SKILL.md

$28

✅ AI WeChat Article Writer installed!

Tell me your topic and I'll write and publish a WeChat article for you.

Try it now: "帮我写一篇关于 AI 编程的公众号文章"

What it does:

Plan topics from trending hotspots and SEO keywords

Write professional articles with de-AI voice

Format with beautiful WeChat-optimized themes

Generate cover images with AI (4 providers supported)

Publish directly to your WeChat draft box

Setup (one-time):

Install & configure: cd toolkit && npm install && npm run build && cd .. && pip install -r requirements.txt && cp config.example.yaml config.yaml

Get YouMind API Key → fill youmind.api_key in config.yaml

Get WeChat AppID & AppSecret from 微信开发者平台 → fill wechat.appid and wechat.secret, add your public IP (curl -s https://ifconfig.me) to the API IP whitelist

No WeChat API yet? You can still write and preview locally — just skip the WeChat config steps.

See the Setup section below for detailed step-by-step instructions with screenshots.

Need help? Just ask!

For first-run setup and client onboarding details, see references/operations.md.

Usage

Provide a topic, brand/client name, or raw Markdown for publishing.

Write from a topic:

帮我写一篇关于 AI 编程趋势的公众号文章

Write for a specific client:

给 demo 客户写一篇推文，主题是远程办公最佳实践

Format and publish raw Markdown:

把这篇 Markdown 排版成公众号样式并发布到草稿箱

Interactive mode:

用交互模式帮我写一篇公众号文章，我想自己选题和框架

Setup

Prerequisites: Node.js ≥ 18, Python ≥ 3.9, a verified WeChat Official Account with API access.

Step 1 — Install Dependencies

cd toolkit &#x26;&#x26; npm install &#x26;&#x26; npm run build &#x26;&#x26; cd ..

pip install -r requirements.txt

Step 2 — Create Config File

cp config.example.yaml config.yaml

Step 3 — Get YouMind API Key (Recommended)

YouMind API Key 用于知识库语义搜索、联网搜索、文章归档、AI 生图（Nano Banana Pro）。不配也能跑，但会丧失这些增强能力。

打开 YouMind API Keys 页面

登录后点击 「Create API Key」 创建新密钥

复制生成的 sk-ym-xxxx 格式密钥

填入 config.yaml 的 youmind.api_key 字段

youmind:

  api_key: "sk-ym-xxxxxxxxxxxxxxxxxxxx"

Step 4 — Get WeChat AppID & AppSecret

打开微信开发者平台，点击 「前往使用」 登录

在「我的业务」面板点击 「公众号」 进入管理页

在 基础信息 页顶部复制 AppID

在「开发密钥」区域点击重置获取 AppSecret（仅展示一次，立即保存）

填入 config.yaml：

wechat:

  appid: "wx_your_appid"

  secret: "your_secret"

  author: "你的作者名"

详细图文步骤见 README.md

Step 5 — Configure IP Whitelist

微信公众号 API 拒绝所有不在白名单中的 IP 请求，必须配置后才能发布。

获取公网 IP：

# macOS / Linux

curl -s https://ifconfig.me

# Windows PowerShell

(Invoke-WebRequest -Uri "https://ifconfig.me" -UseBasicParsing).Content.Trim()

拿到 IP 后，在上一步的微信开发者平台公众号页面 →「开发密钥」→ API IP 白名单 → 点击编辑 → 粘贴 IP 保存。

家庭宽带 IP 会变。发布报 IP 错误时重新获取 IP 并更新白名单即可。云服务器 / CI 环境通常是静态 IP，配一次就行。

Step 6 — Image Provider Keys (Optional)

配图使用降级链：AI 生图 → 图库搜索 → CDN 预制封面下载 → 仅输出 prompt。不配任何 key 也不影响发布。

Provider

获取方式

config.yaml 字段

YouMind (Nano Banana Pro)

使用 Step 3 的 YouMind API Key，无需额外配置

image.providers.youmind.api_key（留空则自动使用 youmind.api_key）

Google Gemini (Imagen)

