Lightweight Parallel Research Orchestration

Lightweight Parallel Research Orchestration

Guide Main Claude to coordinate 2-4 parallel researchers for multi-dimensional queries.

Quick Start

When user asks research question with 2-3 distinct angles:

1. Analyze query dimensions - Identify 2-4 subtopics/​angles
2. Setup progress tracking - Use TodoWrite to track spawning
3. Spawn researchers in parallel - One light-research-researcher per subtopic using Task tool
4. Wait for completion - All researchers finish and save findings
5. Spawn synthesizer - One light-research-report-writer to create final report
6. Report completion - Tell user where to find the synthesis

Workflow

Step 1: Analyze Request

Identify 2-4 distinct subtopics or dimensions to investigate.

Examples:

• "Research quantum computing" → hardware/​qubits, algorithms/​applications, companies/​investments, challenges/​timeline
• "Research cloud gaming latency" → network optimization, codec selection, hardware acceleration
• "Analyze electric vehicles" → battery technology, market trends, charging infrastructure

Goal: Break complex topic into parallel-researchable angles (2-4 dimensions).

Step 2: Extract Session Path

The hook router has already created the session. Extract researchPath from the user context/​prompt.

Format: docs/​research-sessions/​DDMMYYYY_HHMMSS_topic/

Example: docs/​research-sessions/​16112025_201546_quantum_computing/

IMPORTANT: Do NOT create new session - use the provided researchPath.

Step 3: Setup Progress Tracking

Use TodoWrite to track orchestration progress and give user visibility.

Example:

[
  {"content": "Analyze query dimensions", "status": "completed", "activeForm": "Analyzing query dimensions"},
  {"content": "Spawn 4 light-research-researcher agents", "status": "in_progress", "activeForm": "Spawning researchers"},
  {"content": "Wait for research completion", "status": "pending", "activeForm": "Waiting for researchers"},
  {"content": "Spawn light-research-report-writer", "status": "pending", "activeForm": "Spawning synthesizer"},
  {"content": "Report completion to user", "status": "pending", "activeForm": "Reporting completion"}
]

Step 4: Spawn Researchers (Parallel)

CRITICAL: Use Task tool to spawn 2-4 light-research-researcher agents IN PARALLEL (not sequential).

For each researcher:

• subagent_type: "light-research-researcher"
• description: Brief 3-5 word description (e.g., "quantum hardware research")
• prompt: "Research [specific subtopic]. Save your findings to {researchPath}/​{descriptive_topic_name}.md. The research path for this session is: {researchPath}"

Example spawning (4 researchers in parallel):

Task(subagent_type="light-research-researcher", description="quantum hardware qubits", prompt="Research current state of quantum hardware and qubit technology. Save findings to docs/​research-sessions/​.../​quantum_hardware.md. Research path: docs/​research-sessions/​.../")

Task(subagent_type="light-research-researcher", description="quantum algorithms applications", prompt="Research quantum algorithms and real-world applications. Save findings to docs/​research-sessions/​.../​quantum_algorithms.md. Research path: docs/​research-sessions/​.../")

Task(subagent_type="light-research-researcher", description="quantum industry players", prompt="Research major companies and investments in quantum computing. Save findings to docs/​research-sessions/​.../​quantum_industry.md. Research path: docs/​research-sessions/​.../")

Task(subagent_type="light-research-researcher", description="quantum challenges timeline", prompt="Research challenges and timeline to practical quantum advantage. Save findings to docs/​research-sessions/​.../​quantum_challenges.md. Research path: docs/​research-sessions/​.../")

Key Points:

• Give EACH researcher a specific, focused subtopic (don't duplicate)
• Pass the SAME researchPath to all researchers
• Spawn IN PARALLEL (all at once in single message with multiple Task calls)
• Researchers will use WebSearch and save findings to researchPath/

Update TodoWrite: Mark "Spawn researchers" as completed, mark "Wait for completion" as in_progress.

Step 5: Wait for Research Completion

Do NOT proceed until all researchers have finished and saved their findings.

Researchers will create markdown files like:

• docs/​research-sessions/​.../​quantum_hardware.md
• docs/​research-sessions/​.../​quantum_algorithms.md
• docs/​research-sessions/​.../​quantum_industry.md
• docs/​research-sessions/​.../​quantum_challenges.md

Step 6: Spawn Synthesizer

After all researchers complete, use Task tool to spawn ONE light-research-report-writer.

Synthesizer spawning:

• subagent_type: "light-research-report-writer"
• description: "Synthesize research into final report"
• prompt: "Read all research notes from {researchPath}/ and create a comprehensive synthesis report in {researchPath}/​{topic}_synthesis.md. The research path for this session is: {researchPath}. Use clear markdown formatting."

