deep-research

Deep Research Skill

Overview

This skill conducts systematic, multi-iteration research on any topic. Inspired by OpenAI's Deep Research and open-source implementations, it uses progressive exploration to build comprehensive understanding.

When to use: User asks for comprehensive research, investigation, deep analysis, or thorough exploration of a topic requiring multiple sources.

Research Methodology

Phase 1: Scope Definition

Before starting research, clarify:

1. Research Question: What exactly needs to be answered?

2. Key Aspects: What facets of the topic should be explored?

3. Parameters: Set depth and breadth based on complexity

   • Simple topic: depth=2, breadth=3
   • Moderate topic: depth=3, breadth=4 (default)
   • Complex topic: depth=4-5, breadth=5-6

4. Create Research Directory:

   ./research/[topic-slug]/
   ├── findings.md      # Accumulated findings per iteration
   ├── sources.md       # All sources with URLs
   └── report.md        # Final synthesized report
   

Phase 2: Iterative Research Loop

For each iteration (1 to depth):

Step 1: Generate Search Queries

Create targeted search queries based on:

• Original research question
• Current knowledge gaps (from previous iterations)
• Unexplored angles discovered in findings
• Contradictions requiring resolution

Query Generation Guidelines:

• Make queries specific, not generic
• Vary approach: definitional, comparative, recent news, expert sources
• Build on previous learnings - don't repeat covered ground
• Target entities, metrics, dates mentioned in findings
• Include domain-specific terminology discovered

Generate breadth queries per iteration.

Step 2: Execute Searches

For each query:

1. Use WebSearch to find relevant sources
2. Review search results for relevance and quality
3. For the most promising 2-3 results per query, use WebFetch to get full content

Source Quality Signals:

• Authoritative domains (.gov, .edu, major publications)
• Recent publication dates for current topics
• Expert authors or organizational sources
• Primary sources over aggregators