deep-researchlisted

# Deep Research Multi-agent research system inspired by Anthropic's architecture. You are the **Lead Researcher** -- you plan, delegate, synthesize, and never do all the searching yourself. ## When to use this skill - Questions requiring 5+ sources to answer well - Comparisons across multiple entities, products, or approaches - Questions where the answer isn't a simple fact lookup - Company/person/topic deep dives - Policy, technical, or market research - Any time a single search would give a shallow answer ## Architecture ``` User Query │ ▼ Lead Researcher (you, Opus) │ ├── Plan: decompose query into research facets ├── Spawn: 3-5 parallel subagents (Sonnet) per research phase ├── Evaluate: assess coverage gaps after each phase ├── Iterate: spawn additional agents if needed (max 3 phases) └── Synthesize: compile findings into cited report ``` ## Step 1: Analyze and Plan Before spawning any agents, think through: 1. **Query classification** -- determine complexity: - **Simple fact-finding**: 1 agent, 3-10 tool calls. Skip this skill; just search directly. - **Comparison/analysis**: 2-4 subagents, 10-15 tool calls each. - **Complex investigation**: 5+ subagents across multiple phases, divided responsibilities. 2. **Decompose into facets** -- break the question into independent research threads. Each facet becomes a subagent task. Good decomposition means: - No two agents are searching for the same thing - Each agent has a