wxmp-draft

name: wxmp-draft description: End-to-end pipeline to turn a Markdown research report into a WeChat Official Account (公众号) rich-graphics article, upload images, and save as a Draft (草稿) via WeChat APIs. Never publishes. compatibility: opencode metadata: language: zh-CN output: wxmp-draft safety: no-publish license: MIT

What This Skill Does

Turn a .md research report into a WeChat Official Account rich article and save it as a Draft:

1. Rewrite into a WeChat-friendly narrative (70% hardcore tech, 30% sharp metaphors; precise, entrepreneurial).
2. Produce image plan + image captions.
3. (Optional) Generate a cover image and “data card” images locally.
4. Call WeChat APIs to:
   • get access_token
   • upload cover as permanent material and obtain thumb_media_id
   • upload in-article images and obtain URLs
   • create draft via draft/add

Hard rule: DO NOT publish / mass-send. Only draft.

Key Behavior (Per H2 Images + Typography)

• One web-sourced image per H2: with --auto-web-images, the script searches Wikimedia Commons, downloads a suitable image, uploads to WeChat via uploadimg, then inserts it right after each H2 section heading.
• No captions under images: inserted images are <img> only (no visible attribution/caption text).
• Attribution kept in metadata: source/license is stored in cache JSON (not rendered in the article).
• WeChat-stable formatting: headings are rendered as <p style=...> blocks (not relying on <h1>/<h2>/<h3>), so Markdown hierarchy is obvious in the WeChat editor.
• Tech heading styles: Chapter (章) headings use a dark "terminal" panel style; section (节, like 2.1) headings use a light "blueprint" card style; tertiary headings remain minimal.
• Title behavior (from-markdown): WeChat draft title is auto-shortened to ~10 tokens (LLM) while the body keeps the full original Markdown H1 (including bold emphasis).
• No auto subtitle injection (from-markdown): digest is used for WeChat digest field only; it is not inserted under the H1 to avoid duplicating the first paragraph.
• Tables (default): Markdown tables are kept as real HTML <table> and restyled with WeChat-safe inline CSS for mobile readability.
• Tables (legacy image mode): use --table-mode image to convert Markdown tables into {{IMG:table-*}} placeholders (optionally render/upload as images with --gen-images).

Inputs

• Markdown file path (required)
  • Example: /Users/wangweiyang/Downloads/游戏运维监控工具推荐.md
• WeChat config file path (required for live run)
  • Default: ~/Downloads/wxmpdev.config
  • Format supported:
    • env style:
      • APPID=...
      • APPSECRET=...
      • AUTHOR=... (optional)
      • ACCESS_TOKEN=... (optional; temporary token override)
      • PROXY=http://user:pass@host:port (optional; route traffic via whitelisted egress)
      • HTTPS_PROXY=... / HTTP_PROXY=... (optional)
    • JSON style:
      • { "appid": "...", "appsecret": "...", "author": "...", "access_token": "..." }

Safety Rules (Non-negotiable)

• Never log or print secrets:
  • AppSecret, access_token, request bodies containing tokens.
• Do not call publish endpoints:
  • No /cgi-bin/freepublish/*
  • No /cgi-bin/message/mass/*
• WeChat constraints:
  • In-article images must be uploaded via media/uploadimg and use returned URL.
  • Cover must use permanent material media_id as thumb_media_id (commonly uploaded via material/add_material).

Operational Notes (WeChat IP Whitelist)

If your Official Account has IP whitelist enabled, WeChat will reject cgi-bin/token with errcode=40164.

Two ways to proceed:

1. Add this machine's public egress IP to the WeChat IP whitelist.
2. Use ACCESS_TOKEN=... in config (obtain the token from a whitelisted environment).

You can print best-effort public IP via:

wxmp-draft --draft-only --print-egress-ip

Expected Output

• A local result JSON (includes the created draft_media_id on success)
• A local manifest/cache for idempotency (avoid duplicate uploads/drafts)
• A readable out/last_render.html for human inspection

Recommended Execution

1) Draft the article content (LLM rewrite)

Read the input Markdown and produce a structured article spec (JSON) with:

• title (<= 32 chars)
• digest (<= 128 chars)
• author (<= 16 chars, optional)
• content_html (WeChat-safe HTML)
• images: list of assets to upload
  • cover asset: { "id": "cover", "kind": "cover", "path": "...", "alt": "...", "caption": "..." }
  • inline assets: { "id": "table-1", "kind": "inline", "path": "...", "alt": "...", "caption": "..." }

In content_html, reference inline assets using placeholders:

• {{IMG:table-1}}

The uploader will replace placeholders with WeChat-hosted image URLs.

2) Run the uploader (draft-only)

Use the helper script:

wxmp-draft \
  --input "/Users/wangweiyang/Downloads/游戏运维监控工具推荐.md" \
  --config "~/Downloads/wxmpdev.config" \
  --mode from-markdown \
  --table-mode html \
  --gen-images \
  --auto-web-images \
  --web-image-source auto \
  --image-query-mode llm \
  --draft-get-verify \
  --draft-only

Notes:

• To preserve deep-research content, prefer --mode from-markdown and use --image-query-mode llm to let the model generate better image queries only.
• --mode llm rewrites the article body; avoid it if you want the Markdown content kept verbatim.
• --gen-images generates a cover image and renders any remaining non-table placeholders (e.g. diagrams/checklists) into image cards.
• For table images specifically, you must use --table-mode image.
• --inline-cover (optional) inserts the cover image into the body; default is off to keep the strict "one image per H2" rule.

Test Command (No Publish)

After config exists, verify the end-to-end draft save:

wxmp-draft \
  --input "/Users/wangweiyang/Downloads/游戏运维监控工具推荐.md" \
  --config "~/Downloads/wxmpdev.config" \
  --mode from-markdown \
  --gen-images \
  --auto-web-images \
  --draft-get-verify \
  --draft-only

This must print a final line containing: Draft created: media_id=....