Imagine asking your AI assistant, “What did I highlight in The Value?” and getting back a precise timeline of your reading marathon from six years ago. That’s exactly what happened when I installed WeRead’s newly released official AI skill. It connects to your reading account and lets you query six core data sets: bookshelf, book search, reading stats, book details, notes and highlights, and recommendations. The setup is almost trivial—copy a single command, paste it into Claude Code, grab an API key from WeChat Reading, and you’re done in under a minute. No config files, no headaches. In the domestic digital reading space, this level of API integration is unprecedented.
But after an afternoon of fascinated tinkering, I hit a wall. I asked the skill to recommend product management books. It returned a well-structured list of 14 titles, grouped from beginner to advanced. Sounds great—until I realized three of those books were already sitting in my bookshelf with hundreds of notes. Yu Jun’s Product Methodology? I had 68 highlights. Behind the Product? On my shelf for two years. The AI had simply performed a keyword search and returned generic results. It never once checked my reading history, my existing highlights, or my bookshelf. Every recommendation was essentially blind. It was a natural-language wrapper around a basic search API, not a personal reading advisor.
This is the core limitation of raw API skills: the six capabilities are powerful individually, but they lack intelligent orchestration. The skill can pull your stats, search books, and list your notes, but it cannot cross-reference them to understand who you are as a reader. It treats each user as a blank slate, ignoring the invaluable data of what you’ve already read, skipped, or deeply engaged with. In contrast, platforms like Amazon’s Kindle have long used collaborative filtering and purchase history to recommend next reads, but they rarely allow external AI to access that data. WeRead’s move is groundbreaking in its openness, but it stops short of the real value: context-aware personalization.
To bridge this gap, I built an open-source project called huashu-weread. It wraps the official WeRead skill with a “reading consultant” workflow layer. Instead of one-size-fits-all APIs, it offers four specialized workflows: advisor (recommends the next book by cross-referencing your shelves and notes to find gaps), path (designs a tailored reading ladder for a topic, from basics to frontiers, based on your level), alchemy (organizes your scattered highlights into structured summaries), and review (generates ready-to-post reading reflections for social media). The key insight is simple: your bookshelf shows what you’ve collected, but your notes reveal what you actually absorbed. Cross-referencing these two data sources is the difference between a search engine and a personal librarian.
Still, the official skill’s release marks a significant signal. It demonstrates that Chinese reading platforms are finally opening their user data to third-party AI, something even global giants like Kobo or Apple Books have not fully done. The next step is for WeRead to natively embed these reasoning workflows into their skill itself, rather than leaving users to hack their own solutions. If they can combine their rich reading data with a recommendation engine that truly understands your cognitive map, WeRead could redefine what it means to have an “AI reading partner.” Until then, the skill remains a powerful foundation—but it’s a foundation, not a finished building. The real intelligence still lies in what we build on top of it.
The hardest part of recommending books isn’t knowing what’s out there—it’s knowing what you’ve already been through. WeRead’s skill gives us the raw materials; the craft lies in weaving them into a personal narrative of learning. That’s the part AI has yet to master, but the door is now open.