."
  • Appropriate humor and complaints are a plus

Technical content writing standards：

Code examples: use``` to mark code blocks and indicate the language type

# 示例：创建一个skill的eval
python -m scripts.run_loop \
  --eval-set evals/trigger-eval.json \
  --skill-path ./my-skill \
  --max-iterations 5

Command-line operations: wrap in code blocks and add comments for explanation

# 安装依赖
pip install -r requirements.txt
# 启动服务
python main.py --port 8080

Technical concept explanations: first speak plainly, then give the terminology

Simply put, eval is like giving your skill a "mock exam"—throw in a few test questions and see whether the skill can respond correctly. Technically it's called evaluation, but you can understand it as automated testing.

Structure：

  1. Hook opening (50-100 words) → an introduction that grabs the reader, which can be a question, a surprising discovery, or a story
  2. Background setup (100-200 words) → why write this article, the ins and outs of this matter
  3. Core content (main body) → technical details + hands-on process + screenshots/code, divided into sections with subheadings
  4. Personal opinion/summary (100-200 words) → my view, who it suits, what to do next
  5. Interaction prompt → "Have you used it? Welcome to chat in the comments"

Title style(for tech blogs only):

  • Hands-on type: "I rebuilt the eval system with Claude Code Skills 2.0, tripling efficiency"
  • In-depth analysis type: "Skills 2.0 deep dive: a complete practical guide from eval to benchmark"
  • Opinion type: "Why is Skills 2.0 a must-have upgrade for AI developers?"
  • Pitfall type: "Claude Code Skills deployment pitfalls: these 5 problems almost made me give up"
  • Comparison type: "Coze vs Claude Code Skills: the ultimate comparison of two AI workflows"

Negative examples——Technical blogs absolutely must not be written like this:

  • ❌ "In recent years, with the rapid development of artificial intelligence technology..." (empty opening)
  • ❌ "This article will introduce you in detail..." (academic paper style)
  • ❌ Not a single line of code or command in the whole article (it's no longer a technical article)
  • ❌ Only paraphrasing official documentation, with no personal experience or opinions
  • ❌ Official tone: "This feature is designed to provide developers with better..."

🎤 Style C: Event/Community Recap

Applicable to: tech sharing session recaps, community meetups, hackathons, offline events

Tone: Has a sense of being on-site, has human touch, balances technical content and community atmosphere.

Core features：

  • Opening creates a sense of being on-site—time, place, number of people, atmosphere
  • Technical content and people's stories interwoven—neither purely technical nor a plain running account
  • Specific details and moments—a question from an audience member, reactions to a demo, discussions during the tea break
  • Significance and outlook for the community
  • Photos are very important—provide detailed image suggestions (venue panorama, speakers, interactive sessions, group photo)

Structure：

  1. On-site opening → "On the afternoon of March 8, the meeting room at Mofang in Pazhou, Guangzhou was filled with 30 developers..."
  2. Event highlights (2-3 key segments) → each segment: theme + core content + audience reaction
  3. Technical takeaways → extract the most valuable technical points from the talks
  4. Community voices → feedback from participants, interactive highlights
  5. Outlook and preview → next event, follow-up community plans

🔍 Style D: Review Comparison

Applicable to: product reviews, tool comparisons, technical solution selection

Tone: Objective and in-depth, but can include personal preferences.

Core features：

  • Clear review dimensions and scoring system
  • Real usage experience (rather than listing parameters)
  • Both pros and cons must be stated—no advertorial
  • Clear conclusion—"If you're in scenario XX, choose A; if in scenario YY, B is more suitable"

Structure：

  1. Review background—why this comparison is being made
  2. Definition of review dimensions—from which angles to compare
  3. Dimension-by-dimension comparison (table + textual explanation)
  4. Overall recommendation—suggestions for different scenarios

Step 4: Title Creation

Provide for all articles 3 alternative titles：

  • Option A: Best fits the main idea of the content
  • Option B: More shareable/curiosity-driven
  • Option C: More SEO/search-friendly

General title principles：

  • 13-20 characters (optimal length for WeChat official account display)
  • Contains core keywords
  • Content must match; no clickbait titles
  • Use punctuation appropriately (! ? ...) to strengthen tone

Official article titles → refer to references/title-formulas.md Technical blog titles → refer to the title formulas in "Style B" of this document

Step 5: Formatting and Delivery

Formatting Suggestions(common to all styles):

