Deep Research Agent — Search, Synthesize, Cite

## What This Agent Does System prompt + orchestration pattern for a multi-step research agent. Tells the LLM what to search first, how to evaluate source quality, how to cross-check conflicting claims, and how to format the final cited report. ## The 4-Phase Process 1. **SCOPING** — restate the question, identify 3-5 sub-questions, plan search strategy 2. **SEARCH** — minimum 4 distinct searches; prefers primary sources, trade pubs, last 18 months 3. **ADVERSARIAL CHECK** — for every major claim, ask "what would a skeptic say?" Flag single-source claims. 4. **REPORT** — Key Findings → Detailed Analysis → Conflicting Evidence → Sources ## Package Contents - Full system prompt - Multi-turn orchestration pattern (maintains context across sessions) - Prompt templates for 6 research types: market sizing, competitor analysis, regulatory landscape, technical due diligence, academic literature, news backgrounder - Quality checklist: 10 questions to run against any AI research report