Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Query open citation data and reference networks via OpenCitations
.claude/skills/brycewang-stanford-opencitations-api/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 18% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 35% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 42% | 0% |
OpenCitations is an independent infrastructure organization dedicated to open scholarship and the publication of open bibliographic and citation data. Its main product, the OpenCitations Index of Crossref open DOI-to-DOI citations (COCI), contains over 1.6 billion citation relationships harvested from CrossRef metadata. This makes it the largest fully open citation dataset in the world.
The OpenCitations API allows researchers to programmatically access citation and reference data for any DOI in the index. This is valuable for citation network analysis, bibliometric research, impact assessment, and building literature discovery tools. Unlike proprietary citation databases (Web of Science, Scopus), OpenCitations data is fully open under a CC0 public domain dedication.
The API is free, requires no authentication, and has no published rate limits. It returns data in JSON or CSV format, making it easy to integrate into data analysis pipelines.
No authentication required. The OpenCitations API is fully open and free. No API key, registration, or email is needed. There are no published rate limits, but users should implement reasonable request pacing for large-scale queries. For bulk data access, download the complete COCI dataset from https://opencitations.net/download.
GET https://api.opencitations.net/index/v1/citations/{doi}| Param | Type | Required | Description | |-------|------|----------|-------------| | doi | string | Yes | The DOI of the cited paper (URL path parameter) | | format | string | No | Response format header: application/json or text/csv |
bash curl "https://api.opencitations.net/index/v1/citations/10.1038/nature12373"
oci: OpenCitations Identifier for the citation linkciting: DOI of the citing papercited: DOI of the cited paper (the input DOI)creation: date the citation was first recordedtimespan: time between publication of citing and cited papersjournal_sc: whether citing and cited are in the same journal (self-citation indicator)author_sc: whether any author appears in both papers (author self-citation indicator)GET https://api.opencitations.net/index/v1/references/{doi}| Param | Type | Required | Description | |-------|------|----------|-------------| | doi | string | Yes | The DOI of the paper whose references to retrieve |
bash curl "https://api.opencitations.net/index/v1/references/10.1038/nature12373"
citing is the input DOI and cited are the referenced papers.GET https://api.opencitations.net/index/v1/metadata/{doi}| Param | Type | Required | Description | |-------|------|----------|-------------| | doi | string | Yes | The DOI to look up (supports multiple DOIs separated by __) |
bash curl "https://api.opencitations.net/index/v1/metadata/10.1038/nature12373"
title, author, year, source_title (journal), volume, issue, page, doi, citation_count, and reference.GET https://api.opencitations.net/index/v1/citation-count/{doi}| Param | Type | Required | Description | |-------|------|----------|-------------| | doi | string | Yes | The DOI to check |
bash curl "https://api.opencitations.net/index/v1/citation-count/10.1038/nature12373"
count field indicating the number of citations in the index.No published rate limits. OpenCitations does not enforce strict API quotas. The service runs on academic infrastructure, so users should be respectful. Best practices include pacing requests to 1-5 per second for sustained queries, caching results, and using the bulk dataset download for large-scale network analyses. The API may return HTTP 503 under heavy load.
