brycewang-stanford/skills/05-kthorn-research-superpower/research/answering-research-questions

<!-- ╔══════════════════════════════════════════════════════════════╗ ║ 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。 -->

---

name: Answering Research Questions description: Main orchestration workflow for systematic literature research - search, evaluate, traverse, synthesize when_to_use: When user asks a research question. When user wants to find specific data in literature. When starting comprehensive literature review. When user says "find papers about" or "what is known about". version: 1.0.0

Answering Research Questions

Overview

Orchestrate the complete research workflow from query to findings.

Core principle: Systematic, trackable, comprehensive. Search → Evaluate → Traverse → Synthesize.

Announce at start: "I'm using the Answering Research Questions skill to find [specific data] about [topic]."

The Process

Phase 1: Parse Query

Extract from user's request:

Keywords:

• Main concepts (e.g., "BTK inhibitor", "selectivity")
• Synonyms and alternatives (e.g., "Bruton tyrosine kinase")
• Related terms (e.g., "off-target", "kinase panel")

Data types needed:

• Specific measurements (IC50, KD, EC50, etc.)
• Methods or protocols
• Structures or sequences
• Results or conclusions

Constraints:

• Date ranges
• Specific compounds/targets
• Organisms or systems
• Publication types

Ask clarifying questions if needed:

• "Are you looking for in vitro or in vivo data?"
• "Any specific time frame?"
• "Which kinases are you most interested in?"
• "What email address should I use for Unpaywall API requests?" (Required for finding open access papers)

Phase 2: Initialize Research Session

Propose folder name:

research-sessions/YYYY-MM-DD-brief-description/

Example: research-sessions/2025-10-11-btk-inhibitor-selectivity/

Show proposal to user:

📁 Creating research folder: research-sessions/2025-10-11-btk-inhibitor-selectivity/
   Proceed? (y/n)

Create folder structure:

mkdir -p "research-sessions/YYYY-MM-DD-description"/{papers,citations}

Initialize files:

Core files (always create these):

papers-reviewed.json:

{}

citations/citation-graph.json:

{}

SUMMARY.md:

# Research Query: [User's question]

**Started:** YYYY-MM-DD HH:MM
**Keywords:** keyword1, keyword2, keyword3
**Data types sought:** IC50 values, selectivity data, synthesis methods

---

## Highly Relevant Papers (Score ≥ 8)

Papers scored using `evaluating-paper-relevance` skill:
- Score 0-10 based on: Keywords (0-3) + Data type (0-4) + Specificity (0-3)
- Score ≥ 8: Highly relevant with significant data
- Score 7: Relevant with useful data
- Score 5-6: Possibly relevant
- Score < 5: Not relevant

(Papers will be added here as found)

Example format:
### [Paper Title](https://doi.org/10.1234/example)
**DOI:** [10.1234/example](https://doi.org/10.1234/example) | **PMID:** [12345678](https://pubmed.ncbi.nlm.nih.gov/12345678/)

---

## Relevant Papers (Score 7)

(Papers will be added here as found)

---

## Possibly Relevant Papers (Score 5-6)

(Noted for potential follow-up)

---

## Search Progress

- Initial PubMed search: X results
- Papers reviewed: Y
- Papers with relevant data: Z
- Citations followed: N

---

## Key Findings

(Synthesized findings will be added as research progresses)

CRITICAL: Always use clickable markdown links for DOIs and PMIDs

Auxiliary files (for large searches >100 papers):

See evaluating-paper-relevance skill for guidance on when to create:

• README.md - Project overview, methodology, file inventory
• TOP_PRIORITY_PAPERS.md - Curated priority list organized by tier
• evaluated-papers.json - Rich structured data for programmatic access

For small searches (<50 papers), stick to core files only. For large searches (>100 papers), auxiliary files add significant organizational value.

Phase 3: Search Literature

Use searching-literature skill:

1. Construct PubMed query from keywords
2. Execute search (start with 100 results)
3. Save results to initial-search-results.json
4. Report: "🔎 Found N papers matching query"

Phase 4: Evaluate Papers

Use evaluating-paper-relevance skill:

For each paper:

1. Check papers-reviewed.json (skip if already processed)
2. Stage 1: Score abstract (0-10)
3. If score ≥ 7: Stage 2 deep dive
4. Extract findings to SUMMARY.md
5. Download PDF and supplementary if available
6. Update papers-reviewed.json (for ALL papers, even low-scoring ones)
7. If score ≥ 7: proceed to Phase 5 for this paper

CRITICAL: Add every paper to papers-reviewed.json regardless of score. This prevents re-review and tracks complete search history.

