Save To Academic Notion

./save_paper.py https://openreview.net/forum? id =abc123

Manual mode (when auto-extraction fails)

./save_paper.py --manual
--title "Paper Title"
--url "https://example.com/paper"
--authors "Alice; Bob; Charlie"
--abstract "This paper presents..."

JSON output (for automation)

./save_paper.py 2501.12345 --json What it does: Auto-fetches metadata from arXiv API (title, authors, abstract, date) For non-arXiv URLs: attempts web scraping fallback Checks for duplicates (via Link property) Creates page with proper properties and abstract as quote block Uses "Authors txt" field (semicolon-separated text, not multi-select) Returns Notion page URL Output: Fetching metadata for 2501.12345... Creating Notion page for: Paper Title... Authors: Author One, Author Two, ... ✓ Created: https://www.notion.so/... If paper already exists: ✓ Paper already exists: https://www.notion.so/... JSON output: { "ok" : true , "action" : "created" , // or "exists" "page_id" : "..." , "page_url" : "https://www.notion.so/..." , "source" : "arxiv" // or "web_fetch" or "manual" } Error handling (non-arXiv URLs): If web scraping fails, returns structured error with suggestion: { "ok" : false , "error" : "Could not extract metadata from URL" , "suggestion" : "Use browser tools to extract metadata, then use --manual mode" } Non-arXiv Paper Fallback arXiv papers: Metadata auto-fetched from arXiv API (always works) Non-arXiv URLs: Attempts web scraping via openclaw web_fetch Success: Creates page with extracted metadata (may be incomplete) Failure: Returns error with suggestion to use --manual mode Manual mode workflow (for agent): Try auto-save first: ./save_paper.py <url> --json If ok: false : Use browser tools to extract title, authors, abstract Save with --manual mode: ./save_paper.py --manual
--title "Paper Title"
--url "https://..."
--authors "Author1; Author2; Author3"
--abstract "..." \

date "2026-03-10" Supported URLs: arXiv (always works): https://arxiv.org/abs/... OpenReview, ACL Anthology, bioRxiv, journal sites (web scraping fallback) Any URL (manual mode as last resort) Targets Academic paper database: database_id: d3e3be7f-c96a-45de-8e7d-3a78298f9ccd data_source_id (query): 73e9f7f8-c667-4279-a62f-2c16c1885d0f Hard rules Only write when Selina explicitly asks to save/add/log. Deduplicate : do not create a new entry if it already exists. Prefer using the database property Link as the canonical key. Authors field: Uses "Authors txt" (rich text, semicolon-separated), not "Author" (multi-select) Dedup workflow Normalize the URL (trim, remove tracking params when safe). Query the data source filtering: Link equals the normalized URL If arXiv: also try canonical https://arxiv.org/abs/<id> if the user provided a PDF link. If found: return the existing page URL; optionally offer to update metadata if missing. Create workflow If not found: Create a new page in the database: POST /v1/pages with parent: {"database_id": <database_id>} Set properties: Name (title): best available title (from arXiv/metadata) Link (url): normalized URL Optionally: Publisher , Type , Publishing/Release Date , Discipline , Topics , Status Populate authors: Updated (Mar 10, 2026): Uses Authors txt property (rich text, semicolon-separated) Format: "Author1; Author2; Author3" No longer uses: Author property (multi-select) — causes schema overflow with 1000+ unique authors Example: properties["Authors txt"] = {"rich_text": [{"text": {"content": "Alice; Bob; Charlie"}}]} Populate abstract: There is no dedicated "Abstract" property. Put the abstract into the page body as blocks (preferred): quote (common in Selina's pages) OR heading_2 : "Abstract" paragraph (or quote content): abstract text Optionally use TLDR for a short summary (not the full abstract), if Selina wants that later. Metadata extraction If URL is arXiv Accept input forms: https://arxiv.org/abs/<id> https://arxiv.org/pdf/<id>.pdf bare <id> like 2501.01234 Fetch metadata from one of: arXiv API: http://export.arxiv.org/api/query?id_list=<id> (Atom XML) Fallback: scrape the arXiv abs page. Extract: title authors abstract published date (optional) If URL is not arXiv Use web_fetch to get the page and extract: title (best guess) author list if present abstract/summary if present If not reliably available, create entry with just Name + Link and leave placeholders. Suggested property mapping (when confident) Publisher : set to bioRxiv , arXiv , NeurIPS , etc. only if clearly indicated. Type : default to Academic Journal for papers; Blog Post for posts. Status : default to Ready to Start . Discipline : add tags like ML , MechInterp , Geometry , ScienceofDL if strongly implied. UX Before writing: Confirm the normalized link + the guessed title. After writing: Return the created/updated Notion page URL. Briefly list what fields were populated.