<!-- ╔══════════════════════════════════════════════════════════════╗ ║ 本文件为开源 Skill 原始文档，收录仅供学习与研究参考 ║ ║ CoPaper.AI 收集整理 | https://copaper.ai ║ ╚══════════════════════════════════════════════════════════════╝ 来源仓库: https://github.com/kthorn/research-superpower 项目名称: research-superpower 开源协议: MIT License 收录日期: 2026-04-02 声明: 本文件版权归原作者所有。此处收录旨在为社会科学实证研究者 提供 AI Agent Skills 的集中参考。如有侵权，请联系删除。 -->

Getting Started with Research Superpowers

Research Superpowers gives Claude Code systematic workflows for literature searching and review.

Focus: Finding, screening, and extracting data from published papers. NOT for analyzing experimental data or designing experiments.

What You Can Do

Use these skills for systematic literature reviews:

• Search literature - PubMed and Semantic Scholar integration
• Build screening rubrics - Define and test relevance criteria collaboratively
• Screen papers - Two-stage screening (abstract → deep dive) with scoring
• Extract data - Find specific methods, results, measurements from papers
• Traverse citations - Smart backward/forward citation following
• Large-scale screening - Parallel subagent processing for 50+ papers
• Track findings - Organized research sessions with summaries, PDFs, and deduplication

Available Skills

Literature Search & Review Skills (skills/research/)

• answering-research-questions - Main orchestration workflow (search → screen → extract → synthesize)
• building-screening-rubrics - Collaborative rubric design with test-driven refinement
• searching-literature - PubMed search with keyword optimization
• evaluating-paper-relevance - Two-stage screening (abstract → deep dive)
• subagent-driven-review - Parallel screening for large searches (50+ papers)
• checking-chembl - Check if medicinal chemistry papers have curated SAR data in ChEMBL
• traversing-citations - Semantic Scholar citation network traversal
• finding-open-access-papers - Unpaywall API to find free versions of paywalled papers
• cleaning-up-research-sessions - Safe cleanup of intermediate files after research complete

Basic Workflow

When user asks a literature search question:

1. Read answering-research-questions skill - Main orchestration
2. Announce: "I'm using the Answering Research Questions skill"
3. Parse query - Extract keywords, data types, constraints
4. Create research folder - Propose name, initialize tracking
5. Optional: Build rubric - For large searches (50+ papers), use building-screening-rubrics skill
6. Search → Screen → Extract → Traverse - Follow the workflow
7. Check in regularly - Every 10 papers, checkpoint every 50

Research Session Folders

Each query creates a folder in research-sessions/:

research-sessions/YYYY-MM-DD-query-description/
├── SUMMARY.md              # Main findings
├── papers-reviewed.json    # Deduplication tracking (DOI → status)
├── papers/                 # Downloaded PDFs and supplementary data
└── citations/              # Citation graph tracking

Core Principles

For systematic literature review:

• Precision over breadth - Find papers with specific data you need, not just topical matches
• Test-driven screening - Build and validate rubrics before bulk processing
• Smart citation following - Only traverse relevant citations to avoid exponential explosion
• Deduplicate aggressively - Track ALL reviewed papers by DOI (even non-relevant)
• Cache abstracts - Save for re-screening when rubrics change
• Report progress - Update user every 10 papers as work proceeds
• Checkpoint frequently - Ask to continue or stop every 50 papers
• Reproducible - Save rubrics, queries, and methodology with research sessions

API Information

PubMed E-utilities (no key required):

• Search: https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi
• Details: https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esummary.fcgi
• Full text: https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi

Semantic Scholar (free tier works, optional key for higher limits):

• Paper: https://api.semanticscholar.org/graph/v1/paper/DOI:{doi}
• References: https://api.semanticscholar.org/graph/v1/paper/{id}/references
• Citations: https://api.semanticscholar.org/graph/v1/paper/{id}/citations

Finding Skills

Use the find-skills script to search for relevant skills:

# From project directory
./scripts/find-skills              # List all skills
./scripts/find-skills literature   # Search for "literature"
./scripts/find-skills 'cite|ref'   # Regex search

Remember

• Always start by reading the relevant research skill
• Announce skill usage when you begin
• Track everything in the research folder
• Check in with user regularly during long searches
• Deduplicate using papers-reviewed.json (DOI as key)