Report progress for EVERY paper:

📄 [15/100] Screening: "Paper Title"
   Abstract score: 8 → Fetching full text...
   ✓ Found IC50 data for 8 compounds
   → Added to SUMMARY.md

📄 [16/100] Screening: "Another Paper"
   Abstract score: 3 → Skipping (not relevant)

📄 [17/100] Screening: "Third Paper"
   Abstract score: 7 → Relevant, adding to queue...

Every 10 papers, give summary update

Phase 5: Traverse Citations

Use traversing-citations skill:

For papers scoring ≥ 7:

1. Get references (backward)
2. Get citations (forward)
3. Filter for relevance (score ≥ 5)
4. Add to processing queue
5. Evaluate queued papers (return to Phase 4)

Report progress:

🔗 Following citations from highly relevant paper
   → Found 12 relevant references
   → Found 8 relevant citing papers
   → Adding 20 papers to queue

Phase 6: Checkpoint

Check after:

• Every 50 papers reviewed
• Every 5 minutes of processing
• Queue exhausted

Ask user:

⏸️  Checkpoint: Reviewed 50 papers, found 12 relevant
    Papers with data: 7
    Continue searching? (y/n/summary)

Options:

• y - Continue processing
• n - Stop and finalize
• summary - Show current findings, then decide

Phase 7: Synthesize Findings

When stopping (user says no or queue empty):

Option A: Manual synthesis (small research sessions)

1. Review SUMMARY.md - Organize by relevance and topic
2. Extract key findings - Group by data type
3. Add synthesis section:

## Key Findings Summary

### IC50 Values for BTK Inhibitors
- Compound A: 12 nM (Smith et al., 2023)
- Compound B: 45 nM (Doe et al., 2024)
- [More compounds...]

### Selectivity Data
- Compound A shows >80-fold selectivity vs other kinases
- Tested against panel of 50 kinases (Jones et al., 2023)

### Synthesis Methods
- Lead compounds synthesized via [method]
- Yields: 30-45%
- Full protocols in [papers]

### Gaps Identified
- No data on selectivity vs [specific kinase]
- Limited in vivo data
- Few papers on resistance mechanisms

4. Update search progress stats
5. List all files downloaded

Option B: Script-based synthesis (large research sessions >50 papers)

For large research sessions, consider creating a synthesis script:

create generate_summary.py:

• Read evaluated-papers.json from helper scripts
• Aggregate findings by priority and scaffold type
• Generate comprehensive SUMMARY.md with:
  • Executive summary with statistics
  • Papers grouped by relevance score
  • Priority recommendations for next steps
  • Methodology documentation
• Include timestamps and reproducibility info

Benefits:

• Consistent formatting across sessions
• Easy to regenerate as more papers added
• Can customize grouping/filtering logic
• Documents complete methodology

Final report:

✅ Research complete!

📊 Summary:
   - Papers reviewed: 127
   - Relevant papers: 18
   - Highly relevant: 7
   - Data extracted: IC50 values for 45 compounds, selectivity data, synthesis methods

📁 All findings in: research-sessions/2025-10-11-btk-inhibitor-selectivity/
   - SUMMARY.md (organized findings)
   - papers/ (14 PDFs + supplementary data)
   - papers-reviewed.json (complete tracking)

Phase 8: Final Consolidation

CRITICAL: Always consolidate findings at the end

1. Create relevant-papers.json

Filter papers-reviewed.json to extract only relevant papers (score ≥ 7):

# Read papers-reviewed.json
with open('papers-reviewed.json') as f:
    all_papers = json.load(f)

# Filter for relevant papers (score >= 7)
relevant_papers = {
    doi: data for doi, data in all_papers.items()
    if data.get('score', 0) >= 7
}

# Save to relevant-papers.json
with open('relevant-papers.json', 'w') as f:
    json.dump(relevant_papers, f, indent=2)

Format:

{
  "10.1234/example1.2023": {
    "pmid": "12345678",
    "title": "Paper title",
    "status": "highly_relevant",
    "score": 9,
    "source": "pubmed_search",
    "timestamp": "2025-10-11T16:00:00Z",
    "found_data": ["IC50 values", "synthesis methods"],
    "chembl_id": "CHEMBL1234567"
  },
  "10.1234/example2.2023": {
    "pmid": "23456789",
    "title": "Another paper",
    "status": "relevant",
    "score": 7,
    "source": "forward_citation",
    "timestamp": "2025-10-11T16:15:00Z",
    "found_data": ["MIC data"]
  }
}

