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 \x3Cskill-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 \x3Cskill-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)

\x3Cskill-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="\x3Cskill-path>"

# Step 1: Fetch HTML (fast path)
bash "$SKILL_PATH/scripts/fetch.sh" "URL" /tmp/wx_article.html

# Step 2: Inspect metadata
node "$SKILL_PATH/scripts/parse.mjs" /tmp/wx_article.html --json

# Step 3: If title/contentLength is empty or unusable, run browser fallback
python3 "$SKILL_PATH/scripts/fetch-browser.py" "URL" /tmp/wx_article.html
node "$SKILL_PATH/scripts/parse.mjs" /tmp/wx_article.html --json

# Step 4: Parse to Markdown
node "$SKILL_PATH/scripts/parse.mjs" /tmp/wx_article.html > /tmp/wx_article.md

# Step 5: Save with the bundled safe saver
node "$SKILL_PATH/scripts/save.mjs" \
  --markdown /tmp/wx_article.md \
  --config "$SKILL_PATH/config.json" \
  --title "\x3Carticle_title>" \
  --target-path "\x3Ctarget_path>"

The filename should be derived from the article title, keeping it readable: strip special characters, keep Chinese characters. Example: 从Claude Code源码看AI Agent工程架构.md

The bundled save.mjs uses a strict safe save path:

• resolve output only under vault_disk_root
• reject unsafe absolute paths and parent-directory traversal
• create parent folders as needed
• write the Markdown file directly inside the configured vault root

If shell redirection like > /tmp/wx_article.md is blocked by policy, use an equivalent alternate execution shape that still relies on the skill's own parser script, then save the resulting Markdown with the bundled direct-write saver.

Batch save (multiple URLs)

For 2+ URLs, process them sequentially. For 4+ URLs, consider using subagents in parallel (each with its own temp file).

# Per URL:
bash "$SKILL_PATH/scripts/fetch.sh" "$url" "/tmp/wx_${i}.html"
node "$SKILL_PATH/scripts/parse.mjs" "/tmp/wx_${i}.html" > "/tmp/wx_${i}.md"
# Then save each to Obsidian

Output format

The parser produces clean Markdown with YAML frontmatter:

---
title: "Article Title"
author: "公众号名称"
publish_date: "2026-03-31 19:45:08"
saved_date: "2026-03-31"
source: "wechat"
url: "https://mp.weixin.qq.com/s/..."
---

The parser automatically:

• Preserves all article images (WeChat CDN URLs)
• Removes WeChat decoration text (THUMB, STOPPING)
• Merges "PART.XX" + title into proper ## PART.XX Title headings
• Strips promotional tails (关注/点赞/在看, author bios, QR codes)
• Preserves bold, italic, code blocks, blockquotes, lists, links

Post-processing by Claude

After the parser runs, review the output and apply any remaining cleanup:

1. If the user specified tags, add them to the frontmatter
2. Verify the filename is clean and descriptive
3. Confirm save location with the user if ambiguous
4. When saving the final note, resolve the final path as:
   • \x3Cvault_disk_root>/\x3Ctarget_path>/\x3Cfilename>.md
   • Example: \x3Cvault_disk_root>/notes/wechat/文章名.md
   • Do not prepend extra app-specific subfolders like humanlike/ unless the user explicitly asked for that vault subdirectory

Troubleshooting

curl returns empty, shell page, or parser finds no content

Try the browser fallback:

python3 "$SKILL_PATH/scripts/fetch-browser.py" "URL" /tmp/wx_article.html

If browser fallback also fails, the article may require a stronger logged-in context or be a special unsupported page type.

Playwright is missing

If browser fallback fails with missing Playwright modules (for example ModuleNotFoundError: No module named 'playwright'), do not block the whole workflow.

Do this instead:

1. Keep the normal fetch.sh path as the default first attempt
2. Only report Playwright as a missing optional dependency for fallback mode
3. Continue with non-browser flow if the fast path already produced usable article HTML
4. If the user wants, help install Playwright globally/local to improve future fallback success

Policy block vs approval-required

These are different situations and must be treated differently:

• Approval-required: the system created an approval object and the user can explicitly approve that exact command
• Policy-blocked at entry: the command is rejected before an approval object exists

Do not assume every blocked command can be fixed with an approval command. If there is no actual approval object from the runtime, explain that clearly and switch to a different execution shape.

Empty content / no #js_content

Some special article types (mini-programs, video-only) aren't supported. Inform the user.

Parse step blocked by shell shape

If a direct command like:

node "$SKILL_PATH/scripts/parse.mjs" /tmp/wx_article.html > /tmp/wx_article.md

is blocked by policy, do not abandon the skill. Use another execution form that still calls the same parser script and captures its stdout, then save the Markdown with the bundled direct-write saver.

obsidian CLI not available

That is fine. The publish build saves by direct file write to the configured vault disk path.

Important for this setup:

• direct-write saving should use vault_disk_root from config when present
• notes/wechat should resolve inside the configured vault root
• do not prepend extra app-specific subfolders unless the user explicitly asked for them