research-synthesis

poemswe 的头像作者 poemswe

将研究结果综合成连贯的叙述,并量化不确定性。用于整合多个来源的发现、创建研究摘要、从证据中得出结论或传达研究结果。触发短语包括'综合'、'整合发现'、'结论是什么'、'总结研究'、'整体图景'、'汇集起来'

📖 中文导读:将研究结果综合成连贯的叙述,并量化不确定性。用于整合多个来源的发现、创建研究摘要、从证据中得出结论或传达研究结果。触发短语包括'综合'、'整合发现'、'结论是什么'、'总结研究'、'整体图景'、'汇集起来' 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。

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:

  1. How many sources support it?
  2. What is their combined quality?
  3. Are there methodological differences?
  4. 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

  1. Opening: Context and importance
  2. Body: Organized evidence presentation
  3. Integration: How pieces connect
  4. Limitations: What we don't know
  5. 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:

  1. If called from /research orchestration: output directory path will be available in context
  2. If called standalone: create new directory

Standalone Mode (synthesis invoked directly, not via /research):

  1. Create directory: mkdir -p research-synthesis-outputs/$(date +%Y-%m-%d_%H-%M-%S)
  2. Write synthesis to synthesis-report.md in that directory
  3. Notify user of file location

Orchestrated Mode (called from /research):

  1. Use the output directory path from research orchestration context
  2. Write synthesis to final-synthesis.md in that directory
  3. 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.