wechat-md

name

wechat-md

description

Format Markdown articles into WeChat-compatible content using md-wechat.vercel.app via browser automation, then copy the rendered result to the clipboard for pasting into the WeChat public-account editor. Use this skill whenever the user mentions 公众号排版, wechat-md, 微信公众号格式, WeChat formatting, copying to WeChat, or wants any markdown/blog post turned into 微信公众号-ready content. Also triggers on phrases like '排成公众号格式', '复制到公众号', '转微信格式'. Prefer this skill over ad-hoc browser poking — it handles style presets, content injection, and clipboard copying in a reliable automated sequence.

WeChat Markdown Formatter

把 Markdown 变成可直接粘贴到微信公众号后台的排版内容。

Design goals

1. Clipboard is the deliverable — the job isn't done until rendered content is in the clipboard, not just displayed in the editor
2. No lingering processes — the local HTTP server auto-exits after serving the file once, so no cleanup is needed
3. Browser automation agnostic — the workflow uses standard primitives (navigate, evaluate JS, click, snapshot) that work across different automation runtimes
4. Tolerant of broken images — blog articles with relative local image paths will show broken images in the preview, but this never affects the clipboard output

Runtime assumption

This skill requires browser automation capabilities. The exact tool names differ by runtime, but the workflow needs equivalents for:

• Navigate: open or reload a URL in a browser tab
• Evaluate JS: execute JavaScript in the page context and get a return value
• Click: click a DOM element (by ref, selector, or role)
• Snapshot: inspect the current page state (DOM tree / accessibility tree)

Examples:

• OpenClaw: browser(action=navigate), browser(action=act, kind=evaluate), browser(action=act, kind=click), browser(action=snapshot)
• Chrome DevTools MCP: navigate, evaluate, click, snapshot
• Playwright: page.goto(), page.evaluate(), page.click(), etc.

Tab consistency: keep the same tab/target across all calls within one session.

Preferred workflow

Step 1: Prepare a clean markdown file

Strip YAML frontmatter and <!--more--> tags, then write a clean temp file:

node <skill-dir>/scripts/prepare-markdown.js <markdown-file> --strip-frontmatter

Default output: /tmp/wechat-md-input.md

Skip --strip-frontmatter only if the user explicitly wants frontmatter-related content kept.

Step 2: Start a one-shot local HTTP server

The server avoids escaping issues when injecting large markdown into the browser. It auto-exits after serving one request, so no cleanup is needed.

python3 <skill-dir>/scripts/temp-http.py /tmp > /tmp/wechat-server.log 2>&1 &
sleep 0.5
cat /tmp/wechat-server.log

This prints SERVER_URL=http://127.0.0.1:<port>. Capture the URL for Step 4.

Important: The > /tmp/wechat-server.log 2>&1 & pattern with sleep is required to prevent blocking. Read the URL from the log file, not from stdout directly.

Step 3: Open md-wechat and apply the style preset

Navigate to https://md-wechat.vercel.app/, then evaluate this JS to set localStorage style values:

(() => {
  localStorage.setItem('color', '#40B8FA')
  localStorage.setItem('size', '16px')
  localStorage.setItem('codeBlockTheme', 'https://cdn-doocs.oss-cn-shenzhen.aliyuncs.com/npm/highlightjs/11.11.1/styles/atom-one-light.min.css')
  localStorage.setItem('fonts', 'Optima-Regular, Optima, PingFangSC-light, PingFangTC-light, PingFang SC, Cambria, Cochin, Georgia, Times, Times New Roman, serif')
  localStorage.setItem('isMacCodeBlock', 'true')
  localStorage.setItem('isShowLineNumber', 'true')
  localStorage.setItem('isCiteStatus', 'true')
  localStorage.setItem('MD__use_indent', 'false')
  localStorage.setItem('MD__theme', 'default')
  localStorage.setItem('MD__use_justify', 'true')
  return 'done'
})()

Then reload (re-navigate to the same URL) so the preset takes effect.