2. Enhance SUMMARY.md with Methodology Section

Add these sections to the TOP of existing SUMMARY.md (before paper listings):

# Research Query: [User's question]

**Date:** 2025-10-11
**Duration:** 2h 15m
**Status:** Complete

---

## Search Strategy

**Keywords:** BTK, Bruton tyrosine kinase, inhibitor, selectivity, off-target, kinase panel, IC50
**Data types sought:** IC50 values, selectivity data, kinase panel screening
**Constraints:** None (open date range)

**PubMed Query:**

("BTK" OR "Bruton tyrosine kinase") AND (inhibitor OR "kinase inhibitor") AND (selectivity OR "off-target")

---

## Screening Methodology

**Rubric:** Abstract scoring (0-10)
- Key terms: +3 pts each (or Keywords 0-3, Data type 0-4, Specificity 0-3 if using old rubric)
- Relevant terms: +1 pt each
- Threshold: ≥7 = relevant

**Sources:**
- Initial PubMed search
- Forward/backward citations via Semantic Scholar

---

## Results Statistics

**Papers Screened:**
- Total reviewed: 127 papers
- Highly relevant (≥8): 12 papers
- Relevant (7): 18 papers
- Possibly relevant (5-6): 23 papers
- Not relevant (<5): 74 papers

**Data Extracted:**
- IC50 values: 45 compounds across 12 papers
- Selectivity data: 8 papers with kinase panel screening
- Full text obtained: 18/30 relevant papers (60%)

**Citation Traversal:**
- Papers with citations followed: 7
- References screened: 45 papers
- Citing papers screened: 38 papers
- Relevant papers found via citations: 8 papers

---

## Key Findings Summary

### IC50 Values for BTK Inhibitors
- Ibrutinib: 0.5 nM (Smith et al., 2023)
- Acalabrutinib: 3 nM (Doe et al., 2024)
- [Additional findings synthesized from papers below]

### Selectivity Patterns
- Most inhibitors show >50-fold selectivity vs other kinases
- Common off-targets: TEC, BMX (other TEC family kinases)

### Gaps Identified
- Limited data on selectivity vs JAK/SYK
- Few papers on resistance mechanisms
- No in vivo selectivity data found

---

## File Inventory

- `SUMMARY.md` - This file (methodology + findings)
- `relevant-papers.json` - 30 relevant papers (score ≥7)
- `papers-reviewed.json` - All 127 papers screened
- `papers/` - 18 PDFs + 5 supplementary files
- `citations/citation-graph.json` - Citation relationships

---

## Reproducibility

**To reproduce:**
1. Use PubMed query above
2. Apply screening rubric (threshold ≥7)
3. Follow citations from highly relevant papers (≥8)
4. Check Unpaywall for paywalled papers

**Software:** Research Superpowers skills v2025-10-11

---

[Existing paper listings follow below...]

## Highly Relevant Papers (Score ≥ 8)

### [Paper Title]...

Report to user:

✅ Research session complete!

📄 Consolidation complete:
   1. SUMMARY.md - Enhanced with methodology, statistics, and findings
   2. relevant-papers.json - 30 relevant papers (score ≥7) in JSON format

📁 All files in: research-sessions/2025-10-11-btk-inhibitor-selectivity/
   - SUMMARY.md (complete: methodology + paper-by-paper findings)
   - relevant-papers.json (30 relevant papers for programmatic access)
   - papers-reviewed.json (127 total papers screened)
   - papers/ (18 PDFs)

🔍 Quick access:
   - Open SUMMARY.md for complete findings and methodology
   - Use relevant-papers.json for programmatic access

💡 Optional: Clean up intermediate files?
   → Use cleaning-up-research-sessions skill to safely remove temporary files

Workflow Checklist

Use TodoWrite to track these steps:

• [ ] Parse user query (keywords, data types, constraints)
• [ ] Propose and create research folder
• [ ] Initialize tracking files (SUMMARY.md, papers-reviewed.json, citation-graph.json)
• [ ] Search PubMed using searching-literature skill
• [ ] For each paper: evaluate using evaluating-paper-relevance skill
• [ ] For relevant papers (≥7): traverse citations using traversing-citations skill
• [ ] Report progress regularly
• [ ] Checkpoint every 50 papers or 5 minutes
• [ ] When done: synthesize findings and enhance SUMMARY.md with methodology
• [ ] Create relevant-papers.json (filtered JSON for programmatic access)
• [ ] Final report with stats and file locations

