
research-synthesis
by poemswe
Plugin and Skills for Claude Code, Gemini CLI and Codex
SKILL.md
name: research-synthesis description: Synthesizes research findings into coherent narratives with uncertainty quantification. Use when integrating findings from multiple sources, creating research summaries, drawing conclusions from evidence, or communicating research results. Triggers on phrases like "synthesize", "integrate findings", "what's the conclusion", "summarize research", "overall picture", "bring together". tools:
- WebSearch
- WebFetch
- Read
- Grep
- Glob
Research Synthesis
This skill guides the integration of diverse research findings into coherent, actionable conclusions.
Phase 1: Synthesis Preparation
Input Assessment
- What sources/findings need synthesis?
- What is the overarching research question?
- Who is the audience for this synthesis?
- What decisions will this inform?
Source Inventory
| Source | Type | Quality | Key Contribution |
|---|---|---|---|
| [Source] | [Type] | [Rating] | [Main finding] |
Compatibility Check
- Do sources address the same question?
- Are methodologies compatible?
- Can findings be meaningfully compared?
- Are there definitional inconsistencies?
CHECKPOINT: Confirm synthesis scope and purpose with user.
Phase 2: Pattern Recognition
Finding Categorization
Group findings by:
By Conclusion:
- Consistent findings (agree)
- Inconsistent findings (disagree)
- Complementary findings (different aspects)
- Unique findings (only one source)
By Evidence Strength:
- Strong evidence (multiple high-quality sources)
- Moderate evidence (some quality sources)
- Weak evidence (limited or low-quality sources)
- Contested (conflicting strong sources)
Convergence Analysis
For each major finding:
- How many sources support it?
- What is their combined quality?
- Are there methodological differences?
- Do any sources contradict?
Phase 3: Weight Assignment
Evidence Weighting Factors
| Factor | Weight Modifier |
|---|---|
| Sample size | Larger = higher weight |
| Study design | RCT > observational |
| Peer review | Reviewed > not reviewed |
| Replication | Replicated > single study |
| Recency | More recent = higher (usually) |
| Relevance | Direct > indirect evidence |
Confidence Levels
- High confidence: Multiple high-quality sources agree, no major contradictions
- Moderate confidence: Good evidence but some limitations or gaps
- Low confidence: Limited evidence, quality concerns, or contradictions
- Very low confidence: Minimal evidence, major limitations
- Insufficient: Cannot draw conclusions
Phase 4: Contradiction Resolution
When Sources Disagree
Step 1: Verify actual disagreement
- Are they measuring the same thing?
- Are conditions comparable?
- Could both be true in different contexts?
Step 2: Assess relative quality
- Which has stronger methodology?
- Which has larger sample?
- Which is more recent?
Step 3: Identify explanatory factors
- Population differences
- Methodological differences
- Context differences
- Time period differences
Step 4: Synthesis approach
| Situation | Approach |
|---|---|
| Quality difference | Favor higher quality |
| Context difference | Specify conditions |
| Genuine debate | Present both positions |
| Unexplained | Acknowledge uncertainty |
CHECKPOINT: Present contradictions and proposed resolution for user input.
Phase 5: Narrative Construction
Synthesis Structure Options
Conceptual Framework: Organize around theoretical concepts
Concept 1 → Concept 2 → Concept 3
↓ ↓ ↓
[Findings] [Findings] [Findings]
Chronological: Trace evolution of understanding
Early understanding → Key developments → Current state
Problem-Solution: Frame around practical questions
Problem → Evidence → Solutions → Remaining gaps
Argument-Based: Build toward conclusions
Claim → Evidence → Counterclaim → Resolution → Conclusion
Narrative Elements
- Opening: Context and importance
- Body: Organized evidence presentation
- Integration: How pieces connect
- Limitations: What we don't know
- Conclusion: Key takeaways
Phase 6: Uncertainty Quantification
Uncertainty Sources
| Source | Description | Handling |
|---|---|---|
| Measurement | Data collection errors | Acknowledge precision limits |
| Sampling | Non-representative samples | Note generalizability limits |
| Model | Theoretical assumptions | Test sensitivity |
| Conflict | Disagreeing sources | Present range of views |
| Gap | Missing information | Explicitly note unknowns |
Uncertainty Communication
Use calibrated language:
| Confidence | Language |
|---|---|
| Very high (>95%) | "The evidence clearly shows..." |
| High (80-95%) | "The evidence strongly suggests..." |
| Moderate (60-80%) | "The evidence suggests..." |
| Low (40-60%) | "Some evidence indicates..." |
| Very low (<40%) | "Limited evidence hints at..." |
Phase 7: Actionable Conclusions
Conclusion Formulation
For each key conclusion:
- State the finding clearly
- Specify confidence level
- Note key supporting evidence
- Acknowledge limitations
- Identify implications
Recommendation Framework
| Evidence Strength | Recommendation Type |
|---|---|
| Strong | Direct recommendation |
| Moderate | Conditional recommendation |
| Weak | Suggestion for consideration |
| Insufficient | No recommendation (need more research) |
Phase 8: Documentation
Output Structure
# Research Synthesis: [Topic]
## Executive Summary
[2-3 paragraph overview of key findings]
## Purpose
[Research question and synthesis goals]
## Sources Synthesized
[Brief description of evidence base]
## Key Findings
### Finding 1: [Statement]
**Confidence**: [Level]
**Evidence**: [Summary of supporting sources]
**Caveats**: [Limitations or conditions]
### Finding 2: [Statement]
[Same structure]
## Areas of Uncertainty
- [Uncertainty 1]: [Description and implications]
- [Uncertainty 2]: [Description and implications]
## Contradictions and Debates
- [Topic]: [Summary of disagreement and interpretation]
## Conclusions
[Integrated conclusions with confidence levels]
## Implications
- For [audience 1]: [Implications]
- For [audience 2]: [Implications]
## Research Gaps
[What remains unknown and needs investigation]
## References
[Formatted citations]
File Writing Instructions
CRITICAL: After generating the synthesis, you MUST write it to a file.
Context Detection:
- If called from
/researchorchestration: output directory path will be available in context - If called standalone: create new directory
Standalone Mode (synthesis invoked directly, not via /research):
- Create directory:
mkdir -p research-synthesis-outputs/$(date +%Y-%m-%d_%H-%M-%S) - Write synthesis to
synthesis-report.mdin that directory - Notify user of file location
Orchestrated Mode (called from /research):
- Use the output directory path from research orchestration context
- Write synthesis to
final-synthesis.mdin that directory - Add metadata header:
# Research Synthesis: [Topic]
**Completed**: [Timestamp]
**Agents Used**: [List of agents that contributed]
**Research Query**: [Original query]
---
[Rest of synthesis content]
File Writing:
- Use Write tool to save the complete synthesis
- Preserve all markdown formatting
- Include all sections (Executive Summary through References)
- Confirm file creation to user with full path
Error Handling:
- If Write fails: notify user, provide synthesis in conversation
- If directory doesn't exist in orchestrated mode: create it with fallback naming
CHECKPOINT: Review synthesis for accuracy and completeness with user, and confirm file has been written.
Score
Total Score
Based on repository quality metrics
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
3ヶ月以内に更新がある
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
Reviews
Reviews coming soon