WeChat Article → Obsidian

Save WeChat MP articles (微信公众号) as clean Markdown notes in Obsidian. No browser, no CDP, no plugins needed.

How it works

WeChat articles are fetched with a two-stage strategy. First, try a normal curl with browser-like headers. If that only returns a shell page and the parser cannot find usable metadata / #js_content, fall back to a real browser fetch using Playwright + Chrome, extract the rendered article body, then convert it to clean Markdown and save it into Obsidian.

Dependencies

• curl (pre-installed on macOS/Linux)
• Node.js >= 18
• Playwright + a Chromium/Chrome browser (only needed for fallback mode)
• obsidian CLI (optional for vault-side workflows, not required for saving)

First-time setup

On first use, check <skill-path>/config.json. If obsidian_vault or default_path is empty, ask the user:

1. "What is your Obsidian vault name?" — this is the vault name used with obsidian CLI (e.g., vault=MyVault)
2. "Where should I save WeChat articles by default?" — a path inside the vault (e.g., notes/wechat, articles/wechat)

Then write the answers to <skill-path>/config.json:

{
  "obsidian_vault": "MyVault",
  "default_path": "notes/wechat"
}

This only needs to happen once. After that, the skill uses the saved config automatically.

Configuration

Natural language override (per-request)

The user can override the default path anytime:

• "把这篇文章存到 reading/tech 目录"
• "save this under articles/ai/"
• "导入到 Obsidian 的 inbox 文件夹"

Parse the target path from the user's message and use it instead of default_path.

Config file (persistent default)

<skill-path>/config.json:

• obsidian_vault: the vault name for obsidian CLI
• default_path: where to save articles when the user doesn't specify a path
• vault_disk_root: absolute disk path of the vault root for direct-write saving

Recommended setup:

• obsidian_vault: optional vault name reference
• default_path: notes/wechat
• vault_disk_root: absolute disk path to the user's Obsidian vault

Workflow

Preferred execution strategy (important)

Use a step-by-step workflow by default. Do not assume the one-line "fetch + parse + save" pattern is the most reliable option in OpenClaw.

Preferred order:

1. Fetch HTML with fetch.sh
2. Inspect metadata with parse.mjs --json
3. Only if metadata/body is unusable, try browser fallback with Playwright
4. Parse HTML into Markdown
5. Save to Obsidian. For the publish build, use the bundled safe direct-write saver.

Why this matters:

• some exec forms may be blocked by environment policy even when the underlying task is allowed
• a blocked exec does not always mean a user approval can fix it
• splitting the flow makes it much easier to identify whether the failure is in fetching, parsing, browser fallback, or vault writing

Single article

SKILL_PATH="<skill-path>"