Integration Points

Skills used:

1. searching-literature - Initial PubMed search
2. evaluating-paper-relevance - Score and extract from papers
3. traversing-citations - Follow citation networks

All skills coordinate through:

• Shared papers-reviewed.json (deduplication)
• Shared SUMMARY.md (findings accumulation)
• Shared citation-graph.json (relationship tracking)

File organization:

• Small searches (<50 papers): Core files only (papers-reviewed.json, SUMMARY.md, citation-graph.json)
• All searches: Create relevant-papers.json at end; enhance SUMMARY.md with methodology
• Large searches (>100 papers): May add auxiliary files (README.md, TOP_PRIORITY_PAPERS.md, evaluated-papers.json) for better organization

Error Handling

No results found:

• Try broader keywords
• Remove constraints
• Check spelling
• Try different synonyms

API rate limiting:

• Report to user: "⏸️ Rate limited, waiting..."
• Wait required time
• Resume automatically

Full text unavailable:

• Note in SUMMARY.md
• Continue with abstract-only evaluation
• Flag for manual retrieval if highly relevant

Too many results (>500):

• Suggest narrowing query
• Process first 100, ask if continue
• Focus on most recent or most cited

Quick Reference

| Phase | Skill | Output | |-------|-------|--------| | Parse | (built-in) | Keywords, data types, constraints | | Initialize | (built-in) | Folder, SUMMARY.md, tracking files | | Search | searching-literature | List of papers with metadata | | Evaluate | evaluating-paper-relevance | Scored papers, extracted findings | | Traverse | traversing-citations | Additional papers from citations | | Synthesize | (built-in) | Enhanced SUMMARY.md with methodology + findings | | Consolidate | (built-in) | relevant-papers.json (filtered to score ≥7) |

Common Mistakes

Not tracking all papers: Only adding relevant papers to papers-reviewed.json → Add EVERY paper to prevent re-review, track complete history Creating unnecessary auxiliary files for small searches: For <50 papers, stick to core files (papers-reviewed.json, SUMMARY.md, citation-graph.json). For large searches (>100 papers), auxiliary files like README.md and TOP_PRIORITY_PAPERS.md add value. Silent work: User can't see progress → Report EVERY paper, give updates every 10 Non-clickable identifiers: Plain text DOIs/PMIDs → Always use markdown links Jumping to evaluation without good search: Too narrow results → Optimize search first Not tracking papers: Re-reviewing same papers → Always use papers-reviewed.json Following all citations: Exponential explosion → Filter before traversing No checkpoints: User loses context → Report and ask every 50 papers Poor synthesis: Just list papers → Group by data type, extract key findings Batch reporting: Reporting 20 papers at once → Report each one as you go

User Communication (CRITICAL)

NEVER work silently! User needs continuous feedback.

Report frequency:

• Every paper: Brief status as you screen (📄 [N/Total] Title... Score: X)
• Every 5-10 papers: Progress summary with counts
• Every finding: Immediately report what data you found
• Every decision point: Ask before changing direction

Be specific in progress reports:

• ✅ "Found IC50 = 12 nM for compound 7 (Table 2)"
• ❌ "Found data"
• ✅ "Screening paper 25/127: Not relevant (score 3)"
• ❌ Silently skip papers

Ask for clarification when needed:

• ✅ "Are you looking for in vitro or in vivo IC50 values?"
• ❌ Assume and potentially waste time

Report blockers immediately:

• ✅ "⚠️ Paper behind paywall - evaluating from abstract only"
• ❌ Silently skip without mentioning

Periodic summaries (every 10-15 papers):

📊 Progress update:
   - Reviewed: 30/127 papers
   - Highly relevant: 3 (scores 8-10)
   - Relevant: 5 (score 7)
   - Currently: Screening paper 31...

Why: User can course-correct early, knows work is happening, can stop if needed

Success Criteria

Research session successful when:

• All relevant papers found and evaluated
• Specific data extracted and organized
• Citations followed systematically
• No duplicate processing
• Clear SUMMARY.md with actionable findings
• User questions answered with evidence

Next Steps

After completing research:

• User reviews SUMMARY.md and relevant-papers.json
• Optional: Run cleaning-up-research-sessions skill to remove intermediate files
• May request deeper dive into specific papers
• May request follow-up searches with refined keywords
• May archive or share research session folder