WeChat Article Skills
Nine skills take a WeChat Official Account article from topic and text to layout and a draft in the backend
- 716 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- Reads aws.env with model keys and the account AppSecret and sends them to the WeChat API and model APIs
- Creates posts in a live account and, in published mode, sends the article to subscribers
- Uploads article text, cover and illustrations to WeChat servers and to external models
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/wechat-article-skillsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Reads aws.env with model keys and the account AppSecret and sends them to the WeChat API and model APIs
- Creates posts in a live account and, in published mode, sends the article to subscribers
- Uploads article text, cover and illustrations to WeChat servers and to external models
This is third-party code. Review the repository files before installing.
What it does
Nine skills turn the agent into an editor for WeChat Official Accounts. The main skill, aws-wechat-article-main, drives the whole chain: topic, text, review, layout, images, publishing. It pauses for confirmation at each step, so you can step in, edit or restart. The other skills also run on their own when the material already exists. Topics offers 3–5 topic cards based on trends. Writing drafts to your specification, calling an external model or writing the text itself. Review checks for banned words and typos, compares the text with the spec and diagnoses signs of AI writing. Formatting converts Markdown to WeChat HTML with seven built-in themes and seven components such as a quote card, checklist and comparison. Images derives 12 cover forms and 8 in-article illustration forms from the meaning of the article. Publishing uses the official WeChat API. By default the article lands in drafts (publish_method: draft), a mass send is a separate switch, and the cover is cropped to 2.35:1. Assets keeps a product knowledge base in the products folder and imports .aws presets from aiworkskills.cn, where settings are filled in a form: account, strategy, writing, visuals, publishing. Sticker builds multi-image series.
Who it is for. For editors and owners of WeChat Official Accounts who run the channel with Claude Code, Cursor, Codex and compatible agents.
Good fit when
- You want to go from topic to a draft in the Official Account backend in one conversation, confirming every step
- The article is ready in Markdown and only needs converting to WeChat HTML with a theme and components
- Before release the text has to pass a banned-word list and a check for signs of AI writing
- You run several accounts: credentials go into numbered slots such as WECHAT_1_APPID and WECHAT_2_APPID
- A release needs a 2.35:1 cover and a series of illustrations derived from the article text
Not a fit when
- You publish somewhere other than WeChat: layout, covers and publishing target Official Accounts only
- You have no AppID and AppSecret for an account: automatic publishing will not work and has to be switched off with publish_method: none
- You need a single image or a short note without a pipeline: nine skills with configs are too heavy for that
- You cannot clone the repository: the skills read aws.env and .aws-article/config.yaml from the root of the working folder
Example request
Help me write a WeChat article about setting up Claude Code for beginners, format it and put it into the account draftsLimitations
- Python 3.10 or newer and PyYAML are required; Pillow is recommended for image compression and cropping.
- Writing and image model keys are optional: without them the config check only warns. Publishing needs WECHAT_1_APPID and WECHAT_1_APPSECRET.
- The set sends the AppID and AppSecret to api.weixin.qq.com and uploads article text and images to WeChat servers and to external models.
- The skills link to each other by relative paths, so all nine folders go into one skills directory.
- The scripts/install-skills.sh script mentioned in the README is not in the repository tree, so the skills are copied by hand.
- Documentation and SKILL.md files are mostly in Chinese; an English README exists.
How to disable. Delete the aws-wechat-article-* and aws-wechat-sticker folders from the agent's skills directory. The aws.env file in the working folder holds model keys and the AppSecret, so remove it together with the .aws-article folder. To keep the skills without publishing, set publish_method: none in .aws-article/config.yaml.
FAQ
How do I install the WeChat skills in Claude Code?
Clone the repository and copy all nine folders from skills into the project's .claude/skills. Create .aws-article/config.yaml and aws.env from the examples in skills/aws-wechat-article-main/references, fill in the keys and run validate_env.py. When the script exits with code 0, tell the agent to help you write a WeChat article.
Does the set publish to all subscribers right away?
No. The default publish_method: draft means publish.py full only writes a draft into the backend. The published mode needs your explicit request or the --publish flag on the full command.
Which keys does it need?
Publishing needs WECHAT_1_APPID and WECHAT_1_APPSECRET, and extra accounts use higher-numbered slots. WRITING_MODEL_API_KEY and IMAGE_MODEL_API_KEY are optional: without them the check only warns. Any model with an OpenAI-compatible API works, for example DeepSeek, Qwen, GLM or Kimi.
What does the Review skill check?
It looks for banned words from the list in the publishing settings, typos and departures from your specification, and reports in three tiers. Since June 1, 2026 it also has a separate dimension that flags signs of AI writing.
Which agents and models does the set work with?
The skills follow the OpenClaw standard and work in thirteen Claw-family tools as well as Claude Code, Autohand Code, Cursor and Codex. Any model with an OpenAI-compatible API can write text and images, and the keys stay local in aws.env.
Discussion
No questions or reviews yet.
