🔬 COMPREHENSIVE RESEARCH WORKFLOW FOR KAI

YOU (Kai) are reading this because a research request was detected by the load-context hook.

This command provides instructions for YOU to orchestrate comprehensive multi-source research by directly invoking researcher commands (NOT spawning new Claude Code sessions).

🎯 YOUR MISSION

When a user asks for research, YOU must deliver FAST RESULTS through massive parallelization:

1. Launch up to 10 research agents in parallel - Spawn multiple instances of each researcher (perplexity-researcher, claude-researcher, gemini-researcher) with different query angles
2. Each agent does ONE query + ONE follow-up - Quick, focused research cycles
3. Collect results as they complete (typically within 15-30 seconds)
4. Synthesize findings into a comprehensive report
5. Report back using the mandatory response format

Speed Strategy:

• Launch 10 agents simultaneously to cover the entire search space
• Each agent handles a specific angle/sub-question
• Parallel execution = results in under 1 minute
• Follow-up queries only when critical information is missing

📋 STEP-BY-STEP WORKFLOW

Step 1: Decompose Question & Launch Agent Army (10 parallel agents)

SPEED IS THE PRIORITY - Launch up to 10 research agents simultaneously

Step 1a: Break Down the Research Question

First, decompose the user's question into 3-10 specific sub-questions that cover:

• Different angles of the topic
• Specific aspects to investigate
• Related areas that provide context
• Potential edge cases or controversies

Step 1b: Launch Research Agents in Parallel (up to 10 agents)

Use the Task tool to spawn multiple specialized research agents simultaneously:

// Launch 3-10 agents in PARALLEL - each with ONE specific sub-question
// Use a SINGLE message with multiple Task tool calls

Task({
  subagent_type: "perplexity-researcher",
  description: "Research sub-question 1",
  prompt: "Research this specific angle: [sub-question 1]. Do ONE focused search query and ONE follow-up if needed. Return findings quickly."
})

Task({
  subagent_type: "claude-researcher",
  description: "Research sub-question 2",
  prompt: "Research this specific angle: [sub-question 2]. Do ONE focused search query and ONE follow-up if needed. Return findings quickly."
})

Task({
  subagent_type: "gemini-researcher",
  description: "Research sub-question 3",
  prompt: "Research this specific angle: [sub-question 3]. Do ONE focused search query and ONE follow-up if needed. Return findings quickly."
})

// Continue launching up to 10 agents total
// Mix perplexity-researcher, claude-researcher, gemini-researcher
// Each gets ONE focused sub-question

Available Research Agents:

• perplexity-researcher: Fast Perplexity API searches
• claude-researcher: Claude WebSearch with intelligent query decomposition
• gemini-researcher: Google Gemini multi-perspective research

CRITICAL RULES FOR SPEED:

1. ✅ Launch ALL agents in ONE message (parallel execution)
2. ✅ Each agent gets ONE specific sub-question (focused research)
3. ✅ Maximum 10 agents (covers entire search space)
4. ✅ Each agent does 1 query + 1 follow-up max (quick cycles)
5. ✅ Results return in 15-30 seconds (parallel processing)
6. ❌ DON'T launch sequentially (kills speed benefit)
7. ❌ DON'T give broad questions (forces multiple iterations)

Step 2: Collect Results (15-30 seconds)

Wait for agents to complete - they typically return results within 15-30 seconds due to parallel execution.

Each agent returns:

• Focused findings from their specific sub-question
• Source citations
• Confidence indicators
• Quick insights

Step 3: Synthesize Results

Create a comprehensive report that:

A. Identifies Confidence Levels:

• HIGH CONFIDENCE: Findings corroborated by multiple sources
• MEDIUM CONFIDENCE: Found by one source, seems reliable
• LOW CONFIDENCE: Single source, needs verification

B. Structures Information:

## Key Findings

### [Topic Area 1]
**High Confidence:**
- Finding X (Sources: perplexity-research, claude-research)
- Finding Y (Sources: perplexity-research, claude-research)

**Medium Confidence:**
- Finding Z (Source: claude-research)

### [Topic Area 2]
...

## Source Attribution
- **Perplexity-Research**: [summary of unique contributions]
- **Claude-Research**: [summary of unique contributions]

## Conflicting Information
- [Note any disagreements between sources]

C. Calculate Research Metrics:

• Total Queries: Count all queries across all research commands
• Services Used: List unique services (Perplexity API, Claude WebSearch, etc.)
• Total Output: Estimated character/word count of all research
• Confidence Level: Overall confidence percentage
• Result: 1-2 sentence answer to the research question

Step 4: Return Results Using MANDATORY Format

📅 [current date from date command] 📋 SUMMARY: Research coordination and key findings overview 🔍 ANALYSIS: Synthesis of multi-source research results ⚡ ACTIONS: Which research commands executed, research strategies used ✅ RESULTS: Complete synthesized findings with source attribution 📊 STATUS: Research coverage, confidence levels, data quality ➡️ NEXT: Recommended follow-up research or verification needed 🎯 COMPLETED: Completed multi-source [topic] research 🗣️ CUSTOM COMPLETED: [Optional: Voice-optimized under 8 words]