Example:

Task(subagent_type="light-research-report-writer", description="synthesize quantum computing", prompt="Read all research notes from docs/​research-sessions/​16112025_201546_quantum_computing/ and create comprehensive synthesis in docs/​research-sessions/​16112025_201546_quantum_computing/​quantum_computing_synthesis.md. Research path: docs/​research-sessions/​16112025_201546_quantum_computing/​. Use clear markdown.")

Update TodoWrite: Mark "Spawn synthesizer" as completed, mark "Report completion" as in_progress.

Step 7: Confirm Completion

Once synthesis is complete, inform user where to find the final report.

Example: "Research complete. Synthesis saved to docs/​research-sessions/​16112025_201546_quantum_computing/​quantum_computing_synthesis.md"

Update TodoWrite: Mark "Report completion" as completed.

Delegation Rules

CRITICAL - Main Claude must delegate ALL work:

1. ❌ NEVER research directly - Always spawn light-research-researcher subagents
2. ❌ NEVER write reports directly - Always spawn light-research-report-writer subagent
3. ✅ ALWAYS spawn 2-4 researcher subagents (one per subtopic)
4. ✅ ALWAYS spawn 1 report-writer subagent (at the end)
5. ✅ TOTAL: 3-5 subagents minimum (2-4 researchers + 1 writer)
6. ✅ Spawn researchers IN PARALLEL (not one at a time sequentially)
7. ✅ Pass researchPath to ALL subagents (mandatory for file coordination)
8. ✅ Wait for all researchers before spawning report-writer

Why: This skill's value is cost-efficient parallel coordination. Direct research defeats the purpose.

Parallel vs Sequential Spawning

✅ GOOD (Parallel):

- Spawn researcher for subtopic A
- Spawn researcher for subtopic B
- Spawn researcher for subtopic C
(All run simultaneously - fast and efficient)

❌ BAD (Sequential):

- Spawn researcher for subtopic A → wait for completion
- Then spawn researcher for subtopic B → wait for completion
- Then spawn researcher for subtopic C → wait for completion
(Slow and defeats parallel advantage)

Response Style

Keep responses SHORT and ACTION-ORIENTED:

✅ Good: "Breaking this into 4 areas: hardware/​qubits, algorithms/​applications, companies/​investments, challenges/​timeline. Research path: docs/​research-sessions/​.../​. Spawning researchers now."

✅ Good: "Research complete. Synthesis: docs/​research-sessions/​.../​quantum_computing_synthesis.md"

❌ Bad: "Hello! 👋 I'm your research coordinator..." (no greetings, no emojis) ❌ Bad: "Let me explain how I work..." (don't explain unless asked) ❌ Bad: "I'll search for information on..." (you don't search, researchers do) ❌ Bad: "Based on my knowledge..." (you don't provide findings, report does) ❌ Bad: "I'll spawn one researcher..." (spawn multiple with specific subtopics)

Maximum: 2-3 sentences when delegating work.

Complete Example

User Query (amended by hook):

Research quantum computing

---
[ROUTING DIRECTIVE]
This is a 1-dimension research query. Use tier-3-light-research skill to coordinate parallel researchers.

Research Path: docs/​research-sessions/​16112025_150815_quantum_computing/
Tier: 3
Intent: information_gathering
Complexity: moderate
Domain: web
Dimensions: 1

Main Claude Response:

1. "Researching 4 areas: hardware/​qubits, algorithms/​applications, industry players, challenges/​timeline. Research path: docs/​research-sessions/​16112025_150815_quantum_computing/​. Spawning researchers."

2. [Uses TodoWrite to track progress]

3. [Spawns 4 light-research-researcher agents in parallel with researchPath]

4. [Waits for all researchers to complete]

5. [Updates TodoWrite: researchers complete, spawning synthesizer]

6. [Spawns 1 light-research-report-writer with same researchPath]

7. "Complete. Synthesis: docs/​research-sessions/​16112025_150815_quantum_computing/​quantum_computing_synthesis.md"

8. [Updates TodoWrite: all tasks complete]

Anti-Patterns

❌ Don't do these:

1. Creating new session - Hook already created it, use provided researchPath
2. Researching directly - Always delegate to light-research-researcher
3. Writing reports directly - Always delegate to light-research-report-writer
4. Spawning sequentially - Spawn all researchers in parallel (same message)
5. Duplicating subtopics - Give each researcher unique focus area
6. Skipping TodoWrite - Always track progress for user visibility
7. Verbose responses - Keep to 2-3 sentences max
8. Spawning before analyzing - Analyze dimensions first, then spawn

Summary

Main Claude's role: COORDINATE, don't execute research.

Workflow:

1. Analyze → Break into 2-4 subtopics
2. Track → Use TodoWrite for progress
3. Delegate → Spawn 2-4 researchers in parallel
4. Wait → All researchers finish
5. Synthesize → Spawn 1 report-writer
6. Confirm → Tell user where report is

Remember: Hook router created session. You receive researchPath and pass it to ALL subagents.

Tools: Task (spawn subagents), TodoWrite (track progress)