Google AI Studio 创建 API key

image.providers.gemini.api_key

OpenAI (GPT Image)

OpenAI Platform 创建 API key

image.providers.openai.api_key

豆包 (Seedream)

火山引擎控制台创建 API key

image.providers.doubao.api_key

在 config.yaml 中设置 image.default_provider 指定首选 provider，或留空让 Skill 自动选第一个有 key 的。

Verify Setup

配置完成后，对 Agent 说一句试试：

"帮我写一篇关于 AI 编程的公众号文章"

如果配置有问题，Skill 会在对应步骤报错并给出修复提示——不会整体卡死。

For client onboarding and post-setup operations, see references/operations.md.

Skill Directory

This skill is a folder. Read files on demand — do NOT load everything upfront.

Path

Purpose

When to read

references/pipeline.md

Full step-by-step execution (Steps 1–8)

When running the writing pipeline

references/operations.md

Post-publish commands, client onboarding, themes, first-run setup

When handling operational tasks

references/writing-guide.md

Pre-writing framework, depth architecture, de-AI protocol, voice

Step 4 (writing)

references/frameworks.md

5 article frameworks with execution detail

Step 3.5 (framework selection)

references/topic-selection.md

4-dimension topic evaluation model

Step 3 (topic generation)

references/seo-rules.md

Title optimization, keyword density, digest, tags

Step 5 (SEO pass)

references/visual-prompts.md

Cover and inline image design, prompt engineering

Step 6 (visual AI)

references/theme-dsl.md

Custom theme design language (integrates Impeccable if installed)

When creating custom themes — auto-detects & offers to install Impeccable

references/youmind-integration.md

Knowledge base API, search, archiving

When using YouMind features

references/cli-reference.md

All CLI command syntax

When running toolkit commands

references/wechat-constraints.md

WeChat platform technical limits, safe CSS, size caps

When debugging rendering or format issues

references/style-template.md

Client config template with field guide

When onboarding a new client

references/openapi-document.md

YouMind OpenAPI full endpoint schemas

When calling YouMind API directly

references/skill-maintenance.md

Skill self-maintenance, validation, architecture guardrails

When improving or refactoring this skill itself

references/builtin-themes.json

CSS examples for 10 built-in themes

When customizing themes

clients/{client}/style.yaml

Client brand voice, topics, blacklist, theme

Step 1 (load config)

clients/{client}/playbook.md

Client-specific writing rules (if exists)

Step 4 (writing)

clients/{client}/history.yaml

Published article history

Step 2.5 (dedup)

config.yaml

API credentials (WeChat, YouMind, image providers)

Step 1 (first-run check)

toolkit/dist/*.js

Executable scripts (run from toolkit/)

Various steps

scripts/*.py

Python scripts (trending topics, SEO keywords)

Steps 2, 2.5

Execution Modes

Auto (default): Run Steps 1–5 automatically. Before Step 6 image generation, proactively ask once about image scope and style unless the user already specified them. Then continue through Steps 6–8. Only pause elsewhere if a step AND its fallback both fail, required info is missing, or user explicitly asks to pause.

Interactive: Triggered by "interactive mode", "let me choose", "show me the topics/frameworks/themes". Pauses at: topic selection, framework choice, image plan, theme selection. All other steps run automatically.

Critical Quality Rules

Non-negotiable. Violating any one means the article has failed:

**Read references/writing-guide.md BEFORE writing.** The pre-writing framework and de-AI protocol are mandatory.

Zero AI-sounding text. Run the full 4-level de-AI protocol from writing-guide.md.

H1 title: 20–28 Chinese characters. The converter extracts H1 as the WeChat title.

Digest: ≤54 Chinese characters. WeChat enforces a 120 UTF-8 byte limit.

Word count: 1,500–2,500. Sweet spot for completion rate is 1,500–2,000.

Specificity over abstraction. Every claim must be grounded in concrete detail.

Depth over polish. Run the Depth Checklist (writing-guide.md) before the De-AI pass. If the article's core thesis is something from the top 3 Google results, it needs a rewrite, not a polish.

