huashu-weread

huashu-weread is a Chinese-language add-on for WeRead, the WeChat reading app. WeRead recently shipped its own official AI skill that exposes eight account APIs, like the user's bookshelf, notebooks, reading statistics, and store recommendations. The README says that the official skill is really just a search interface with a natural-language wrapper. If you ask it to recommend a book, it does not check what you already own or what notes you have made, so it often suggests books that have been sitting on your shelf for years.

This repository adds a layer of prompt and workflow logic on top of those eight official APIs. It does not replace the official skill: the official one still runs underneath. The core idea is what the author calls bookshelf-plus-notes cross analysis. Your shelf categories show what you have actively sorted as interesting, and your notebooks show which books you actually read. Looking at only one side gives a misleading picture, so the skill joins them, classifies each book as truly read, sitting unread, hidden deep reading, or skim-and-stop, and uses that view to drive its replies.

The package ships four workflows. Advisor recommends what to read next by looking for gaps in your shelf and notes. Path turns a topic you want to learn into a beginner-to-advanced reading ladder, after first checking what level you are at. Alchemy reorganises your highlights into structured notes grouped by theme. Review writes a year-in-reading style article for sharing on social platforms like Moments, WeChat Official Accounts, or Xiaohongshu. Each workflow has its own document inside the workflows folder, and the system routes a vague request like which book should I read next to the right workflow automatically.

Installation runs through skills.sh: you install the official WeRead skill first, get an API key, then run npx skills add alchaincyf/huashu-weread. The README notes that the skill is agent-agnostic and works with Claude Code, Cursor, Codex, OpenClaw, and Hermes. Edge cases such as a missing API key, empty notebooks for new users, or upgrade prompts each have explicit fallback behaviour rather than silent failures. Output formatting rules force timestamps into YYYY-MM-DD, reading durations into hours and minutes, and book IDs into weread:// deep links.

The license is MIT and the project is made by the author of an earlier skill called Nuwa, which distilled a person's methodology into a skill. This one applies the same distillation idea to a WeChat reading account.