Biorxiv Search

bioRxiv Search

Search bioRxiv through its official API for recent-preprint discovery, date-range scans, DOI lookups, author shortlisting, and local keyword filtering over title, abstract, and author metadata.

Instructions

1. Prefer this skill when the request is about bioRxiv-native preprints, recent biology submissions, or preprint metadata that may lag in PubMed, PMC, or Crossref.
2. Use the bundled CLI:
   • In this repository: skills/biorxiv-search/scripts/search
   • After installation: ~/.agents/skills/biorxiv-search/scripts/search
3. The official bioRxiv API does not provide a general server-side keyword search endpoint.
   • Use the CLI to fetch metadata from a bounded recent window or explicit date range, then filter locally.
4. When keywords are provided, search title, abstract, and authors by default.
   • If the user wants abstract-only matching, pass --fields abstract.
5. Keep the search window bounded.
   • Use --days N for recent scans or --start-date YYYY-MM-DD --end-date YYYY-MM-DD for explicit intervals.
   • If you omit the interval, the CLI defaults to the most recent 30 days.
   • The CLI converts --days N into an explicit date range before calling the API so pagination stays predictable.
6. Use --category <name> when the topic should stay narrow.
   • The API accepts the bioRxiv category as a query parameter such as cell_biology, genomics, or neuroscience.
7. Use --author for author-specific requests.
   • By default, consider both the supplied full-name form and an abbreviated-first-name form, for example --author "Peter Nugent" and --author "P. Nugent".
   • Do not silently merge these in the final answer. Report full-name matches and abbreviated-first-name matches in separate groups because initials can be ambiguous.
   • The CLI normalizes both given-name-first and API-style Surname, F. M. forms. Prefer separate passes or a local partition of returned records by the literal authors text when you need clean buckets.
8. The API currently returns up to 30 records per page.
   • The API orders an interval from oldest to newest. The CLI reads count and total, starts at the newest page, and scans backward.
   • Increase --scan-limit when a broad query needs more of the interval inspected.
9. By default, the CLI collapses multiple versions of the same preprint and keeps the latest version for each DOI.
   • Use --all-versions only when version-by-version output matters.
10. Treat the API output as discovery metadata.

• If exact citation details or the latest abstract-page presentation matter, verify the shortlisted candidates on bioRxiv or the DOI landing page before finalizing the answer.

11. If the user wants peer-reviewed biomedical literature or PMC full text rather than bioRxiv preprints, use polars-dovmed instead.
12. The CLI retries HTTP 429 and transient 5xx responses with bounded exponential backoff. Output records the retry policy, the exact latest-version rule, and separate match groups for every requested author form; initial-only groups remain explicitly ambiguous.

Quick Reference

Input Requirements

• Python 3
• One of:
  • a keyword query
  • a bioRxiv DOI via --doi
  • a request for recent/date-bounded preprints with no keyword query
• Interval, category, author, field, and version flags are listed in Quick Reference; --phrase treats the whole query as one phrase.
• If the user asks for very old or very broad searches, widen the date range deliberately and be explicit that recall depends on the chosen interval and --scan-limit.

Search Semantics

• The official bioRxiv API supports:
  • recent-post windows such as 30d
  • explicit date ranges
  • DOI lookup
  • subject-category filtering
• The API does not support a general server-side keyword query for title or abstract.
  • The CLI performs local filtering after fetching metadata.
• For predictable paging, the CLI implements --days N as an explicit UTC date range instead of relying on the API's relative-date shorthand.
• Interval searches start from the newest cursor and scan backward, so a bounded --scan-limit covers recent records first.
• Plain multi-word queries are local AND queries.
  • single cell atlas means all three terms must appear somewhere in the selected search fields.
• OR must be written explicitly to broaden synonyms or alternate phrasings.
  • "organoid OR spheroid"
  • "CRISPR OR Cas9"
• Quoted phrases are preserved when possible.
  • "\"single cell\" atlas" keeps single cell as one phrase and also requires atlas.
• --fields abstract restricts keyword filtering to abstracts only.
  • This is the flag to use when the user explicitly cares about abstract matches.
• Author filters can fragment across name variants.
  • For person-specific searches, check the full-name form and abbreviated-first-name form separately and keep those buckets separate in the final answer.

Output

• JSON with:
  • request metadata (query, query_groups, interval, category, author filters, search fields)
  • API metadata (pages_fetched, records_scanned, total_available, request_urls)
  • warnings about defaulted windows, scan-limit truncation, or API limitations
  • normalized result records with:
    • doi
    • title
    • authors
    • date
    • version
    • category
    • abstract
    • published
    • doi_url
    • biorxiv_url
    • matched_in

Quality Gates

• The request uses a bounded recent window or explicit date range
• The chosen --scan-limit is large enough for the query breadth
• The selected search fields match the user request, especially when abstract matching matters
• Author-specific requests use one or more reasonable name variants
• The final answer keeps abbreviated-name matches separate and labels them as potentially ambiguous
• The answer does not overstate recall for a broad historical search
• Final candidate metadata is verified on bioRxiv when exact citation/version details matter
• Output states the latest-version policy, keeps requested author forms in separate match groups, and uses bounded retry/backoff for 429 and transient 5xx responses.

Examples

Example 1: Recent keyword scan over title + abstract

skills/biorxiv-search/scripts/search "single cell atlas" 10 --days 30

Example 2: Broaden with OR

skills/biorxiv-search/scripts/search 'organoid OR spheroid' 15 \
  --days 90 \
  --category developmental_biology

Example 3: Abstract-only keyword filtering

skills/biorxiv-search/scripts/search "CRISPR screen" 10 \
  --days 60 \
  --fields abstract

Example 4: Author-specific search with separate variant reporting

skills/biorxiv-search/scripts/search "supernova" 20 \
  --days 365 \
  --author "Peter Nugent" \
  --author "P. Nugent"

Example 5: DOI lookup

skills/biorxiv-search/scripts/search --doi 10.1101/682021

Troubleshooting

Issue: Results are too broad
Solution: Narrow the interval, add --category, restrict with --fields, or replace a loose query with a phrase or explicit OR terms.

Issue: Results are too sparse
Solution: Increase --days or widen the date range, raise --scan-limit, and add alternate query terms with explicit OR.

Issue: Need abstract matches, not title matches
Solution: Use --fields abstract.

Issue: Author search looks incomplete
Solution: Repeat --author with explicit variants such as "Peter Nugent" and "P. Nugent". If a middle initial is known, add that too, for example "Peter E. Nugent" and "P. E. Nugent". Keep these result sets separate in the final answer because abbreviated forms can be ambiguous.

Issue: The API returns multiple versions of the same preprint
Solution: Keep the default deduped output, or pass --all-versions if version-level output matters.

Issue: Broad historical search may be missing expected hits
Solution: This usually means the interval or --scan-limit was too narrow. Widen them deliberately and say so in the final answer.

Issue: Need peer-reviewed literature rather than preprints
Solution: Use polars-dovmed or another peer-reviewed-literature workflow instead of bioRxiv metadata search.

Related Skills

• /crossref-lookup — resolve citation metadata from bioRxiv DOIs
• /polars-dovmed — switch here if the query wants peer-reviewed PMC full text instead of preprints