Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Discover open access research outputs via the SHARE notification API
.claude/skills/brycewang-stanford-share-research-api/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 35% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 42% | 0% |
SHARE (SHared Access Research Ecosystem) aggregates metadata from 200+ research repositories, preprint servers, and publishers into a unified search API. Operated by the Center for Open Science, it tracks research outputs as they move through the scholarly communication cycle — from preprint to publication. Free, no authentication for search.
https://share.osf.io/api/v2bash# Text search across all sources curl "https://share.osf.io/api/v2/search/creativeworks/?q=climate+change&page[size]=20" # Filter by type curl "https://share.osf.io/api/v2/search/creativeworks/?q=neural+networks&filter[type]=preprint" # Filter by source curl "https://share.osf.io/api/v2/search/creativeworks/?q=genomics&filter[sources]=PubMed+Central" # Filter by date curl "https://share.osf.io/api/v2/search/creativeworks/?q=COVID-19&filter[date][gte]=2024-01-01" # Filter by tag/subject curl "https://share.osf.io/api/v2/search/creativeworks/?q=machine+learning&filter[tags]=deep+learning"
| Parameter | Description | Example | |-----------|-------------|---------| | q | Search query | q=CRISPR | | filter[type] | Output type | preprint, article, dataset, thesis | | filter[sources] | Source repository | PubMed Central, arXiv, Zenodo | | filter[date][gte] | From date | 2024-01-01 | | filter[date][lte] | Until date | 2026-12-31 | | filter[tags] | Tag filter | open+data | | page[size] | Results per page | page[size]=50 | | sort | Sort order | -date_updated |
| Source | Type | |--------|------| | arXiv | Preprints | | PubMed Central | Biomedical articles | | Zenodo | Multi-discipline repository | | Figshare | Data/figures | | SSRN | Social science preprints | | DataCite | Research data | | Institutional repositories | Various |
pythonimport requests BASE_URL = "https://share.osf.io/api/v2" def search_share(query: str, output_type: str = None, source: str = None, from_date: str = None, page_size: int = 20) -> list: """Search SHARE for research outputs.""" params = {"q": query, "page[size]": page_size} if output_type: params["filter[type]"] = output_type if source: params["filter[sources]"] = source if from_date: params["filter[date][gte]"] = from_date resp = requests.get( f"{BASE_URL}/search/creativeworks/", params=params, ) resp.raise_for_status() data = resp.json() results = [] for item in data.get("data", []): attrs = item.get("attributes", {}) results.append({ "title": attrs.get("title"), "description": (attrs.get("description") or "")[:300], "type": attrs.get("type"), "date": attrs.get("date_updated", "")[:10], "sources": attrs.get("sources", []), "tags": attrs.get("tags", []), "identifiers": attrs.get("identifiers", []), }) return results # Example: find recent preprints on a topic preprints = search_share( "transformer architecture", output_type="preprint", from_date="2024-01-01", ) for p in preprints[:5]: print(f"[{p['date']}] {p['title']}") print(f" Type: {p['type']} | Sources: {', '.join(p['sources'][:3])}")
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 13,388 | 8,607 | -36% | 1 | 1 | 0% | 1,947 | 2,226 | +14% | 0 | 0 | — |
case-02 | fail→pass | 15,880 | 8,714 | -45% | 1 | 1 | 0% | 2,495 | 2,710 | +9% | 0 | 0 | — |
case-03 | fail→fail | 18,355 | 9,492 | -48% | 1 | 1 | 0% | 3,665 | 2,741 | -25% | 0 | 0 | — |
case-09 | fail→pass | 8,198 | 3,494 | -57% | 1 | 1 | 0% | 1,213 | 1,642 | +35% | 0 | 0 | — |
case-04 | pass→pass | 7,942 | 6,388 | -20% | 1 | 1 | 0% | 1,242 | 2,081 | +68% | 0 | 0 | — |
case-05 | pass→pass | 12,712 | 11,761 | -7% | 1 | 1 | 0% | 2,647 | 3,722 | +41% | 0 | 0 | — |
case-06 | pass→pass | 13,213 | 8,746 | -34% | 1 | 1 | 0% | 2,108 | 2,890 | +37% | 0 | 0 | — |
case-07 | fail→pass | 7,669 | 4,222 | -45% | 1 | 1 | 0% | 1,511 | 1,669 | +10% | 0 | 0 | — |
case-08 | fail→pass | 6,780 | 2,401 | -65% | 1 | 1 | 0% | 1,100 | 1,563 | +42% | 0 | 0 | — |
case-10 | fail→pass | 10,492 | 3,073 | -71% | 1 | 1 | 0% | 1,914 | 1,643 | -14% | 0 | 0 | — |
case-11 | fail→pass | 10,801 | 3,822 | -65% | 1 | 1 | 0% | 1,768 | 1,682 | -5% | 0 | 0 | — |
case-12 | fail→pass | 6,993 | 2,766 | -60% | 1 | 1 | 0% | 1,275 | 1,544 | +21% | 0 | 0 | — |
case-13 | fail→pass | 14,523 | 10,127 | -30% | 1 | 1 | 0% | 2,675 | 3,120 | +17% | 0 | 0 | — |
case-14 | fail→pass | 10,708 | 4,208 | -61% | 1 | 1 | 0% | 1,692 | 1,897 | +12% | 0 | 0 | — |
case-15 | pass→pass | 12,079 | 3,224 | -73% | 1 | 1 | 0% | 1,757 | 1,647 | -6% | 0 | 0 | — |
case-16 | fail→pass | 10,932 | 4,806 | -56% | 1 | 1 | 0% | 1,980 | 2,106 | +6% | 0 | 0 | — |
case-17 | fail→pass | 8,496 | 4,514 | -47% | 1 | 1 | 0% | 1,299 | 1,740 | +34% | 0 | 0 | — |
case-18 | pass→pass | 2,621 | 1,280 | -51% | 1 | 1 | 0% | 396 | 1,262 | +219% | 0 | 0 | — |
case-23 | fail→pass | 7,570 | 4,392 | -42% | 1 | 1 | 0% | 1,261 | 1,748 | +39% | 0 | 0 | — |
case-19 | pass→pass | 9,083 | 3,187 | -65% | 1 | 1 | 0% | 1,468 | 1,731 | +18% | 0 | 0 | — |
case-20 | fail→pass | 11,652 | 5,087 | -56% | 1 | 1 | 0% | 2,419 | 2,200 | -9% | 0 | 0 | — |
case-21 | fail→pass | 14,133 | 5,479 | -61% | 1 | 1 | 0% | 2,514 | 2,139 | -15% | 0 | 0 | — |
case-22 | pass→pass | 6,599 | 3,273 | -50% | 1 | 1 | 0% | 1,355 | 1,734 | +28% | 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 +65 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.