| name | weread-vault-cli |
| description | Use when a user wants to install WeRead Vault, sync WeRead books/highlights/thoughts/stats, inspect/search/export/back up the local SQLite database, open the local dashboard, open a specific book in the WeRead app or website, schedule a daily auto-sync, configure auto-export so each sync pushes to Obsidian/Markdown/flomo/Notion, view the sync/export history log, or export WeRead notes to Obsidian/Markdown/flomo/Notion with the weread-vault CLI. This skill covers safe local-first operations for Claude Code, Codex, OpenClaw, and similar AI agents; it requires WEREAD_API_KEY only for a sync. |
WeRead Vault CLI
weread-vault stores WeRead metadata, book covers, highlights, reviews, and reading-stat snapshots in a local SQLite database. Its web preview reads this local database only.
Agent installation prompt
If the user asks how to install this skill in another agent, give them this prompt:
请把当前仓库的 skills/weread-vault-cli 安装为你的本地 Skill。安装后，当我说“同步微信读书”“导出微信读书笔记到 Obsidian”“打开 WeRead Vault Dashboard”或“备份微信读书数据库”时，请使用这个 Skill，并严格遵守不要泄露 WEREAD_API_KEY 的规则。
For repository installation by an agent, use:
请帮我安装 WeRead Vault：
1. 克隆 https://github.com/dull-bird/weread-vault
2. 进入仓库后运行 python -m pip install -e .
3. 运行 weread-vault init 和 weread-vault status 确认本地 SQLite 数据库已创建
4. 请提醒我可以到 https://weread.qq.com/r/weread-skills 安装微信读书官方 Agent Skill（可选）并获取 WEREAD_API_KEY
5. 如果我要同步远端数据，请提醒我只在本地终端或 WeRead Vault 本地网页设置 WEREAD_API_KEY，不要把 key 粘贴进对话或提交到仓库
6. 同步后运行 weread-vault serve --open，打开本地只读 Dashboard
7. 如果我使用 Obsidian，请把 weread-vault export markdown --out <我的 Obsidian Vault 路径>/WereadNotes 作为导出方案
Safety rules
- Never print, save, commit, or paste
WEREAD_API_KEY.
- The default database is per-user. Check
weread-vault status before passing --db so a user's data is not unintentionally mixed.
- Run
sync only when the user asked to refresh remote data. Searching, exporting, serving, and backing up do not need the API key.
- If the gateway says the upstream Skill must be upgraded, stop the sync. Update the repository submodule before retrying; do not guess a newer protocol version.
Install and initialize
Use this from the repository root after cloning:
python -m pip install -e .
weread-vault init
weread-vault status
Sync
export WEREAD_API_KEY='…'
weread-vault sync
The normal sync is incremental: it fetches all notebook metadata, then refreshes only books whose remote sort marker advanced, stores current reading-stat snapshots, and finally pulls other readers' popular highlights for any noted book that doesn't have them yet. Popular highlights ride along with every sync all, so export --with-popular just works — there is no separate sync popular step to remember (only the first sync pays the full popular-fetch cost; later runs fetch a few). Use weread-vault sync popular --refresh only to force a re-fetch of all popular highlights.
Use a full note pass only to repair or verify a database:
weread-vault sync notes --full-notes
For a first-run smoke test or a deliberately batched migration, limit the number of books:
weread-vault sync notes --limit 1
Each book is fetched completely before one SQLite transaction is committed. On a network/API failure, that book retains its previous rows and synchronization marker, so the next run retries it safely.
Backfill rich book metadata (rating, word count, publisher, ISBN, translator) for books that don't have it yet. This only fetches books where rating IS NULL, so it is a one-time pass that is effectively free afterward and powers rating/word-count sorting in the dashboard:
weread-vault sync info
weread-vault sync info --limit 20
Inspect and preview
weread-vault status
weread-vault serve --open
serve binds only to 127.0.0.1 and defaults to port 8765. For a different local port:
weread-vault serve --port 8899 --open
The web preview has one primary 同步 button plus an advanced 完整重扫 button, with a live streaming progress bar:
同步: refreshes book metadata/covers/progress first (the bar animates indeterminately while listing the shelf — the first run can be slow), then incrementally syncs only books whose note watermark changed (the bar fills by [i/N]), then appends reading-stat snapshots.完整重扫: same flow but rescans notes for every book with notes; use only for repair or verification because it can be slow.