  • Short paragraphs: no more than 3-4 lines per paragraph on mobile
  • Subheadings: one subheading about every 500 words; emoji can be used to enhance recognizability
  • Bold key points: core viewpoints, key data
  • Image rhythm: 1 image every 500-800 words (position suggestions)
  • Code blocks: display code in technical articles in a monospaced font

Final Deliverables：

  1. Complete article body (including title)
  2. 3 alternative title options
  3. Summary (within 100 words, used as the description when sharing)
  4. Image suggestions (cover image + image positions in the body)
  5. Publishing time suggestions
  6. Suggested interactive topics for the comment section

Output Format：

  • Short article (<1500 words) → output directly in the conversation
  • Long article (>1500 words) → recommend generating a .md or .docx file

Step 6: Quality Self-Check

Refer to references/quality-checklist.md for the final review, focusing on:

All Styles：

  • Are facts and data accurate? Are cited sources reliable?
  • No typos or grammatical errors?
  • Is the article structure clear and are paragraph transitions natural?
  • Do the title and content match? Is it not clickbait?

Additional Checks for Official Style：

  • No absolute wording (best, first, only)?
  • No sensitive words? Compliant with platform rules?
  • Consistent tone—always formal yet warm?

Additional Checks for Technical Blogs：

  • Are the code/commands runnable? Is the syntax correct?
  • Are technical concepts explained accurately?
  • Is there hands-on experience or personal opinion? (Pure relay of information has low value)

Anti-Translationese Check (applicable to all styles)：

Refer to references/quality-checklist.md §2.3 and execute item by item the scan for the five types of translationese markers (physical action verbs, adjective + colon opening pattern, abstract noun skeletons, translatable English words, paragraph rhythm). The criterion is not the word itself, but whether the word in the current sentence comes from direct English translation—"站不住脚" is an idiom, keep it; "这个论证被击穿了" is a direct translation of break through, replace it.

Mode Selection: default to conservative mode (accuracy first; if uncertain, keep the original expression). For WeChat official account opinion pieces, literary creation, and marketing copy scenarios, use aggressive mode (allow reorganizing entire paragraphs with a native Chinese mindset, sacrificing a small amount of literal precision in exchange for naturalness). The user can specify the mode.

The above five categories are the most frequent translationese markers; if after reading through it still feels awkward, return to the principle: say it again the way it would originally be said in Chinese.

Step 7 (Optional): L4 Anti-Translationese Polish

When an article pursues a high standard of Chinese naturalness (ChaoGeek in-depth articles, opinion pieces, brand copy), enable this step after passing the Step 6 self-check. This corresponds to the self-refinement path validated by Li et al. (ACL 2025) and Raunak et al. (2024)—have the model rewrite its own draft as an independent reviewer, and naturalness jumps up a notch.

How It Works: Take the first draft as input, switch to the "Anti-Translationese Editing" role, and perform the following operations:

  1. Checkup: Scan the first draft item by item according to the five categories of markers in §2.3, and list the problem location and fragment for each item. If there are no problems, write "None detected"
  2. Rewrite: Provide a thoroughly idiomatic rewritten version. When rewriting, use aggressive mode—allow sentences or even entire paragraphs to be reorganized with a native Chinese mindset
  3. Sentence-by-Sentence Verification: After rewriting, check: each sentence is ≤ 25 characters or has a natural pause; each sentence has ≤ 2 instances of "的"; verbs are content verbs, not "进行/作出/予以"; subjects are people or concrete things, not abstract nouns; sentence lengths within a paragraph alternate between long and short
  4. Accuracy Fallback: For each change, if the original meaning becomes vague, roll back that change and keep the first draft's expression

Trigger Conditions：

  • The user specifies "L4" or "deep polishing" or "polish"
  • The article belongs to the ChaoGeek brand (Style B personal tech blog + brand tone copy)
  • The user explicitly requests "remove all translationese"

Does Not Trigger: For scenarios where accuracy takes priority, such as technical documentation, PRDs, and academic translation, the conservative mode in Step 6 is sufficient.

Resource Documents

  • references/style-guide.md — Official copy language style guide. When creating in 🏢official style,must be referenced。
  • references/writing-techniques.md — WeChat Official Account writing techniques and opening formulas. Can be referenced for all styles.
  • references/title-formulas.md — 50+ title templates. Mainly applicable to official and mass-oriented articles.
  • references/quality-checklist.md — Content quality review checklist. Referenced before publishing all articles.
  • references/linguistic-background.md — The linguistic root causes of anti-translationese. Does not affect execution; provides maintainers with the "why".
  • assets/template-structure.md — Reusable article structure templates.
  • assets/emoji-library.txt — An emoji asset library suitable for WeChat Official Accounts.