AgentSkillsCN

scientific-paper-searcher

在 PubMed、arXiv、bioRxiv 以及 Semantic Scholar 上搜索科研论文。当用户需要查找某一主题的科学文献时,可使用此技能。

SKILL.md
--- frontmatter
name: scientific-paper-searcher
description: Search PubMed, arXiv, bioRxiv, and Semantic Scholar for research papers. Use when user needs to find scientific literature on a topic.

Scientific Paper Searcher

When to Use

  • User asks to find papers on a scientific topic
  • User needs references for research
  • User asks "what papers exist about X"

Don't use for: Full systematic reviews (use literature-review skill), meta-analyses, citation verification only.

Prerequisites

Run ./scripts/check-prereqs - needs curl and internet connectivity.

Workflow

1. Clarify Request

Extract: topic, domain (biomedical/CS/physics/general), recency needs, quantity (default 10-15).

2. Select Databases

DomainPrimarySecondary
BiomedicalPubMed, bioRxivSemantic Scholar
CS/AI/MLarXivSemantic Scholar
Physics/MatharXivSemantic Scholar
InterdisciplinarySemantic ScholarPubMed, arXiv

Search at least 2 databases.

3. Search

Use WebFetch with these URL patterns:

code
PubMed:    https://pubmed.ncbi.nlm.nih.gov/?term=QUERY&size=20
arXiv:     https://arxiv.org/search/?query=QUERY&searchtype=all
bioRxiv:   https://www.biorxiv.org/search/QUERY
Semantic:  https://www.semanticscholar.org/search?q=QUERY

4. Output Format

markdown
## Search Results: [Topic]

**Databases:** PubMed, arXiv | **Papers found:** X

### Key Papers

#### 1. [Title]
- **Authors:** First A, Second B, et al. (Year)
- **Source:** Journal/arXiv
- **Link:** https://doi.org/...
- **Why relevant:** [1 sentence]

[repeat for top 10-15 papers, ranked by relevance + citations]

### Synthesis
[2-3 sentences on what the literature shows]

### Recommended Reading Order
1. Start with...

5. Edge Cases

IssueAction
No resultsBroaden terms, try synonyms
Too many (>100)Add date filter, narrow terms
Database downNote limitation, use others
Wants seminal papersSort by citations, no date filter

Feedback

Run echo "$OUTPUT" | ./scripts/evaluate.sh then ask user: "Were these papers relevant to your research question?"

Log to FEEDBACK.jsonl:

json
{"ts":"<ISO8601>","skill":"scientific-paper-searcher","version":"1.0.0","prompt":"<query>","outcome":<1-5>,"note":"","source":"script|user","schema_version":1}