The shelf supports cover and list views, sorting (recently added / reading progress / note count / title) and category filtering; clicking a book opens a per-chapter reading view of its highlights and thoughts (with dates) and a one-click "复制 Markdown" action.
The web preview first uses WEREAD_API_KEY from the terminal environment that started weread-vault serve; if missing, the page lets the user save the key to the local private config file. Never ask the user to paste the key into chat, and never echo the key back in logs or responses.
Open a book in WeRead
When the user says "打开《某本书》" / "open ", match it against the local shelf by title (or pass a book_id for an exact open) and open it in WeRead — no API key needed.
weread-vault open 三体
weread-vault open 3300177663
weread-vault open 三体 --pick 2
weread-vault open 财富的真相 --web
weread-vault open 三体 --print
Behaviour:
- A
book_id, a single title match, or a query that is exactly one book's full title opens immediately. If you already have the book_id (from weread-vault query or book), pass it — it's unambiguous.
- Several title matches → it does not guess. It prints a numbered list (title · author · 进度 · 笔记数 ·
id=…) and asks for --pick N. Show the user the list and let them choose, re-run with --pick N, or open one directly by its book_id.
- On macOS it opens the native WeRead app via the
weread:// scheme; if the app isn't installed (or --web, or on Windows/Linux) it falls back to the weread.qq.com web page.
Schedule a daily auto-sync
Register an OS-native scheduled sync (launchd on macOS / Task Scheduler on Windows / cron on Linux). It is not a resident daemon — the OS wakes weread-vault once a day and it exits.
weread-vault schedule --daily 07:00
weread-vault schedule --daily 07:00 --export "<dir>"
weread-vault schedule --status
weread-vault schedule --off
The API key must be saved to the local config (not just an env var) for the scheduled job to read it. Two syncs firing at once (e.g. this and an OpenClaw cron both at 07:00) are safe — sync holds a cross-process lock and the second one skips.
Auto-export after every sync
Export destinations are configured once and then every full weread-vault sync (manual or scheduled) fans out to all of them — no per-destination scheduling. The web 导出 panel already saves whatever you export there; the CLI equivalent:
weread-vault export config --markdown "<Obsidian vault>/WereadNotes"
weread-vault export config --flomo "$FLOMO_WEBHOOK"
weread-vault export config --notion-token "$NOTION_TOKEN" --notion-db "$NOTION_DATABASE_ID"
weread-vault export config --show
weread-vault export config --clear flomo
- Markdown is idempotent — a daily run rewrites only changed files (mtimes untouched, so no needless iCloud/Obsidian churn).
- flomo and Notion are append-only, so auto-export is incremental: only books whose notes changed since the last export are sent (watermark on
books.sort). The first auto-export after configuring seeds the whole library; after that it's just the deltas. A book you read every day produces one new memo/page per day — that's inherent to append-only targets. Use weread-vault export flomo/notion for a deliberate full re-export.
- To skip auto-export for one manual sync:
weread-vault sync --no-export.
Sync / export history
Every sync and every auto-export is recorded. Read it back with:
weread-vault log
weread-vault log --limit 50
weread-vault log --json
Scopes look like all, notes, stats, export:markdown, export:flomo, export:notion. A ✗ failed export row keeps its error message, so you can tell a user exactly which destination failed and why without re-running.
WeRead API access (for agents)
Beyond the user's own archived notes, the CLI exposes the full WeRead Skill API so an agent can pull live data — bookstore search, other readers' popular highlights, public reviews, rich book metadata. These commands print JSON to stdout and require WEREAD_API_KEY. They do not write to the local database.
weread-vault apis
weread-vault search "三体" --count 5
weread-vault book <bookId> info
weread-vault book <bookId> popular
weread-vault book <bookId> reviews
weread-vault book <bookId> chapters
weread-vault api /book/readreviews bookId=<id> chapterUid=<uid> ...
