Deep research

Deep Research

Conduct comprehensive, multi-source research on complex topics. Systematically gather, evaluate, and synthesize information into structured reports with proper citations.

When to Use

• User needs thorough research on a complex topic
• User asks "research this," "find out about," or "do a deep dive on"
• User needs a literature review, market analysis, or technology evaluation
• User wants to understand a topic from multiple angles with cited sources
• User needs to verify claims or compare conflicting information

When NOT to Use

• Simple factual lookups (just use web-search directly)
• Searching within the user's own codebase (use grep/glob)
• Looking up Replit-specific features (use replit-docs skill)
• Product recommendations without research depth (use a more specific skill)

Research Architecture

This skill follows a tree-like exploration pattern inspired by leading open-source research tools:

• GPT Researcher (github.com/assafelovic/gpt-researcher, ~17k stars) -- uses "plan and execute" with parallel sub-question research
• STORM (github.com/stanford-oval/storm, ~18k stars) -- Stanford's perspective-guided research that simulates multiple expert viewpoints
• open_deep_research (github.com/langchain-ai/open_deep_research) -- LangChain's iterative search-and-synthesize approach

The core pattern: decompose the question -> search broadly -> read deeply -> identify gaps -> refine queries -> synthesize with citations.

Methodology

Phase 1: Scope Definition

Before starting research, clearly define:

• Research question: What specific question(s) are you answering?
• Scope boundaries: What is in/out of scope?
• Depth level: Overview, moderate analysis, or exhaustive deep-dive?
• Output expectations: Report format, length, audience

Phase 2: Parallel Source Discovery via Subagents

Decompose the topic into 5 distinct focus areas, then launch 5 research subagents in parallel using startAsyncSubagent. Each subagent gets a specific focus area and set of search terms, searches independently, and returns its findings with citations.

How to decompose: After the broad landscape search in Phase 1, identify 5 non-overlapping angles. For example, researching "state of electric vehicles 2026" might decompose into:

1. Market & Competition — market share, sales figures, manufacturer rankings
2. Technology — battery chemistry, charging standards, range improvements
3. Policy & Regulation — government incentives, emissions mandates, trade tariffs
4. Infrastructure — charging network growth, grid capacity, urban vs rural
5. Consumer & Economics — total cost of ownership, resale value, adoption demographics

Launch all 5 in parallel:

// Launch 5 research subagents simultaneously
await startAsyncSubagent({
    task: `Research FOCUS AREA 1: [Market & Competition]

Topic context: [brief description of the overall research question]

Your job: Search for information specifically about [focus area]. Run at least 3-4 webSearch queries with different angles:
- [specific search term 1]
- [specific search term 2]
- [specific search term 3]
- [specific search term 4]

For the most promising results, use webFetch to read the full article.

Return your findings as a structured summary with:
- Key facts and data points (with source URLs)
- Notable claims that need cross-referencing
- Gaps or unanswered questions
- At least 5 distinct sources with URLs`
});

// Repeat for focus areas 2-5 with their own tailored search terms
await startAsyncSubagent({ task: `Research FOCUS AREA 2: [Technology] ...` });
await startAsyncSubagent({ task: `Research FOCUS AREA 3: [Policy & Regulation] ...` });
await startAsyncSubagent({ task: `Research FOCUS AREA 4: [Infrastructure] ...` });
await startAsyncSubagent({ task: `Research FOCUS AREA 5: [Consumer & Economics] ...` });

// Wait for all subagents to complete
const results = await waitForBackgroundTasks();

Each subagent should:

• Run 3-4 webSearch queries with different phrasings and angles
• Use webFetch on the 2-3 most relevant results to extract detailed data
• Return structured findings with source URLs
• Flag any claims that conflict with other results

This approach gathers 25+ distinct sources across 5 focus areas simultaneously, producing far more comprehensive coverage than sequential searching.

After collecting all subagent results, proceed to Phase 3 to evaluate and cross-reference.