Related
SkillsA WeChat article layout skill: turns Markdown into paste-ready HTML with inlined styles that survives pasting into the official account editor
Ten skills run social media content from brief and strategy through platform copy and short video to publishing and review
A CLI for agents that turns Markdown into WeChat HTML, prepares covers and images, and saves drafts into an official account
Content skills: article illustrations, covers, infographics, comics, translation, URL to Markdown and social posting
Agent skill that builds social card carousels and paired covers from an article, screenshots, notes or video in two visual systems
Human Writing
human-writing
A skill for lively Chinese writing: helps the agent write and edit so a specific person seems to speak, not a model
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| WeChat Article Skills (this entry) | no | 716 | |
| md2wechat | no | 3,774 | only converts Markdown to WeChat HTML, prepares covers and saves drafts, with no topic research, writing or text review |
| WeChat Intelligence Hub | no | 2,744 | reads your own WeChat chats for search and digests and publishes nothing |
| Self Media Skills | no | 627 | ten skills run social media content from brief and strategy through |
README in short
The README presents the set as a digital employee for Official Accounts and opens with links to Chinese tutorials and a five-step quick start through the aiworkskills.cn site. It then lists the supported tools, from thirteen Claw-family clones to Claude Code, Autohand Code, Cursor and Codex, and the models: DeepSeek, Qwen, GLM, Kimi, GPT, Claude and anything with an OpenAI-compatible API. A table describes the nine skills and their trigger phrases. Below come notes on layout components and cover derivation, a developer path with hand-edited YAML and a changelog.
README badge
Original SKILL.md
---
name: aws-wechat-article-main
description: 公众号运营｜微信公众号｜公众号一条龙｜公众号全流程｜自媒体运营｜wechat automation｜content pipeline｜AIGC workflow — 公众号一条龙运营总控入口，选题→写稿→审稿→排版→配图→发布串联 8 个子 skill，单条指令完成整篇图文从 0 到上架。面向公众号小编、自媒体、品牌内容。触发词分层：**一条龙流程**「一条龙」「完整流程」「从头做」「从 0 到发布」；**新做新发**「帮我写篇公众号文章」「做一篇公众号文章」「我想发一篇」「帮我发一篇」「再来一篇」；**选题起点**「今天写什么好」「有什么好写的」「找个话题」「爆款选题」「热点选题」「起个爆款标题」；**策划起点**「内容日历」「系列策划」「专栏规划」「连载」；**流程恢复**「接着上次那篇」「继续昨天的」「继续上次的」「接着之前的进度」；**显式模型新写**「用 GPT 写一篇」「用 DeepSeek 写一篇」「把提纲写成文章」。子 skill（topics/writing/review/formatting/images/publish/sticker/assets）单独触发仅限对**已有产物**的修改场景（如"改标题""润色这段""排版""审稿""加封面""发布"）；新做/策划/多环节串联一律走本入口。
homepage: https://aiworkskills.cn
url: https://github.com/aiworkskills/wechat-article-skills
metadata:
openclaw:
requires:
env:
- WRITING_MODEL_API_KEY
- IMAGE_MODEL_API_KEY
- WECHAT_1_APPID
- WECHAT_1_APPSECRET
bins:
- python3
primaryEnv: aws.env
---
# 公众号运营总览
**一键式公众号 AI 内容流水线** —— 从选题到上架 8 个子 skill 串联。
> **套件说明** · `aws-wechat-article-*` 共 9 个 slug：`main / topics / writing / review / formatting / images / publish / assets`，外加 `aws-wechat-sticker`。跨 skill 的相对引用依赖同一 `skills/` 根目录；推荐 `clawhub sync` 一次性全装。源码：<https://github.com/aiworkskills/wechat-article-skills>
**Agent 执行**：确定本 SKILL.md 所在目录为 `{baseDir}`。
## 能力披露（Capabilities）
本 skill 是**编排入口**，真正调用外部 API 的是子 skill（writing / images / publish / sticker / review）。本入口自身只跑 `validate_env.py`：
- **凭证**：读 `aws.env` 的各 `*_API_KEY` 与 `WECHAT_{N}_*`，**仅校验键存在且非空**，值不用于任何网络请求
- **网络**：本入口无外发；**子 skill 有外发**，见各自的能力披露
- **文件读写**：`.aws-article/`、本篇 `article.yaml`
- **shell**：仅 `{python} {baseDir}/scripts/validate_env.py`
> 整体套件会调用外部 LLM、图像与微信 API，并在调用时外发 API key 与本篇内容。
只装 main 一个时仍可做环境校验；进入流水线会因子 skill 缺失而 `file not found`。
## 配置检查 ⛔ BLOCKING
**进入「全局账号约束」「本篇准备」及内容流水线之前**，必须按 **[首次引导](references/first-time-setup.md)** 完整走一遍「检测顺序」——判断 OS、探测 `{python}`、检查两份配置文件、跑 `validate_env.py`、创建预设目录。任一步失败**不得继续**。
那份文档同时是**唯一**的失败引导来源与行为约束（禁止自作主张、不得索取密钥）出处，本文件不再复述。
三个要点在这里说一次：
- **`publish_method` 默认 `draft`**（`publish.py full` 只写公众号草稿箱）。用户明确要求「发出去」才改 `published`，或临时用 `full --publish`。
- **用户明确不接微信** → 先设 `publish_method: none` 再过校验，之后 `full` 会直接跳过。
- **写作 / 图片模型是可选项**，缺失时只警告、不阻断。