# Research Paper Data > Search scholarly literature and research data through discoverable providers, including public metadata APIs, optional user-configured sources, legal open-access full-text discovery, identifier resolution, and citation/reference triage. Use when the user asks for paper search, topic or author literature discovery, DOI/arXiv/PMID/PMCID lookup, dataset DOI metadata, legal OA PDF or full-text locations, references, citations, related work, paper metadata, or evidence-backed literature triage. - Skill: `svd-ai-lab/research-paper-data` (Agent Skill, multi-file: 4 files) - Install (CLI): `npx skillmds@latest add svd-ai-lab/research-paper-data` - Raw SKILL.md: https://api.skillmd.com/api/skills/svd-ai-lab/research-paper-data/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Product & Planning - Author: svd-ai-lab (https://skillmd.com/u/svd-ai-lab) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/svd-ai-lab/research-paper-data --- # Research Paper Data ## Overview Use this skill to search scholarly metadata, resolve persistent identifiers, and find legal open-access full text using providers that are visible in the current runtime. Prefer capability discovery and provenance over assuming a specific service, credential, or edition. When `OPENCODE_CONFIG_DIR` is set, first look for `$OPENCODE_CONFIG_DIR/research-capabilities.json`. Treat that file as the edition's provider declaration. If it is missing, continue with public defaults listed below. Read `references/public-apis.md` when making direct API requests or debugging a source-specific response. Use `scripts/academic_search.py` for broad topic, author, affiliation, ORCID, or OpenAlex/Crossref discovery when Python is available. ## Provider Policy Provider declarations use: - `public`: no user secret required. - `user-configured`: requires a visible user-provided setting, usually an environment variable. - `managed`: an edition-provided provider; use only when its declared base URL and auth requirements are visible in the current runtime. Default public sources: - Crossref for article, book, proceedings, and reference metadata. - arXiv for preprints and arXiv PDFs. - Europe PMC for PMID, PMCID, biomedical metadata, and OA links. - DataCite for dataset, software, preprint, and non-article DOI metadata. - OpenAlex discovery through `scripts/academic_search.py` for broad topic, author, affiliation, ORCID, and citation-aware search. Treat OpenAlex results as discovery candidates; resolve selected DOI/arXiv/PMID records through an authoritative source before claiming final metadata or OA availability. Optional user-configured sources: - Use Unpaywall only when `UNPAYWALL_EMAIL` is visible. - For polite OpenAlex/Crossref pool access, pass `--mailto` to `scripts/academic_search.py` or use `OPENALEX_MAILTO`, `CROSSREF_MAILTO`, or `RESEARCH_CONTACT_EMAIL` when visible. - Use enhanced citation, graph, or full-text providers only when declared in `research-capabilities.json` and their required configuration is visible. Use legal open-access/source-discovery paths only. Do not use paywall bypass, credential sharing, scraped account credentials, or undeclared hidden secrets. ## Workflow 1. Discover providers. - Read `research-capabilities.json` when present. - Check required environment variables or local configuration before using a `user-configured` or `managed` provider. - If a declared provider is unavailable, skip it and say why when it affects coverage. 2. Classify the request. - Natural-language topic search: run `scripts/academic_search.py` first for ranked discovery when Python is available; then resolve selected identifiers through Crossref, arXiv, Europe PMC, or DataCite. - Author search or disambiguation: use `--author`, `--affiliation`, `--orcid`, `--author-id`, or `--list-authors`; report ambiguity instead of guessing. - DOI: normalize and query Crossref; if the item looks like a dataset, software, preprint, or Crossref fails, fall back to DataCite. - arXiv ID: query arXiv and construct canonical abstract and PDF URLs. - PMID or PMCID: query Europe PMC. - Citation/reference graph: use declared reference-capable providers and source-specific reference fields; do not present citation relationships as exhaustive unless a source proves them. 3. Find legal full text. - Prefer arXiv PDFs for arXiv records. - Prefer Europe PMC OA links for PMCID/biomedical records. - Use Unpaywall only when configured. - Otherwise report legal OA links found in source metadata, or state that no legal open full text was found. 4. Return compact, auditable results. - Include title, authors, year, venue/container, identifier, source name, source URL, OA URL/PDF URL when found, and confidence notes. - Mark ambiguous matches and do not collapse multiple plausible papers into one answer. - Cite provider provenance and source failures when they affect trust, coverage, or next steps. ## Query Hygiene - Use conservative result limits first, usually 5 to 10 records. - Prefer `python scripts/academic_search.py "" --limit 5 --compact` for broad discovery smoke tests and quick triage. Use JSON output for downstream parsing. - URL-encode user queries and identifiers. - Respect rate limits and retry-after headers. - Prefer HTTPS JSON APIs. For arXiv, parse Atom/XML carefully and preserve the `id`, `published`, `updated`, authors, categories, and PDF link. - For DOI comparisons, normalize by removing URL prefixes and lowercasing the DOI string. ## Response Shape For searches, provide: - `Best matches`: short ranked list with title, year, authors, venue, DOI or other identifier, and source. - `Full text`: legal PDF/full-text candidates or a clear not-found statement. - `Coverage notes`: skipped optional sources, ambiguity, rate limits, or source failures. - `Next step`: one focused recommendation, such as resolving one DOI, searching arXiv variants, or downloading an OA PDF with user approval. For identifier resolution, provide one record when confident, or a small candidate set when not confident. Never fabricate abstracts, citations, PDFs, or author names when the source does not provide them.