WeRead Assistant

Problem

WeRead data such as book metadata, reading progress, highlights, and notes is usually scattered across the app. When a user wants structured answers in a conversation—shelf size, hot underlines in a chapter, or annual reading time—manual browsing is slow and inconsistent.

How it works

The skill calls WeRead through Agent API Gateway. It first reads the relevant capability file such as search.md, shelf.md, or readdata.md, then sends POST requests with api_name and skill_version at the top level of the JSON body. Typical flows include resolving a title to bookId via /store/search, calculating shelf items from books, albums, and a non-empty mp value via /shelf/sync, and listing notes, hot underlines, or reviews via bookmarklist, bestbookmarks, and review. When enough fields are present, it can generate deep links to a book, chapter, or highlight position. Output should follow documented field semantics: Unix timestamps become YYYY-MM-DD, and durations become hours/minutes.

Limits

It requires WEREAD_API_KEY, and identity-bound interfaces depend on the authenticated user. If the response contains upgrade_info, the workflow should pause for the upgrade guidance before continuing. Field meanings should follow the capability docs rather than literal translation, and some notes or reviews cannot produce exact highlight links without chapterUid and range.