Skill v1.0.0
currentAutomated scan100/100version: "1.0.0" name: ai4scholar-paper-search description: "Use AI4Scholar to search real scholarly papers across Semantic Scholar, PubMed, Google Scholar, arXiv, bioRxiv, and medRxiv. Trigger for AI4Scholar paper search, latest literature, top journal literature discovery, cross-database search, paper search prompts, or when the user needs traceable real references rather than invented citations."
AI4Scholar Paper Search
Use this skill to search real scholarly literature through AI4Scholar MCP or the OpenClaw AI4Scholar plugin.
Source guide: https://lifu-coze.feishu.cn/wiki/WOaewK33Ei2g1nkt44kcRXiBnze AI4Scholar MCP endpoint: https://mcp.ai4scholar.net/sse
Non-Negotiable Rules
- Never fabricate references.
- Never expose an API key. Use
${AI4SCHOLAR_API_KEY}or "your AI4Scholar API key". - Prefer verified identifiers: DOI, PMID, arXiv ID, Semantic Scholar ID, publisher URL.
- Mark unverifiable or weakly matched results as needing manual verification.
- For management, psychology, neuroscience, BCI, HCI, and text-mining topics, search both classic and recent papers when possible.
Tools From The Source Guide
Use the exact schemas exposed by the MCP server or plugin. Common search tools described by the source guide include:
| Tool | Source | Best use | |
|---|---|---|---|
search_semantic | Semantic Scholar | Broad semantic literature search with year filters | |
search_pubmed | PubMed | Biomedical, psychology-adjacent, neuroscience, clinical, health topics | |
search_google_scholar | Google Scholar proxy | Broad cross-disciplinary discovery | |
search_arxiv | arXiv | AI, ML, CS, BCI algorithms, HCI preprints | |
search_biorxiv | bioRxiv | Biology and neuroscience preprints | |
search_medrxiv | medRxiv | Medical and clinical preprints | |
search_semantic_snippets | Semantic Scholar | Search text snippets inside paper full text | |
search_semantic_bulk | Semantic Scholar | Large search, up to 1000 results if supported | |
search_semantic_paper_match | Semantic Scholar | Exact title matching |
Search Workflow
- Convert the user question into 2-4 precise English queries. Add Chinese queries only when Chinese literature is relevant.
- Choose databases:
- Management, psychology, HCI, IS: Semantic Scholar + Google Scholar.
- Neuroscience, EEG, ERP, BCI: PubMed + Semantic Scholar + arXiv when algorithms are involved.
- AI/ML/text mining: arXiv + Semantic Scholar + Google Scholar.
- Run at least two sources when available.
- Deduplicate by DOI, title, year, and first author.
- Prioritize:
- recent five-year papers,
- top journals or major conferences,
- highly cited classics,
- papers that match the user's method and variables.
- Output a literature matrix before writing narrative synthesis.
Screening Criteria
Score each candidate paper from 0-2 on:
| Criterion | 0 | 1 | 2 | |
|---|---|---|---|---|
| Topic fit | unrelated | adjacent | directly relevant | |
| Method fit | wrong method | partially similar | same or transferable method | |
| Evidence quality | unclear | usable | strong design/data | |
| Venue quality | unknown | field-normal | top or authoritative | |
| Citation traceability | no identifier | partial metadata | DOI/PMID/arXiv/publisher URL |
Keep papers scoring at least 7/10 unless the user requests broad exploration.
Output Template
Research question:AI4Scholar mode: MCP / OpenClaw plugin / unknownDatabases searched:Query strings:Filters:Top papers:| # | Title | Authors | Year | Venue | DOI/ID | Source | Why it matters | Verification |Excluded or weak papers:| Title | Reason |Next search move:
Beginner Prompt
Use AI4Scholar to search real English-language papers on "[topic]".Prioritize the most recent five years, while keeping necessary classic papers.Use at least Semantic Scholar plus one of PubMed, Google Scholar, or arXiv.Return title, authors, year, venue, DOI/PMID/arXiv ID, abstract, source, and manual-verification status.
Quality Checks
- If results are empty, broaden the query and try another database.
- If the same paper appears across sources, merge the metadata instead of duplicating it.
- If a result has no DOI or stable identifier, do not cite it as verified.
- If the user asks for "top journal" evidence, explain the venue screen used.