科研 Skill
学术文献全文合法获取
food-fetch
Lawfully acquires full-text PDFs via open access, the user’s reference manager, or their logged-in library session; never bypasses paywalls. Use to fetch papers before research or review.
SKILL.md
Food-Fetch — Lawful Full-Text Acquisition
Turn a list of references or DOIs into read full text, so food-research,
food-deep-research, and food-review build their knowledge base from the actual
papers rather than abstracts. Original work; the routing/manifest architecture is
informed by the open-source nature-downloader skill (see the repo README
Acknowledgements). Legal-access-only, and it never fabricates — a paper that was
not obtained is recorded as not obtained, never summarized as if read.
First-run setup — ask once, remember, remind if skipped
On the first run that needs full text, help the user set up access — this is what
decides whether the review reads real papers. Check
python3 scripts/food_fetch_setup.py status (exit 3 = not set up):
-
If not set up, surface a highlighted request offering the choices, then save the answer so it isn't asked again:
📚 Set up full-text access (one-time). To read papers, not just abstracts, pick one:
- Reference-manager library — send the path to your EndNote
.Datafolder (or Zoterostorage// Mendeley). Best coverage for cited work. - Institutional access — I'll reach entitled full text through your own logged-in library session in the browser when needed.
- Open access only — no setup, but see the warning below.
⚠️ Without access to non-open-access articles, the accuracy of the results is substantially limited — most published papers are paywalled, so the knowledge base and every claim's grounding would rest on abstracts + the ~half of literature that is open access.
Save the choice:
food_fetch_setup.py set --library <path>/--institutional/--open-access-only. - Reference-manager library — send the path to your EndNote
-
If already set up (exit 0), use the saved mode silently — do not re-ask.
-
If the saved mode is
open_access_only, remind at the start of each full-text run — briefly, once — that non-OA access is not configured and that this substantially limits accuracy, and offer to switch (options 1–2). Don't block; proceed at open-access + abstract level.
The store holds only a folder path and a mode — never a password, cookie, or token.
Access ladder (each article, in order) — see references/access-routing.md
- Open access — resolve a legal free PDF (Unpaywall → OpenAlex; PMC/Europe PMC
full text; arXiv/preprints) and download it.
oa_fetcherrunsscripts/fetch_oa.py. Open-access full text is always fetched and read — never left at abstract-level. - The user's reference-manager library —
library_fetcherreads the PDFs the user already downloaded in EndNote (<Library>.Data/PDF/), Zotero (storage/), or Mendeley, mapping each reference by DOI/title. Read-only. - The user's entitled institutional access —
institutional_fetcherreuses the user's own logged-in library session in the browser (Claude-in-Chrome) to open the article through their library resolver / EZproxy / Shibboleth-CARSI / WebVPN and save the entitled PDF. User-driven login only. Seereferences/institutional-access.md. - User-supplied PDFs — anything the user drops in the project folder.
If none reaches it, the article is oa_not_found / library_no_permission — the
user is told plainly, and the source stays unread (not summarized).
Subagents (dispatch via the Agent tool)
fetch_coordinator— takes the reference list / DOIs, confirms the Supporting Information choice once, runs the batch, and returns the manifest + a plain summary of read vs unread.access_router— per article, normalize the DOI/title, detect open-access vs paywalled and publisher, and choose the ladder step (references/access-routing.md).oa_fetcher— download every open-access PDF viascripts/fetch_oa.py; verify each is a real PDF (%PDF); record source + SHA-256.library_fetcher— match references to PDFs in the user's reference-manager library folder and read them (read-only).institutional_fetcher— with the user's own browser session, reach the library-entitled full text and save the PDF; hand off to the user for any login/CAPTCHA/2FA (references/institutional-access.md).content_reader— normalize each obtained article (XML/HTML/PDF) into clean sectioned full text, preferring structured JATS XML over scraped PDF (references/format-reading.md); never treat an abstract or a login page as the paper.
Deliverable — a coverage manifest (references/manifest-and-status.md)
Always return one manifest listing every requested reference with a typed status
(open_access_downloaded · library_pdf_read · institutional_downloaded ·
oa_not_found · library_no_permission · pdf_fetch_failed · not_a_pdf), its
access route, and file/SHA-256 where downloaded — plus a one-line count of how many
were read in full vs not. This is what makes the review's grounding honest and
auditable. Run scripts/privacy_scan.py on anything delivered.
Boundaries (non-negotiable) — references/boundaries.md
Lawful access only: legal OA, the user's own library files, and the user's own authorized institutional session. Never bypass a paywall, DRM, or 2FA; never read, type, export, or ask for passwords / cookies / OTP codes / session tokens; never scrape against a site's terms; no bulk/indiscriminate downloading — only the confirmed list. For a login or verification challenge, hand the browser to the user.
Called by
food-research (screener_appraiser/data_extractor), food-deep-research
(investigator/source_verifier), and food-review (knowledge_builder) call
food-fetch to obtain full text before extracting; the agri-* skills via delegation.