Map the citation relationships around a seminal paper:
bash# Get all papers citing the target paper curl -s "https://api.opencitations.net/index/v1/citations/10.1145/3292500.3330672" | jq '.[].citing' # Get all papers referenced by the target paper curl -s "https://api.opencitations.net/index/v1/references/10.1145/3292500.3330672" | jq '.[].cited'
Filter out self-citations when computing impact metrics:
bashcurl -s "https://api.opencitations.net/index/v1/citations/10.1038/nature12373" | jq '[.[] | select(.author_sc == "no")] | length'
Retrieve citation counts for multiple papers in a batch:
bash# Multiple DOIs separated by double underscore curl -s "https://api.opencitations.net/index/v1/metadata/10.1038/nature12373__10.1126/science.aaa8685__10.1016/j.cell.2015.05.002" | jq '.[] | {doi: .doi, title: .title, citations: .citation_count}'
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 17,331 | 13,014 | -25% | 1 | 1 | 0% | 2,843 | 3,366 | +18% | 0 | 0 | — |
case-02 | fail→pass | 29,026 | 6,854 | -76% | 1 | 1 | 0% | 5,132 | 2,680 | -48% | 0 | 0 | — |
case-03 | fail→pass | 16,537 | 6,978 | -58% | 1 | 1 | 0% | 2,756 | 2,602 | -6% | 0 | 0 | — |
case-04 | fail→pass | 11,218 | 5,066 | -55% | 1 | 1 | 0% | 1,788 | 2,422 | +35% | 0 | 0 | — |
case-05 | fail→pass | 8,690 | 4,717 | -46% | 1 | 1 | 0% | 1,395 | 1,983 | +42% | 0 | 0 | — |
case-06 | fail→pass | 14,085 | 9,251 | -34% | 1 | 1 | 0% | 2,571 | 2,963 | +15% | 0 | 0 | — |
case-07 | pass→pass | 12,596 | 16,424 | +30% | 1 | 1 | 0% | 1,980 | 3,927 | +98% | 0 | 0 | — |
case-08 | pass→pass | 3,961 | 3,802 | -4% | 1 | 1 | 0% | 743 | 2,280 | +207% | 0 | 0 | — |
case-09 | pass→pass | 7,689 | 2,603 | -66% | 1 | 1 | 0% | 1,066 | 2,013 | +89% | 0 | 0 | — |
case-10 | pass→pass | 9,409 | 1,979 | -79% | 1 | 1 | 0% | 1,368 | 1,802 | +32% | 0 | 0 | — |
case-11 | pass→pass | 4,024 | 1,564 | -61% | 1 | 1 | 0% | 701 | 1,775 | +153% | 0 | 0 | — |
case-12 | pass→pass | 7,459 | 5,133 | -31% | 1 | 1 | 0% | 1,120 | 2,197 | +96% | 0 | 0 | — |
case-13 | pass→pass | 14,665 | 2,854 | -81% | 1 | 1 | 0% | 2,036 | 1,949 | -4% | 0 | 0 | — |
case-14 | pass→pass | 4,760 | 4,311 | -9% | 1 | 1 | 0% | 847 | 2,080 | +146% | 0 | 0 | — |
case-15 | pass→pass | 4,875 | 3,108 | -36% | 1 | 1 | 0% | 780 | 1,973 | +153% | 0 | 0 | — |
case-16 | pass→pass | 6,135 | 1,894 | -69% | 1 | 1 | 0% | 981 | 1,767 | +80% | 0 | 0 | — |
case-17 | pass→pass | 4,313 | 2,155 | -50% | 1 | 1 | 0% | 674 | 1,757 | +161% | 0 | 0 | — |
case-18 | pass→pass | 3,559 | 1,887 | -47% | 1 | 1 | 0% | 567 | 1,752 | +209% | 0 | 0 | — |
case-19 | pass→pass | 8,996 | 6,953 | -23% | 1 | 1 | 0% | 1,750 | 2,655 | +52% | 0 | 0 | — |
case-20 | fail→pass | 9,211 | 3,769 | -59% | 1 | 1 | 0% | 1,809 | 2,133 | +18% | 0 | 0 | — |
case-21 | pass→pass | 12,411 | 6,227 | -50% | 1 | 1 | 0% | 2,004 | 2,946 | +47% | 0 | 0 | — |
case-22 | pass→pass | 18,241 | 15,672 | -14% | 1 | 1 | 0% | 2,674 | 4,131 | +54% | 0 | 0 | — |
case-23 | pass→pass | 13,285 | 16,823 | +27% | 1 | 1 | 0% | 2,538 | 4,128 | +63% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +30 percentage points is the difference between those two pass rates over the 23 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.