Create Research Brief

A comprehensive two-phase skill for designing multi-LLM research strategies (Phase 1) and consolidating multi-model outputs into actionable intelligence (Phase 2).

1. Purpose

This skill provides 9 core capabilities:

Checkpoints

This skill uses interactive checkpoints (see

references/checkpoints.yaml

• research_type_classification — When research type is ambiguous
• risk_depth_selection — When risk assessment depth not specified
• model_mode_selection — When model execution mode not specified
• hypothesis_priors_required — When multi_hypothesis enabled but priors missing
• conflict_resolution_approach — When model outputs have significant conflicts (Phase 2)

2. Two-Phase Workflow

Phase 1: Research Design (Before Research)

multi_hypothesis=true

expert_panel=true

Phase 2: Consolidation (After Research)

3. Parameters

research_objective

research_type

market

model_mode

parallel

openai_depth

balanced

risk_depth

standard

multi_hypothesis

false

expert_panel

false

context

""

4. Model Strengths & Assignment

Model Profiles

Default Category Assignments

5. Risk Assessment Depths

Quick (5 Factors)

Basic risk identification for time-sensitive research:

• Top 3 risks with likelihood/impact
• No mitigations or scenarios

Standard (+ Bias Audit)

Adds mitigation planning and cognitive bias check:

• Mitigations and contingencies per risk
• Early warning signals
• Bias audit: confirmation, availability, anchoring

Comprehensive (+ Base Rates)

Full risk analysis with historical grounding:

• Risk scenarios with trigger conditions
• Risk dependencies and cascades
• Base rate comparison from similar research
• Pre-mortem analysis

6. MECE Decomposition Patterns

Pattern 1: Market Research

Pattern 2: Competitive Intelligence

Pattern 3: Technology Evaluation

Pattern 4: Strategic Research

7. Multi-Hypothesis Framing

When to Enable

• Testing predictions or forecasts
• Evaluating competing theories
• Decision involves binary or multi-way choice
• Need to avoid confirmation bias

Process

1. Define core question as testable prediction
2. Generate 2-4 MECE hypotheses covering all outcomes
3. Assign prior probabilities (must sum to 100%)
4. Define supporting and refuting evidence for each
5. Research gathers evidence against criteria
6. Update posteriors based on evidence strength

Example

<hypotheses question="Will enterprise adopt GenAI for customer service by 2027?">
  <hypothesis id="H1" position="broad" prior="30%">
    >50% enterprise adoption
  </hypothesis>
  <hypothesis id="H2" position="selective" prior="50%">
    10-50% adoption in specific use cases
  </hypothesis>
  <hypothesis id="H3" position="limited" prior="20%">
    <10% adoption due to barriers
  </hypothesis>
</hypotheses>

8. Evidence Strength Tribunal

5-point scale for evaluating source quality:

Time Decay: Apply -1 for technology data >6 months, market data >1 year.

Reference: See

references/evidence-strength-rubric.md

9. Conflict Resolution: WWHTBT

When models or sources disagree and resolution isn't clear, apply What Would Have To Be True analysis:

<conflict claim="Market size for X">
  <position holder="Gartner" value="$50B">
    <evidence score="4">2024 market report with methodology</evidence>
  </position>
  <position holder="IDC" value="$35B">
    <evidence score="4">Different scope definition</evidence>
  </position>

  <wwhtbt>
    <for_gartner>
      <condition>Adjacent markets included in scope</condition>
      <condition>Projected vs. realized revenue counted</condition>
    </for_gartner>
    <for_idc>
      <condition>Only core product category</condition>
      <condition>Realized revenue only</condition>
    </for_idc>
  </wwhtbt>

  <recommendation>
    Report range ($35-50B) with scope dependency noted.
    For our purposes, IDC definition more aligned.
  </recommendation>
</conflict>

10. Uncertainty Decomposition

Classification Questions

• Epistemic: "Does someone, somewhere know this?"
• Aleatory: "Even with perfect info, would this still be uncertain?"
• Model: "Would a different definition change the answer?"

Reference: See

references/uncertainty-taxonomy.md

11. Gap Analysis

Part 1: MECE Coverage Audit

Compare findings against expected coverage matrix for research type. Flag:

• Critical gaps: Core dimensions missing or Score ≤2
• Significant gaps: Supporting dimensions weak
• Minor gaps: Context items missing

Part 2: Unknown Unknowns Probes

Reference: See

references/gap-analysis-protocol.md

12. Output Specifications

Phase 1 Deliverable: Research Brief

research-brief.xml
├── Header (ID, type, mode, parameters)
├── Section 1: Research Classification
├── Section 2: MECE Question Decomposition
├── Section 3: Multi-Hypothesis Framing (if enabled)
├── Section 4: Risk Assessment
├── Section 5: Expert Panel (if enabled)
├── Section 6: Model Role Assignments
├── Section 7: Ready-to-Execute Prompts
├── Section 8: Consolidation Strategy
├── Section 9: Verification Priorities
└── Section 10: Effort Estimates

Phase 2 Deliverable: Consolidated Report

consolidated-report.xml
├── Header (quality summary)
├── Part 1: Executive Summary (≤5 findings, bottom line)
├── Part 2: Tiered Findings (1: >75%, 2: 50-75%, 3: <50%)
├── Part 3: Evidence Quality Assessment
├── Part 4: Contested Claims & Conflict Resolution
├── Part 5: Uncertainty Analysis
├── Part 6: Gap Analysis
├── Part 7: Model Contribution Analysis
├── Part 8: Decision Support (if-then tree)
├── Part 9: Kill Criteria
├── Part 10: Methodology Transparency
├── Part 11: Appendices
└── CRITICAL CONSTRAINTS (at end for context retention)

Templates: See

templates/research-brief-template.md

templates/consolidated-report-template.md

13. Expert Panel Integration

When to Enable

• High-stakes decisions
• Multi-disciplinary topics
• Need for challenge/red-teaming
• Regulatory or compliance implications

Process

1. Identify panel size (3-8 experts) and balance
2. Select domain-appropriate experts
3. Define deliberation format (round-robin, debate, Delphi)
4. Assign challenger role for assumption testing
5. Synthesize panel perspectives into findings

Expert Selection by Domain

14. Quality Gates

Phase 1 Gates (Research Design)

Phase 2 Gates (Consolidation)

15. Use Cases

16. Workflow Integration

This skill integrates with the broader research workflow:

┌─────────────────────┐
│ research-interviewer│  Elicit research requirements
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│create-research-brief│  ◀── THIS SKILL (Phase 1)
│     (Phase 1)       │  Design multi-LLM research strategy
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│   Execute Research  │  Run prompts across models
│  (Manual or Agent)  │
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│create-research-brief│  ◀── THIS SKILL (Phase 2)
│     (Phase 2)       │  Consolidate into report
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│ consolidate-research│  Additional synthesis if needed
└─────────────────────┘

17. References and Templates

Reference Files

references/evidence-strength-rubric.md

references/uncertainty-taxonomy.md

references/gap-analysis-protocol.md

references/mece-decomposition-guide.md

Template Files

templates/research-brief-template.md

templates/consolidated-report-template.md

Quick Start

Phase 1: Create Research Brief

/create-research-brief
research_objective: "What is the market opportunity for AI legal research tools?"
research_type: market
risk_depth: standard

Phase 2: Consolidate Research

/create-research-brief --phase=2
input: [model outputs from Phase 1 execution]