Prerequisites
scripts/fetch_oa.py(added in this override) needs network access and talks to the free Unpaywall and OpenAlex APIs — no API key, but Unpaywall asks for a contact email on every request (--email, defaults to a placeholder; pass the user's own email for real use).- For step 2 (reference-manager library), the user's EndNote
.Datafolder, Zoterostorage/, or Mendeley folder must already exist locally with downloaded PDFs; this skill only reads it, it does not install or configure a reference manager.
Limitations
- Step 3 (institutional access) assumes the current agent can drive the user's own logged-in browser session ("Claude-in-Chrome" in the original text) — a Claude.ai/Claude Code browser-automation feature. In an agent without an equivalent browser-control tool, this route is unavailable; fall back to open access (step 1) and user-supplied PDFs (step 4), and tell the user so instead of silently skipping it.
- EZproxy/Shibboleth are common outside China/Taiwan; CARSI/WebVPN are specific to
Mainland Chinese university libraries. All four are named in
references/institutional-access.md, but a library using a different resolver needs its own URL pattern added there.
示例提示词
帮我通过开放获取渠道找到这几篇营养学论文的全文PDF。
所属插件
安装该插件会一并获得同领域的 20 个 Skill / MCP。
同插件的其他科研 Skill
- 农业深度研究综述Deep research producing a source-validated literature review for an agricultural question, ranked by journal quartile. Reuses food-deep-research machinery. Use for an agronomy deep dive.
- 农业科学论文撰写系统Multi-subagent manuscript system for agricultural science with field-trial/soil/animal reporting standards. Reuses food-paper machinery. Use for an agronomy or soil-science manuscript.
- 农业研究全流程编排Orchestrates the full agricultural research-to-publication workflow: journal selection, research, writing, figures, peer review, revision. Use to manage an agronomy project end to end.
- 农业综述转可编辑PPTConverts an agricultural-science report or manuscript into a fully editable PowerPoint deck, reusing food-ppt machinery with agronomy framing. Use for slides from a field-trial report.
- 农业遥感作物制图Maps crop types from Sentinel-2 imagery with Random Forest, extracts NDVI phenology, estimates LAI, and predicts yield via Google Earth Engine. Use for field/county crop remote sensing.
- 农业文献证据综合Multi-source literature and evidence-synthesis workflow for agricultural science, ranked by journal quartile. Reuses food-research machinery. Use for an agronomy or soil-science evidence review.
- 农业论文多审稿人模拟Simulates an editorial peer-review panel for an agricultural manuscript, grounded in cited and field literature. Use for a pre-submission review or mock referee report in agriculture.
- 作物生长模拟Simulates crop growth with WOFOST/AquaCrop via PCSE, pulls weather from NASA POWER, and runs yield-gap analysis under climate scenarios. Use for agronomic intervention evidence.
- 田间试验混合模型分析Analyzes RCBD, split-plot, and multi-site field-trial data with mixed models in R (lme4) or Python (statsmodels), using correct error strata. Use for a crop or soil field-trial analysis.
- 通用深度文献综述研究General-purpose deep research producing a source-validated, fully written literature review via a 12-subagent team. Use for a thorough review or deep-dive needing verified sources.
- 食品营养论文图表系统Builds publication-ready figures (bar/box, dose-response, PCA, heatmaps, forest plots) to a target journal’s spec, with a privacy scan before delivery. Use for a journal-ready figure.
- 食品科学论文撰写系统Multi-subagent manuscript system for food/nutrition science: frames questions, analyzes data, writes statistics, drafts, polishes, and reformats for a journal. Use to write or revise a paper.
- 食品研究全流程编排Orchestrates the whole food/nutrition research-to-publication workflow with quality gates: journal selection, research, writing, figures, review, revision. Use to manage a paper end to end.
- 食品研究报告转可编辑PPTConverts a food/agriculture report or manuscript into a fully editable PowerPoint deck with python-pptx: outline, grounded slide text, themed layout. Use to make slides from a paper.
- 科研基金申请书编排Orchestrates a food/agriculture funding proposal to a scheme’s format (headings, length limit, criteria), drafting or revising via the research/paper/review skills. Use for a grant application.
- 食品营养文献综合研究Runs a multi-source literature and evidence-synthesis workflow for food/nutrition science: search, two-phase screening, synthesis, evidence grading. Use for a deep review or systematic-review scope.
- 食品论文多审稿人模拟Simulates an editorial peer-review panel for a food/nutrition manuscript, grounded in cited and field literature first. Use for a pre-submission review or mock referee report.
- FAOSTAT食品系统数据分析Analyzes FAOSTAT food-balance, dietary-diversity, and trade/production data for cross-country food-systems indicators. Use for a country-comparison food-security dataset.
- 土壤数据空间分析Pulls and analyzes soil properties from SoilGrids (global) and SSURGO (US-only), computes SOC stock, and runs kriging interpolation. Use for a soil-property or soil-carbon map.
常见问题
学术文献全文合法获取能做什么?
Lawfully acquires full-text PDFs via open access, the user’s reference manager, or their logged-in library session; never bypasses paywalls. Use to fetch papers before research or review。
如何安装学术文献全文合法获取?
在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「学术文献全文合法获取」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。
学术文献全文合法获取需要哪些权限?
该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。
学术文献全文合法获取经过官方审核吗?
是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。