AIhunannan/wechat-article-workflow-skill

WeChat Article Workflow

This skill is for personal-brand WeChat articles. Default to editing the author's own reflections, not generic ghostwriting.

Voice Rules

• Start from the user's notes and preserve his personal judgment.
• Keep the article concise, usually under 1000 Chinese characters unless requested otherwise.
• Prefer plain, sharp, reflective writing over polished marketing language.
• Avoid AI-heavy patterns: formulaic openings, over-explaining, slogan endings, excessive lists, and generic inspirational tone.
• Use a local humanizer-zh skill when available and apply its rules.

Title Rules

• Treat the title as a growth lever, not a label. A good title can matter more than a good paragraph.
• Before finalizing, generate 5-10 candidate titles and choose the one with the strongest curiosity gap, specificity, and forwarding impulse.
• Prefer titles with concrete hooks: numbers, extreme contrast, hidden origin stories, surprising causal claims, famous names, or high-stakes consequences.
• Use the user's proven pattern as a benchmark: 一篇文章到底讲了啥？让硅谷集体失眠，24小时内疯传，7000万阅读.
• Good title directions:
  • X诞生前，先发生了一个没人重视的小故事
  • 真正改变X的，不是Y，而是Z
  • 为什么我越来越相信：X比融资更重要
  • 一个被忽视的小实验，后来长成了X
• Keep the final WeChat title within 32 Chinese characters when possible, but do not make it bland just to be short.
• Avoid flat summary titles such as 关于XX的思考, XX给我的启发, 读XX有感, unless the user explicitly wants a low-key diary tone.

Standard Workflow

1. Clarify the article angle only if genuinely needed. Otherwise make a reasonable editorial judgment.
2. Write or polish article.md with frontmatter:
   • title: final WeChat title, max 32 chars.
   • author: configured author name.
   • digest: max 128 chars.
3. Before locking the title, create a short title shortlist internally and pick the strongest one. If the user criticizes the title, revise the title first and republish if needed.
4. Check length with non-whitespace count and keep it near the requested limit. Around 1000 Chinese characters is acceptable when the story needs breathing room.
5. Generate or select images:
   • Prefer Codex built-in image generation for article art unless the user explicitly requests a provider.
   • Generated images are usually under $CODEX_HOME/generated_images/...; copy the chosen images into the article images\ folder and leave originals in place.
   • Default to 3-5 total in-body images per article, including the cover if it should appear in the body.
   • Use one strong 16:9 cover plus 2-4 body images that match specific paragraphs or arguments.
   • Each image must have a job: prove authenticity, explain a concept, create emotional atmosphere, show a product/object, or summarize a key model. Do not add generic decoration.
   • When the user provides real photos, prefer them for authenticity and place them early if they strengthen trust.
   • For AI-generated images, make prompts specific to the paragraph: people, objects, scene, metaphor, mood, color palette, and explicit no text, no logos constraints.
   • Avoid cheap futuristic imagery, random robots, generic business icons, and images that only loosely match the paragraph.
   • If the user expects every generated image to appear in the article body, embed the cover as the first body image too; WeChat cover material alone does not display inside article content.
6. Build WeChat-safe HTML:
   • Use inline styles only.
   • Avoid relying on <style> tags for final WeChat body.
   • Keep the official WeChat title out of the body H1 to avoid duplicate title risk.
7. Run md2wechat inspect before publishing:
   • Confirm title, author, digest, images, upload readiness, draft readiness.
   • If md2wechat API theme conversion fails, use the local inline HTML path and direct WeChat draft script.
8. Publish to WeChat draft:
   • Prefer a direct draft script that explicitly sets title, author, digest, content, and cover media id.
   • Never use a flow that may create a default title such as "AI测试文章".
   • If WeChat returns 40164 invalid ip ... not in whitelist, report the exact IP shown by WeChat. This IP can differ from generic public-IP lookup services, so trust the WeChat error.
9. Archive to Obsidian:
   • Base path: the configured local knowledge-base root.
   • Article path: 50_输出成果\IP文章成品\YYYY\YYYY-MM\YYYY-MM-DD_文章标题\ when using the default Obsidian layout.
   • Save 原始感想.md, 文章.md, article_wechat.html, 发布信息.md, and 素材\.
   • Update 50_输出成果\IP文章成品\IP文章时间轴.md.

Directory Convention

Desktop working directory:

<workspace>\YYYY-MM-DD_slug\

Recommended files:

article.md
article_wechat.html
publish_wechat_direct.py
draft_result.json
draft_error.json
images\cover.png
images\body_01.png
images\body_02.png
images\body_03.png

Publishing Helper

Use scripts\publish_wechat_direct.py from this skill when md2wechat conversion or test-draft is unreliable.

Example:

python scripts\publish_wechat_direct.py `
  --root C:\path\to\article_folder `
  --html article_wechat.html `
  --cover images\cover.png `
  --inline images\kpt_formula.png `
  --title "别迷信天赋，成功靠知识和实践" `
  --author "Author Name" `
  --digest "MIT教授Patrick Winston的一个公式，让我重新理解创业：别迷信天赋，先积累知识，再走进真实世界练习。"

The helper reads WeChat appid and secret from ~\.config\md2wechat\config.yaml, uploads body images with uploadimg, uploads cover as permanent material, creates a draft via /cgi-bin/draft/add, and writes draft_result.json or draft_error.json.

Final Response Checklist

• State whether the WeChat draft was created.
• If successful, include draft media_id.
• Include the final title.
• Mention the number of in-body images if the article was prepared for publishing.
• Link to local article and Obsidian archive paths.
• Mention any real blocker, especially IP whitelist errors.