deep-research

name	deep-research
description	Multi-source parallel research via sub-agents

Deep Research

Conduct thorough multi-source research by spawning parallel sub-agents.

When to Use

• User asks a complex question requiring multiple perspectives or sources
• User says "research this", "deep dive", or "find out everything about"
• Topic requires cross-referencing multiple sources for accuracy

Procedure

1. Decompose the question into 2-4 independent research angles
2. Spawn sub-agents — use spawnAgent for each angle with:
   • Clear, focused task description
   • tools: ["webSearch", "extractUrl"] at minimum
   • Relevant context from the user's question
3. Wait for results — all sub-agents run concurrently
4. Synthesize findings:
   • Cross-reference facts across sources
   • Identify consensus vs. conflicting information
   • Note gaps or areas with limited data
5. Present the synthesis — organized by theme, not by source
6. Store key findings in memory for follow-up questions

Guidelines

• Use 2-4 sub-agents — more than 4 rarely adds value
• Give each sub-agent a distinct research angle, not overlapping queries
• Prefer authoritative sources (official docs, research papers, established publications)
• Clearly distinguish facts from opinions and speculation
• Include source attribution for key claims
• If one sub-agent fails, work with what the others found