Step 4: Inject content via fetch + execCommand

The CodeMirror 6 editor on md-wechat.vercel.app does not expose its internal view API through the DOM — cmView, cmTile, and view.dispatch() all fail silently because the framework doesn't attach these properties to DOM elements in this deployment. Instead, use the browser's built-in editing API which CodeMirror listens to:

(async () => {
  const res = await fetch('<SERVER_URL>/wechat-md-input.md')
  const text = await res.text()
  const cmContent = document.querySelector('.cm-content')
  if (!cmContent) return 'error: .cm-content not found'
  cmContent.focus()
  document.execCommand('selectAll')
  const success = document.execCommand('insertText', false, text)
  return success ? 'success:' + text.length : 'insertText failed'
})()

Replace <SERVER_URL> with the URL from Step 2. The server auto-exits after this fetch completes.

execCommand('insertText') works because CodeMirror intercepts native input events from the contenteditable element. The focus → selectAll → insertText sequence is the same thing that happens when a user pastes text, so it goes through CodeMirror's normal input pipeline and properly updates both the editor state and the preview pane.

Step 5: Verify injection succeeded

Inspect the page (snapshot, compact, maxChars ~500). Check:

• The first paragraph matches the article's opening text
• Article headings appear in the preview pane
• Footer shows a reasonable word/character count

Do not fail because:

• Relative local image paths don't render (expected, doesn't affect output)
• Previous session's content still showing (means injection didn't take — retry Step 4)

Step 6: Click 复制

Click the 复制 button (in the top banner area).

The success toast ("已复制渲染后的内容到剪贴板") appears briefly and auto-dismisses. Do not wait for or depend on catching the toast — a successful click is sufficient confirmation.

Step 7: Done

The HTTP server already exited after serving the file in Step 4. No cleanup needed.

---

Compact execution sequence

For quick reference, the entire flow:

1. exec: node prepare-markdown.js <file> --strip-frontmatter
2. exec: python3 temp-http.py /tmp > /tmp/wechat-server.log 2>&1 & then sleep 0.5 && cat /tmp/wechat-server.log → capture SERVER_URL
3. Navigate to https://md-wechat.vercel.app/
4. Evaluate JS: set localStorage preset
5. Navigate (reload) same URL
6. Evaluate JS: fetch from SERVER_URL + selectAll + insertText (server auto-exits after this)
7. Snapshot: verify content
8. Click: 复制 button
9. Report to user

Steps 3-4-5 can merge: set localStorage first, then navigate once (the navigate acts as the reload).

---

Fallbacks

Fallback A: direct inline injection for small files (<3000 chars)

If the file is short enough, skip the HTTP server entirely:

(() => {
  const text = `<escaped markdown content>`
  const cmContent = document.querySelector('.cm-content')
  if (!cmContent) return 'error'
  cmContent.focus()
  document.execCommand('selectAll')
  return document.execCommand('insertText', false, text) ? 'ok' : 'fail'
})()

Only for short articles — long payloads break due to escaping issues in the evaluate call.

Fallback B: legacy create-injection.js (deprecated)

The older create-injection.js script uses cmTile.view.dispatch() which no longer works on md-wechat.vercel.app. Avoid unless the site changes its CodeMirror configuration.

---

Troubleshooting

Problem	Solution
.cm-content not found	Page not ready. Wait 2s and retry.
insertText returns false	.cm-content not focused. Ensure cmContent.focus() runs first.
Old content still showing	selectAll didn't cover everything. Navigate to a fresh page first.
Server log empty after sleep	Increase sleep to 1s. Or check python3 is available.
fetch fails in evaluate	Server may have exited early or not started. Re-run Step 2.
Relative images fail in preview	Expected for local blog images. Does not affect clipboard output.
Toast not visible after 复制	Normal — auto-dismisses in ~2s. Click success = copy success.

---

Reporting format

- source: <markdown-file>
- WeChat formatting: copied to clipboard ✅
- note: relative local images may not render in preview; does not affect pasted output