Run Research Pipeline
Execute a complete research workflow from problem definition through consolidated findings.
When to Use
Use this skill when you need:
- •End-to-end research on a complex topic
- •Structured progression through all research phases
- •Multi-LLM research design and synthesis
- •Comprehensive analysis with quality gates at each phase
Pipeline Overview
code
┌─────────────────────────────────────────────────────────────┐ │ RESEARCH PIPELINE │ ├─────────────────────────────────────────────────────────────┤ │ │ │ Phase 1: ELICITATION │ │ ┌─────────────────────┐ │ │ │ research-interview │ → PROBLEM-STATEMENT │ │ └─────────────────────┘ │ │ ↓ │ │ Phase 2: DESIGN │ │ ┌─────────────────────┐ │ │ │ research-brief │ → Model-optimized prompts │ │ └─────────────────────┘ │ │ ↓ │ │ Phase 3: EXECUTION (User) │ │ ┌─────────────────────┐ │ │ │ Run prompts in │ → Raw model outputs │ │ │ Claude, Gemini, GPT │ │ │ └─────────────────────┘ │ │ ↓ │ │ Phase 4: SYNTHESIS │ │ ┌─────────────────────┐ │ │ │ consolidate-research│ → CONSOLIDATED-REPORT │ │ └─────────────────────┘ │ │ │ └─────────────────────────────────────────────────────────────┘
Workflow
Execute complete research pipeline for: "$ARGUMENTS"
Phase 1: Knowledge Elicitation
Run research-interviewer to produce PROBLEM-STATEMENT:
Objective: Define the research question clearly with:
- •Core question and sub-questions
- •Key constraints and assumptions
- •Success criteria
- •Known information vs gaps
Output: <problem-statement> artifact with confidence scores
Quality Gate:
- • Research question is specific and answerable
- • Assumptions surfaced and examined
- • Scope boundaries defined
- • Success criteria measurable
Phase 2: Research Design
Run create-research-brief Phase 1 to generate:
Objective: Design multi-LLM research strategy with:
- •MECE question decomposition (5 categories)
- •Model-optimized prompts for Claude, Gemini, GPT
- •Risk assessment for the research approach
Model Assignment Strategy:
| Model | Strength | Best For |
|---|---|---|
| Claude Opus 4.5 | Judgment, synthesis | Strategic questions, nuance |
| Gemini Pro 3 | Breadth, citations | Factual lookup, sourcing |
| GPT-5.2 Deep | Recency, depth | Technical details, edge cases |
Output: <research-brief> with prompts for each model
Quality Gate:
- • MECE decomposition complete (no overlaps/gaps)
- • Each sub-question independently researchable
- • Model assignments match model strengths
- • Prompts optimized for each model's style
Phase 3: User Executes Prompts
PAUSE POINT - Display prompts for user to execute:
markdown
## Ready for Multi-Model Research Execute the following prompts in each model: ### Claude (Opus 4.5) [Display Claude prompt] ### Gemini (Pro 3) [Display Gemini prompt] ### GPT (5.2 Deep) [Display GPT prompt] --- **Instructions:** 1. Copy each prompt to its designated model 2. Collect the responses 3. Paste all responses back here to continue to Phase 4
Wait for user to provide model outputs before continuing.
Phase 4: Consolidation
Run create-research-brief Phase 2 to synthesize:
Objective: Produce unified findings with:
- •Evidence scoring (5-point scale)
- •Conflict resolution (WWHTBT protocol)
- •Uncertainty classification
- •MECE coverage audit
Output: <consolidated-report> with:
- •Executive summary
- •Findings by category with evidence scores
- •Conflicts resolved and rationale
- •Remaining gaps and recommendations
Quality Gate:
- • All model outputs incorporated
- • Evidence scores assigned
- • Conflicts explicitly resolved
- • Actionable recommendations provided
Output Format
Final pipeline output includes all artifacts:
xml
<research-pipeline-output>
<metadata>
<topic>$ARGUMENTS</topic>
<pipeline_id>[unique identifier]</pipeline_id>
<completed_phases>[1,2,3,4]</completed_phases>
</metadata>
<phase-1-output>
<problem-statement>...</problem-statement>
</phase-1-output>
<phase-2-output>
<research-brief>...</research-brief>
</phase-2-output>
<phase-3-inputs>
<model-response model="claude">...</model-response>
<model-response model="gemini">...</model-response>
<model-response model="gpt">...</model-response>
</phase-3-inputs>
<phase-4-output>
<consolidated-report>...</consolidated-report>
</phase-4-output>
<next-steps>
[Recommended follow-up actions]
</next-steps>
</research-pipeline-output>
Quality Gates (Pipeline-Level)
- • All four phases completed
- • Each phase passed its quality gates
- • Artifacts linked and traceable
- • Final recommendations actionable
- • Confidence levels appropriate for decision-making
Post-Pipeline Options
After completing the research pipeline:
| If you need to... | Use skill... |
|---|---|
| Evaluate options from findings | /compare-options |
| Get expert validation | /run-expert-panel |
| Document the methodology | /write-reference |
| Create implementation guide | /write-howto |
| Optimize follow-up prompts | /optimize-prompt |