Parallel Research

Before spawning subagents, ALWAYS ask user for confirmation:

"I'll use the parallel-research skill to investigate [TOPIC] by spawning [N] subagents simultaneously. This will take ~4-6 minutes. Proceed?"

Why: Subagent spawning consumes API resources and requires user awareness of parallel execution.

The Pattern

1. Define Scope

Create a clear research brief with:

• Specific questions to answer
• Areas to investigate (market, technical, legal, etc.)
• Deliverable format
• Success criteria

2. Spawn Subagents

Spawn 2-3 subagents with identical or complementary prompts:

sessions_spawn:
  task: "Research [TOPIC] focusing on [ANGLE]. 
         
         Key questions:
         1. [Question 1]
         2. [Question 2]
         
         Provide specific data, cite sources, avoid hype."
  label: "Research-[ANGLE]"
  thinking: "high"

Example angles for emerging tech:

• Subagent 1: Market sizing & competition
• Subagent 2: Technical feasibility & architecture
• Subagent 3: Legal/regulatory landscape

3. Parallel Execution

• All subagents run simultaneously (4-5 min each)
• No coordination needed during execution
• Save results to workspace files

4. Synthesize Findings

Review all outputs for:

• Convergence: What do multiple agents agree on? (High confidence)
• Divergence: Where do they disagree? (Needs deeper investigation)
• Gaps: What did no one cover? (Missing angles)
• Opportunities: What emerges from the intersection?

5. Deliver Synthesis

Present executive summary with:

• Key findings (prioritized by confidence)
• Actionable insights
• Recommended next steps
• Links to full reports

Benefits

Real Example: Voice NFTs × AI Payments

Prompt (all 3 subagents):

Research [DOMAIN] deeply. Focus on:
1. Market size and growth drivers
2. Existing players (what's working/failing)
3. Technical architecture
4. Legal/regulatory considerations
5. Real use cases (not hype)
6. Opportunities for builders

Provide specific data, cite sources. Save full report to workspace.

Results:

• All 3 converged: "Voice NFTs barely exist as category" (blue ocean confirmed)
• Market estimates aligned: $2-5B TAM by 2030
• Technical assessment + Market analysis = Convergence opportunity: Voice auth for AI payments

Outcome: 10x better insight than single-agent research.

Prompt Templates

Template A: Identical Prompts (for cross-validation)

Research [TOPIC] comprehensively. Cover:
- What is it? (clear definition)
- Market opportunity (size, growth, timing)
- Technical feasibility (what's possible now vs future)
- Existing solutions (who's building, what's working/failing)
- Challenges and risks
- Opportunities for [user context]

Be thorough. Use web_search, web_fetch. Cite sources.
Save full report to: research/[topic]-[timestamp].md

Template B: Complementary Angles

Subagent 1 (Market):

Research [TOPIC] from MARKET perspective:
- TAM/SAM/SOM analysis
- Competition mapping
- Growth drivers and timing
- Investment/funding landscape
- Go-to-market strategies

Subagent 2 (Technical):

Research [TOPIC] from TECHNICAL perspective:
- Current state of the art
- Architecture patterns
- Implementation complexity
- Performance benchmarks
- Integration challenges

Subagent 3 (Legal/Regulatory):

Research [TOPIC] from LEGAL perspective:
- Regulatory status by jurisdiction
- Compliance requirements
- Liability frameworks
- Intellectual property considerations
- Emerging legislation

Tool Usage

Recommended tools for research subagents:

• web_search - Current information, market data
• web_fetch - Deep dive into specific sources
• read - Analyze existing workspace files
• write - Save findings to structured reports

NOT recommended:

• sessions_spawn (avoid recursion)
• browser (unless specific site analysis needed)
• External API calls (adds latency)

Output Format

Subagents should save structured reports:

# [Topic] Research Report

## Executive Summary
2-3 sentences on key findings

## 1. [Section]
Detailed findings with data and citations

## 2. [Section]
...

## Sources
- [Title](URL) - Key insight
- [Title](URL) - Key insight

## Confidence Assessment
- High confidence: [findings with multiple sources]
- Medium confidence: [findings with limited sources]
- Speculative: [emerging trends, projections]

Synthesis Template

After reviewing all subagent outputs:

## Research Synthesis: [Topic]

### 🎯 Key Findings
What converged across all agents (highest confidence)

### 📊 Market Opportunity
Size, timing, positioning

### 🔧 Technical Reality
What's possible today vs 2-3 years

### 💡 Opportunities
Specific, actionable paths forward

### ⚠️ Challenges
Known blockers and risks

### 📄 Full Reports
- [Report 1](path)
- [Report 2](path)
- [Report 3](path)

Performance Notes

• Runtime: 4-6 minutes total (parallel)
• Cost: ~40-80k tokens per subagent (3x single agent)
• Quality: Significantly higher due to cross-validation
• Best for: Complex, ambiguous, or emerging topics

Variations

Quick Validation (2 subagents)

For time-sensitive research:

• Spawn 2 agents with identical prompts
• Focus on convergence only
• Skip detailed synthesis

Deep Investigation (3+ subagents)

For critical decisions:

• 3+ agents with complementary angles
• Include contradictory viewpoints intentionally
• Extended synthesis with decision matrix

Intersection Analysis

For exploring domain convergence:

• Agent 1: Domain A deep dive
• Agent 2: Domain B deep dive
• Agent 3: Explicit intersection analysis

Common Pitfalls

❌ Vague prompts → Agents wander, low-quality output
✅ Specific questions → Focused, actionable findings

❌ Too many agents → Diminishing returns, coordination overhead
✅ 2-3 agents optimal → Good coverage, manageable synthesis

❌ Ignoring divergence → Miss important nuances
✅ Highlight disagreements → Indicates uncertainty or opportunity

Integration with Self-Improvement

After using this skill:

1. Log workflow effectiveness to .learnings/LEARNINGS.md
2. Note which angles produced most value
3. Refine templates based on results
4. Promote proven patterns to AGENTS.md

Example Session

User: "Research Voice NFTs and AI Agent Payments"

Action:
1. Spawn subagent: "Research Voice NFTs" → 4 min
2. Spawn subagent: "Research AI Agent Payments" → 4 min  
3. Spawn subagent: "Research voice authorization" → 4 min
4. All complete simultaneously
5. Synthesize convergence findings
6. Identify intersection opportunity

Result: Voice NFT signatures for AI payment authorization

Version: 1.0.0
Last Updated: 2026-02-17
Tested On: Voice NFTs, AI Agent Payments, Voice Authorization research