📈 RESEARCH METRICS:

• Total Queries: [X] (Primary: [Y], Secondary: [Z])
• Services Used: [N] (List: [service1, service2])
• Total Output: [~X words/characters]
• Confidence Level: [High/Medium/Low] ([percentage]%)
• Result: [Brief summary answer]

🚨 CRITICAL RULES FOR KAI

1. LAUNCH UP TO 10 AGENTS IN PARALLEL - Use a SINGLE message with multiple Task tool calls
2. DECOMPOSE the question first - Create 3-10 focused sub-questions
3. ONE QUERY + ONE FOLLOW-UP per agent - Quick, focused research cycles
4. MIX agent types - Use perplexity-researcher, claude-researcher, gemini-researcher
5. WAIT for ALL agents to complete (15-30 seconds) before synthesizing
6. SYNTHESIZE results - Don't just concatenate outputs
7. USE the mandatory response format - This triggers voice notifications
8. CALCULATE accurate metrics - Count queries, agents, output size
9. ATTRIBUTE sources - Show which agent/method found each insight
10. MARK confidence levels - Based on multi-source agreement

SPEED CHECKLIST:

• ✅ Launched agents in ONE message? (parallel execution)
• ✅ Each agent has ONE focused sub-question?
• ✅ Using up to 10 agents for broad coverage?
• ✅ Agents instructed to do 1 query + 1 follow-up max?
• ✅ Expected results in under 1 minute?

🚧 HANDLING BLOCKED OR FAILED CRAWLS

If research commands report being blocked, encountering CAPTCHAs, or facing bot detection, note this in your synthesis and recommend using:

• mcp__Brightdata__scrape_as_markdown - Scrape single URLs that bypass bot detection
• mcp__Brightdata__scrape_batch - Scrape multiple URLs (up to 10)
• mcp__Brightdata__search_engine - Search Google, Bing, or Yandex with CAPTCHA bypass
• mcp__Brightdata__search_engine_batch - Multiple search queries simultaneously

💡 EXAMPLE EXECUTION

User asks: "Research the latest developments in quantum computing"

Your workflow:

1. ✅ Recognize research intent (hook loaded this command)

2. ✅ Decompose into focused sub-questions:

   • What are the major quantum computing breakthroughs in 2025?
   • Which companies are leading quantum computing development?
   • What are the current limitations and challenges?
   • What practical applications are emerging?
   • What's the state of quantum error correction?
   • How close are we to quantum advantage?
   • What are the latest quantum algorithms?
   • What's happening in quantum cryptography?

3. ✅ Launch 8 agents in PARALLEL (ONE message with 8 Task calls):

   Task(perplexity-researcher, "2025 quantum breakthroughs")
   Task(claude-researcher, "Leading quantum companies")
   Task(gemini-researcher, "Quantum limitations 2025")
   Task(perplexity-researcher, "Practical quantum applications")
   Task(claude-researcher, "Quantum error correction state")
   Task(gemini-researcher, "Quantum advantage timeline")
   Task(perplexity-researcher, "Latest quantum algorithms")
   Task(claude-researcher, "Quantum cryptography developments")
   

4. ✅ Wait for ALL agents to complete (15-30 seconds)

5. ✅ Synthesize their findings:

   • Common themes → High confidence
   • Unique insights → Medium confidence
   • Disagreements → Note and flag

6. ✅ Calculate metrics (8 agents, ~16 queries, 3 services, output size, confidence %)

7. ✅ Return comprehensive report with mandatory format

8. ✅ Voice notification automatically triggered by your 🎯 COMPLETED line

Result: User gets comprehensive quantum computing research from 8 parallel agents in under 1 minute, with multi-source validation, source attribution, and confidence levels.

🎤 VOICE NOTIFICATIONS

Voice notifications are AUTOMATIC when you use the mandatory response format. The stop-hook will:

• Extract your 🎯 COMPLETED line
• Send it to the voice server with Jamie (Premium) voice at 228 wpm
• Announce "Completed multi-source [topic] research"

YOU DO NOT NEED TO MANUALLY SEND VOICE NOTIFICATIONS - just use the format.

🔄 BENEFITS OF THIS ARCHITECTURE

Why parallel agent execution delivers speed:

1. ✅ 10 agents working simultaneously - Not sequential, truly parallel
2. ✅ Results in under 1 minute - Each agent does 1-2 quick searches
3. ✅ Complete coverage - Multiple perspectives from different services
4. ✅ Focused research - Each agent has ONE specific sub-question
5. ✅ No iteration delays - All agents launch at once in ONE message
6. ✅ Multi-source validation - High confidence from cross-agent agreement

Speed Comparison:

• ❌ Old way: Sequential searches → 5-10 minutes
• ✅ New way: 10 parallel agents → Under 1 minute

This is the correct architecture. Use it for FAST research.