| name | readwise |
| description | Use this skill to work with the Readwise Original API for creating, organizing, and reviewing highlights and books. |
| allowed-tools | Bash(uv run *) |
Readwise Original Skill
Use this skill to automate work with the Readwise "Original" API that powers highlight exports and metadata management.
Quick start
- Generate a token from https://readwise.io/access_token and store it in
READWISE_TOKEN.
- Use
uv run --project ${CLAUDE_PLUGIN_ROOT} python ${CLAUDE_PLUGIN_ROOT}/skills/readwise/scripts/readwise_client.py ... instead of calling the API directly; it handles retries, pagination, rate-limit surfacing, --dry-run, and tagging rules.
- Keep requests below the documented rate limit (currently 60 req/min). Batch operations and pause between pages when processing large libraries.
Example workflows
- "Summarize my daily review and save highlights to a note" — uses
highlights review to fetch today's highlights, then formats them
- "Find all highlights from Meditations and export as markdown" — uses
books to find the book ID, then highlights list to fetch
- "Save this quote to my Readwise" — uses
highlight create with the quote text
CLI commands
Global options --dry-run and --raw can appear anywhere in the command.
uv run --project ${CLAUDE_PLUGIN_ROOT} python ${CLAUDE_PLUGIN_ROOT}/skills/readwise/scripts/readwise_client.py [--dry-run] [--raw] highlight create --text ... [--book-id ID] [--title ...] [--author ...] [--category ...] [--image-url ...] [--generated] [--tags t1,t2]
Creates a highlight or batch (--bulk-file ndjson). Use --book-id to target an existing book (the CLI resolves it to title/author/category for the API). Always specify --category when creating new sources—see "Category is critical" below. Use --image-url for cover images (recommended for books and podcasts, not needed for articles or tweets). If --generated is set, .generated is appended to tags and location_type defaults to none.... [--dry-run] highlight update <id> [--title --note --tags --generated] – partial updates.... highlight show <id> – fetch highlight details.... highlight delete <id> [--yes] – delete a highlight, --yes skips confirmation.... highlights list [--book-id 123] [--tag focus] [--category books] [--updated-after 2026-02-01] [--limit 50] – cursor-based listing. Dates are ISO format (e.g. 2026-02-01 or 2026-02-01T10:30:00Z).... highlights review – fetches today's daily review highlights (no parameters; returns the curated spaced-repetition set).... books [--title ...] [--author ...] [--limit N] – search books by title/author (case-insensitive substring match).... book <id> – fetch a single book's metadata.... auth validate – confirms your token works by hitting the /api/v2/auth/ endpoint.
Default output is human-readable markdown with only key fields. Use --raw to get full JSON with all fields.
Disambiguating book matches
books --title performs a case-insensitive substring match and often returns multiple books with the same title. Readwise creates separate book entries per source (Kindle, API, supplemental, etc.), so duplicates are common. When multiple results come back, review them all and use judgement to determine whether they are duplicates of the same work or entirely separate books:
- Duplicates (same author, different sources): Readwise creates one entry per import source. A Kindle book and an API-created entry for the same work will share a title/author. Among duplicates, the entry with the most
num_highlights is usually the primary one.
- Distinct books: Different authors or categories mean genuinely different works. Treat these as separate and ask the user which one they mean if the intent is ambiguous.
Listing highlights for a book by name
- Find the book ID:
... books --title "book name"
- If multiple results, review all entries to determine which are duplicates vs distinct books. For duplicates, query the entry with the most highlights. For distinct books, clarify with the user.
- List its highlights:
... highlights list --book-id <id>
Adding a highlight to an existing book
- Search for the book:
... books --title "when breath becomes air"
- Note the book
id from the output.
- Create the highlight:
... highlight create --text "the verbatim quote" --book-id <id>
The CLI looks up the book's title/author/category and injects them into the payload so the Readwise API matches the highlight to the correct book.
Only add --generated if the text is a summary or paraphrase, not a direct quote.
How Readwise matches sources
Readwise groups highlights into "books" (sources) using (title, author, source_url). The API de-duplicates highlights by matching all four fields: (title, author, source_url, text)—including nulls.
Critical: All highlights from the same content—whether verbatim quotes or generated summaries—MUST use identical (title, author, source_url) values to be grouped together. Inconsistent attributes will scatter highlights across separate book entries.
- If you omit
--title, highlights go into a generic "Quotes" book
- If you omit
--author, it stays blank (or uses the URL domain if source_url is set)
- Use
--book-id when adding to an existing book—the CLI fetches and injects the correct title/author/source_url
Category is critical
Always specify --category explicitly. The API defaults to articles if source_url is present, otherwise books—but these defaults often miscategorize content. Valid categories: books, articles, tweets, podcasts.
| Content type | Category |
|---|
| Physical/ebook | books |
| Blog post, web article | articles |
| Tweet/X post | tweets |
| Podcast episode | podcasts |
| YouTube video | podcasts |
YouTube videos should use podcasts, not articles. Readwise treats video content as podcast-like (time-based, spoken word). Using articles will cause display and organizational issues.
Mismatched categories create duplicate book entries and break the user's library organization. When in doubt, ask the user or check existing entries with books --title.
Agent interaction rules
- Always quote highlights in full: Never reference a highlight in the abstract. When displaying or communicating about a highlight to the user, always include the full highlight text in its entirety as a quoted markdown block (using
>).
- Never save without confirmation: Never save a highlight (create or update) without explicit user confirmation first. Draft the highlight, show it quoted, and wait for approval before calling the API.
Data entry guidance
- Generated snippets: Use
--generated ONLY for highlights that are NOT verbatim quotes from the source content—e.g., AI-generated summaries, paraphrases, or synthesized insights. Do NOT use it for actual quotes or excerpts copied directly from the text. The CLI injects .generated for discoverability and sets location_type=none since generated content has no source location. Generated highlights should still use the SAME (title, author, source_url) as verbatim highlights from the same source.
- Highlight text: supply
--text, --text-file, or pipe content via stdin. The CLI refuses to guess. Bulk imports accept NDJSON rows with text, title, and tags.
- Location best practices: omit
--location unless you can provide an absolute, client-agnostic reference (page number, character offset). When --location is omitted the payload leaves location blank so Readwise can reconcile it later. Valid --location-type values: page (books), time_offset (podcasts/videos), order (articles). For generated quotes the CLI defaults to location_type=none.
- Dry runs: add
--dry-run to print the exact payload without calling the API. Dry-run output always shows the full payload (no field filtering).
Scripts
${CLAUDE_PLUGIN_ROOT}/skills/readwise/scripts/readwise_client.py: full-featured CLI covering highlight create/read/update, daily review, and books list/detail. Commands surface remaining rate-limit headers when provided so agents can throttle work.- Shared helpers live in
${CLAUDE_PLUGIN_ROOT}/readwise_common/ (auth, HTTP retries, tag/location utilities); import from there when extending functionality to keep behavior consistent.
Testing & validation
- Use the dry-run flag in the client to print payloads instead of sending them when iterating on workflows.
- Run
uv run --project ${CLAUDE_PLUGIN_ROOT} python -m compileall ${CLAUDE_PLUGIN_ROOT}/skills/readwise ${CLAUDE_PLUGIN_ROOT}/readwise_common before shipping changes to catch syntax issues.
- For live smoke tests, set
READWISE_TOKEN and exercise --dry-run highlight create, highlights list, and highlights review against a small limit to verify pagination + rate-limit logging.