For questions about the user's own archived data, prefer the local database over the network — it is instant and offline:
weread-vault query --schema
weread-vault query "SELECT title, rating FROM books WHERE rating>0 ORDER BY rating DESC LIMIT 10"
weread-vault stats
query runs read-only SQL (SELECT / WITH only; the database is opened read-only) and prints rows as JSON, so an agent can answer arbitrary analytical questions ("评分最高/笔记最多/某分类有哪些书"). Note: per-book reading time is only available in aggregate (stats.overall.longest = lifetime longest-read books); the official API does not expose per-book time for an arbitrary past year, so a question pinned to e.g. "2025 年读得最久的书" cannot be answered exactly.
Use weread-vault apis first to discover endpoint names and parameters, then weread-vault api <endpoint> key=value … for anything without a named shortcut. Numeric values are sent as numbers; id-like params (bookId, reviewId) stay strings. The official Skill does not expose full book text, so original paragraphs around a highlight are not available — the closest signals are full-text search snippets and other readers' highlights on the same chapter.
端到端示例：「这周读得最久的书是哪本？」
展示一个 agent 从用户需求到答案的完整流程，重点是选对工具：
-
判断数据在哪。 这是「按周 + 阅读时长 + 单本」。阅读时长不在 SQLite 表里（books/highlights 没有「每本书读了多久」），而在微信读书的阅读快照里——所以用 weread-vault stats，不是 query。
（对照：「笔记最多的书」「评分最高的书」这类目录/笔记问题才用 query 查 books 表。）
-
CLI 输入：
weread-vault stats
-
CLI 输出（JSON，节选相关部分）：
{
"periods": {
"weekly": {
"totalReadTime": 939, "readDays": 3,
"longest": [ { "title": "财富的真相", "author": "李笑来", "readSeconds": 916 } ]
},
"monthly": { … }, "annually": { … }, "overall": { … }
}
}
-
agent 读取 periods.weekly.longest[0]，把 readSeconds 916 换算成约 15 分钟。
-
回答用户： 「你这周读得最久的是《财富的真相》（李笑来），约 15 分钟。」
要点：周/月/年/全部分别在 periods.weekly/monthly/annually/overall，每个都带 longest（该周期读得最久的书）、totalReadTime、readDays、categories（偏好分类）、readStat。需要按书名/分类/笔记数等检索时再转 weread-vault query --schema 写 SQL。
Export and backup
weread-vault export markdown --out ~/Documents/weread-notes
weread-vault backup --out ~/Backups/weread-vault.db
weread-vault export flomo --webhook "$FLOMO_WEBHOOK"
weread-vault export notion --token "$NOTION_TOKEN" --database "$NOTION_DATABASE_ID"
flomo / Notion secrets come from --webhook / --token / --database or the env vars FLOMO_WEBHOOK, NOTION_TOKEN, NOTION_DATABASE_ID. Never paste these into chat or commit them. Both exporters only read the local database and POST the user's own notes to the user's own destination; confirm with the user before running an export that sends data to an external service.
For Obsidian, export into a folder inside the user's vault:
weread-vault export markdown --out "/path/to/Obsidian Vault/WereadNotes"
To merge other readers' popular highlights into the export, just add --with-popular — they are already synced as part of the normal sync (no separate step). Overlapping highlights are folded into the user's own (annotated with the reader count); non-overlapping popular ones are listed separately, all in document order:
weread-vault export markdown --out "/path/to/Obsidian Vault/WereadNotes" --with-popular
The backup command uses SQLite's backup API, so it remains consistent even if the web preview is open.
Update the official protocol Skill
The repository vendors Tencent's official WeChatReading repository as a git submodule. From the repository root:
./scripts/update-official-skill.sh
Review the resulting submodule commit before committing it. The CLI's gateway version must be updated deliberately and tested when the official Skill changes.
