Use this skill whenever the user mentions 微信读书, WeRead, 书架, 读书时间, 读书笔记, 划线, or wants to do anything with their personal reading data. This includes: book search (搜书/找书), shelf inspection, reading progress and time, notes and highlights export, book reviews, reading statistics (weekly/monthly/annual/

WeRead The local weread CLI is the only interface you need. It handles authentication, request shape, skill version , JSON parsing, upgrade checks, and normalized errors — writing ad hoc curl requests would mean reimplementing all of that and losing the normalized error layer. Agents should use json by default. Use human readable output only when the user explicitly wants terminal readable command output. For large outputs such as shelf, notebooks, bookmarks, or exports, write JSON or Markdown to a file first, then summarize the result in chat. Do not paste full large JSON payloads or full exported notes into the conversation. For notes export , use output <path by default. Only stream the export to stdout when the user explicitly asks to see the full export inline. First Decision Before any WeRead work, check auth and CLI health: If the weread command is missing, or if auth configured is false, read references/first use.md and guide the user through setup. Command Map When a supported API isn't covered by a first class command, use the raw escape hatch: When To Read References References are loaded on demand to keep startup context lean. Load them only when the task requires it: First time setup, missing auth, or API Key questions : read references/first use.md Shelf totals, public/private counts, audiobook or article collection handling : read references/domain rules.md Notes, highlights, bookmarks, personal ideas, public reviews, or exports : read references/domain rules.md Reading statistics, historical periods, cross year ranges, or time unit interpretation : read references/domain rules.md Deep links to books, chapters, highlights, or ideas : read references/domain rules.md Intent Routing Search Find a book, get a bookId , or user says 搜书 / 找书 : weread search "<keyword " scope book Generic 搜一下 or mixed intent: scope all Web fiction or 网文 : scope fiction Audiobooks, podcasts, 听书 , 有声书 , or 专辑 : scope audio Authors: scope author Full text search or 书里提到 : scope fulltext Book lists: scope list Official accounts: scope mp Articles: scope article Book Details If the user gives a title rather than an ID, prefer weread json book resolve "<title " limit 5 and use items[].bookId . Metadata: book info Chapter UIDs (needed for notes/highlights by chapter): book chapters Reading progress: book progress — progress is an integer percent; 1 means 1%, only 100 means finished Shelf Use weread json shelf list . For recent reading, use weread json shelf recent limit 10 . For anything nuanced (public/private split, audiobook handling), read references/domain rules.md . Reading Statistics Use weread json readdata summary mode monthly for common summaries, or weread json readdata detail with mode weekly , monthly , annually , or overall . Time fields in structured output are seconds. For historical or cross period calculations, read references/domain rules.md . Notes and Highlights Overview across all books: weread json notes notebooks Books with the most personal notes/highlights: weread json notes top limit 20 Single book export: weread notes export <bookId format markdown output <path ; choose a local Markdown path and summarize counts/content instead of pasting the whole file If manually combining data, use both weread json notes bookmarks <bookId and weread json notes mine <bookId For counting rules, export limits, or popular highlight queries, read references/domain rules.md Reviews Public reviews: weread json reviews list <bookId with type 0 (all) through type 4 . Single review: weread json reviews single <reviewId . Recommendations Personalized: weread json discover recommend Similar books: weread json discover similar <bookId Friends' reading activity: weread json discover friends ; continue with nextMaxIdx as max idx and reuse synckey for refreshes Pagination Stay shallow by default — only paginate further when the user explicitly asks for a complete export, a ranking, or a total that requires all pages. Use limit for display size and all where supported. For manual pagination, use json and pass the native cursor from the previous JSON result: Search: hasMore == 1 → pass last item searchIdx as max idx Notebooks: hasMore == 1 → pass last books[].sort as last sort Reviews: pass last review idx as max idx and returned synckey as synckey Similar: pass last item idx as max idx and booksimilar.sessionId as session id Friends' activity: pass nextMaxIdx as max idx ; reuse the returned synckey when refreshing the feed Error Handling The CLI normalizes errors to JSON: missing auth : read references/first use.md and help configure the key upgrade required : stop immediately and follow upgrade info.message ; don't continue the original task until upgraded upstream timeout , network error : these are retryable; the CLI already retries transient failures before returning the error api error , http error , invalid json : report the failure; retry only when repeating is safe User Facing Output Default to json ; use compact when a command returns too much metadata for the task Convert Unix timestamps to readable dates Convert seconds to hours and minutes Use numbered lists for search results, shelf entries, notes, reviews, and recommendations Include WeRead deep links when useful; format rules are in references/domain rules.md