**Obey the client's blacklist** — both words and topics. No exceptions.

Playbook overrides writing-guide. If playbook.md exists for this client, it takes priority for voice and style decisions.

Before generating visuals, proactively ask about image scope and style. Do not silently assume. If the host supports AskUserQuestion, use it. Otherwise ask a concise plain-text question.

Always publish to drafts. Step 7 publishes directly to WeChat draft box. Do NOT ask — this is mandatory and automatic.

Pipeline Overview

Read references/pipeline.md for full execution details of each step.

Step

Action

Key reference

Load client style.yaml + routing

—

1.5

Mine YouMind knowledge base for source material

youmind-integration.md

Fetch trending topics via fetch_hotspots.py

—

2.5

Dedup against history.yaml + SEO keyword scoring

—

Generate 10 topics, score, select best

topic-selection.md

3.5

Generate 5 framework proposals, select best

frameworks.md

Write article with pre-writing thinking + depth check

writing-guide.md

SEO optimization + full de-AI pass

seo-rules.md

Design and generate cover + inline images

visual-prompts.md

Publish to WeChat drafts (mandatory, automatic)

cli-reference.md

7.5

Append to history + archive to YouMind

youmind-integration.md

Report results: title, digest, tags, media_id

—

Routing shortcuts:

User gave a specific topic → Skip Steps 2–3, go 1.5 → 3.5

User gave raw Markdown → Skip to Step 7

Resilience: Never Stop on a Single-Step Failure

Every step has a fallback. If a step AND its fallback both fail, skip that step and note it in the final output.

Step

Fallback

1.5 Knowledge mining

Skip, empty knowledge_context

2 Trending topics

YouMind web-search → WebSearch → ask user

2.5 SEO scoring

Self-estimate, mark "estimated"

3 Topic generation

Ask user for a manual topic

6 Image generation

Output prompts, skip images

7 Publishing

Generate local HTML preview

7.5 History/Archive

Warn, continue

Python/Node missing

Tell user install command

Operations

For post-publish commands (polish, rewrite, change theme, stats review), client onboarding, learn-from-edits, custom themes, and first-run setup, read references/operations.md.

If the request is about improving this skill itself, refactoring its structure, or checking for documentation drift, read references/skill-maintenance.md.

Gotchas — Common Failure Patterns

"The AI Essay": The article reads like a well-organized explainer piece — correct, comprehensive, boring. Fix: re-read writing-guide.md's voice architecture and pre-writing framework. The article needs a PERSON behind it, not an information system.

"The Generic Hot Take": Writing about a trending topic without adding any insight beyond what is already in the top 10 search results. If you cannot identify your unique angle in one sentence, pick a different topic.

"The Word-Count Pad": Hitting 2,000 words by being verbose instead of being deep. Every paragraph should survive the test: "if I delete this, does the article lose something specific?" If not, delete it.

"The Pretty But Empty Article": Beautiful formatting, nice images, zero substance. Visual quality cannot compensate for thin content. Get the writing right first.

"The Blacklist Miss": Forgetting to check style.yaml blacklist against the final article. Always do a final scan before publishing.

"The Broken Pipeline Halt": Stopping the entire flow because one step failed. NEVER do this. Use the fallback. If the fallback fails, skip and note it. The user can always fix individual pieces manually.

Comparison

Feature

This Skill

Manual Writing

135 Editor

Xiumi

AI topic mining

✅

❌

De-AI voice protocol

✅

N/A

❌

AI cover generation

✅ (4 providers)

❌

One-click draft publishing

✅

❌

YouMind knowledge base

✅

❌

Learn from your edits

✅

N/A

❌

Multi-client management

✅

❌

Custom themes

✅ 4 built-in + DSL

N/A

✅

References

YouMind API: see references/openapi-document.md

CLI commands: see references/cli-reference.md

YouMind Skills gallery: https://youmind.com/skills?utm_source=youmind-wechat-article

Publishing: shared/PUBLISHING.md

youmind-wechat-article