Phase 3: Source Evaluation

Assess each source for credibility:

• Authority: Who published it? What are their credentials?
• Currency: When was it published? Is the information still current?
• Objectivity: Is there obvious bias? Is it sponsored content?
• Accuracy: Can claims be cross-referenced with other sources?
• Coverage: Does it cover the topic in sufficient depth?

Use webFetch to read full articles from the most promising search results.

Phase 4: Information Synthesis

Organize findings thematically (what separates deep research from simple search):

• Group related findings across sources
• Identify areas of consensus and disagreement
• Note gaps in available information -- conduct follow-up searches to fill them
• Cross-reference critical claims across at least 2-3 independent sources
• Build a narrative that answers the research question
• Distinguish between established facts, expert opinions, and speculation
• Draw connections between sources that reveal patterns not visible in any single source

Phase 5: Report Writing

Structure the final report clearly:

• Lead with the most important findings
• Support claims with specific sources
• Acknowledge limitations and uncertainties
• Provide actionable recommendations where appropriate

Output Format

Research Report Structure

# [Research Topic]

## Executive Summary
[2-3 paragraph overview of key findings and conclusions]

## Background
[Context needed to understand the topic]

## Key Findings

### Finding 1: [Theme]
[Detailed analysis with source citations]

### Finding 2: [Theme]
[Detailed analysis with source citations]

### Finding 3: [Theme]
[Detailed analysis with source citations]

## Analysis
[Cross-cutting analysis, patterns, implications]

## Limitations
[What couldn't be determined, data gaps, caveats]

## Recommendations
[Actionable next steps based on findings]

## Sources
[Numbered list of all sources with URLs]

Best Practices

1. Cast a wide net first, then narrow -- start with broad searches before diving into specifics
2. Cross-reference critical claims -- never rely on a single source for important facts
3. Cite everything -- every factual claim should trace back to a source
4. Note disagreements -- when sources conflict, present both sides and analyze why
5. Timestamp your research -- note when the research was conducted, as information changes
6. Separate facts from analysis -- clearly distinguish between what sources say and your interpretation

Example Workflow

// Phase 1: Broad landscape search to identify focus areas
const overview = await webSearch({ query: "state of electric vehicle market 2026" });

// Phase 2: Launch 5 parallel research subagents
await startAsyncSubagent({
    task: `Research EV Market & Competition: search for "EV market share by manufacturer 2025 2026",
    "electric vehicle sales global rankings", "Tesla BYD market share comparison".
    Use webFetch on best results. Return findings with source URLs.`
});
await startAsyncSubagent({
    task: `Research EV Battery Technology: search for "solid state battery progress 2026",
    "EV battery cost per kwh trend", "lithium iron phosphate vs NMC comparison".
    Use webFetch on best results. Return findings with source URLs.`
});
await startAsyncSubagent({
    task: `Research EV Policy & Regulation: search for "EV tax credit policy 2026",
    "emissions regulations electric vehicles", "EV tariffs trade policy".
    Use webFetch on best results. Return findings with source URLs.`
});
await startAsyncSubagent({
    task: `Research EV Charging Infrastructure: search for "EV charging network growth statistics",
    "NACS vs CCS charging standard adoption", "fast charging stations by country".
    Use webFetch on best results. Return findings with source URLs.`
});
await startAsyncSubagent({
    task: `Research EV Consumer Economics: search for "EV total cost of ownership vs gas 2026",
    "electric vehicle resale value trends", "EV adoption demographics income".
    Use webFetch on best results. Return findings with source URLs.`
});

// Collect all results
const results = await waitForBackgroundTasks();

// Phase 3-5: Evaluate sources, cross-reference claims, synthesize into structured report
// Write comprehensive report with all findings and citations from all 5 subagents

Limitations

• Cannot access paywalled academic journals or subscription databases
• Cannot access social media content (LinkedIn, Twitter, Reddit)
• Web sources may have varying levels of reliability
• Research is a snapshot in time -- findings may change
• Cannot conduct primary research (surveys, interviews, experiments)