Deep Research

Multi-source research engine that replaces hours of manual research with one prompt. Combines semantic search, web search, technical documentation, and community discussions to produce comprehensive, properly-cited research reports.

When to Use This Skill

• Deep investigation of any topic (technical, market, academic, competitive)
• Understanding a new technology, framework, or concept
• Market research and competitive analysis
• Academic research with proper citations
• Technical evaluation of tools, libraries, or architectures
• Trend analysis and emerging topic exploration
• Building knowledge base entries with verified sources

What This Skill Does

1. Multi-Source Intelligence Gathering

   • Exa: Neural/semantic search for papers, documentation, and code
   • Brave Search: General web for news, articles, and current information
   • DeepWiki: GitHub repositories and technical documentation
   • HackerNews: Tech community discussions, opinions, and real-world experiences

2. Intelligent Source Routing

   • Automatically prioritizes sources based on research mode
   • Cross-references findings across sources for validation
   • Identifies consensus and disagreements between sources

3. Structured Synthesis

   • Produces organized reports with clear sections
   • Includes proper citations and source links
   • Highlights key findings and actionable insights

4. Knowledge Base Integration

   • Saves research as notes to Supabase items table
   • Supports multi-session research continuation
   • Tags and categorizes for easy retrieval

Research Modes

Technical Mode (Default for tech topics)

Prioritizes: Documentation, code examples, papers, GitHub repos Best for: Evaluating frameworks, understanding APIs, learning new technologies

Market Mode

Prioritizes: News, analysis, trends, company information Best for: Industry research, startup evaluation, market sizing

Academic Mode

Prioritizes: Papers, citations, scholarly sources, research findings Best for: Literature reviews, scientific topics, formal research

Competitive Mode

Prioritizes: Company websites, product comparisons, pricing, reviews Best for: Competitive analysis, vendor evaluation, buy vs. build decisions

Output Formats

Quick Summary (2-3 paragraphs)

Fast overview for initial understanding or time-constrained situations.

Deep Dive (Full report with sections)

Comprehensive analysis with executive summary, detailed findings, and recommendations.

Annotated Bibliography

Source-focused output listing key resources with annotations and relevance notes.

Key Takeaways Only

Bullet-point format highlighting the most important findings.

How to Use

Basic Research

/deep-research [topic]

With Mode Specified

/deep-research --mode technical "WebSocket scaling patterns"
/deep-research --mode market "AI code assistant market 2024"
/deep-research --mode academic "transformer architecture attention mechanisms"
/deep-research --mode competitive "Supabase vs Firebase vs PlanetScale"

With Output Format

/deep-research --format quick "Rust for backend development"
/deep-research --format deep "Vector databases comparison"
/deep-research --format bibliography "LLM fine-tuning techniques"
/deep-research --format takeaways "GraphQL vs REST in 2024"

With Save Option

/deep-research --save "Kubernetes autoscaling strategies"

Continue Previous Research

/deep-research --continue [topic or research-id]

Instructions

When a user requests deep research on any topic:

Step 1: Parse Request and Determine Parameters

Extract from user request:

• Topic: The subject to research
• Mode: technical (default) | market | academic | competitive
• Format: quick | deep (default) | bibliography | takeaways
• Save: Whether to save to items table (default: ask after completion)
• Continue: Whether this continues previous research

If parameters are ambiguous, infer based on topic:

• Code/framework/API topics -> technical mode
• Company/industry/pricing topics -> market mode
• Scientific/research topics -> academic mode
• Product comparison topics -> competitive mode

Step 2: Execute Multi-Source Search

Perform searches in parallel using MCP tools based on mode priority:

Technical Mode Search Pattern

1. Exa (highest priority):
   - Search: "[topic] documentation tutorial"
   - Search: "[topic] best practices examples"
   - Search: "[topic] architecture design patterns"

2. DeepWiki:
   - Search for relevant GitHub repositories
   - Extract README content and documentation
   - Find code examples and implementations

3. HackerNews:
   - Search: "[topic]" for community discussions
   - Look for experience reports and gotchas
   - Find real-world usage stories

4. Brave Search:
   - Search: "[topic] guide 2024"
   - Search: "[topic] comparison alternatives"

Market Mode Search Pattern

1. Brave Search (highest priority):
   - Search: "[topic] market analysis 2024"
   - Search: "[topic] industry trends"
   - Search: "[topic] funding news startups"

2. Exa:
   - Search: "[topic] market research report"
   - Search: "[topic] industry analysis"

3. HackerNews:
   - Search for discussions about the market/industry
   - Find insider perspectives and predictions

4. DeepWiki:
   - If relevant, find technical details about products

Academic Mode Search Pattern

1. Exa (highest priority):
   - Search: "[topic] research paper"
   - Search: "[topic] study findings"
   - Search: "[topic] academic review"
   - Use semantic search for related concepts

2. Brave Search:
   - Search: "[topic] research university"
   - Search: "[topic] peer reviewed"

3. HackerNews:
   - Search for discussions of papers/research
   - Find critiques and implementations

Competitive Mode Search Pattern

1. Brave Search (highest priority):
   - Search: "[product A] vs [product B]"
   - Search: "[topic] pricing comparison"
   - Search: "[topic] reviews pros cons"

2. Exa:
   - Search company documentation
   - Find feature comparisons

3. HackerNews:
   - Search: "[topic]" for user experiences
   - Find migration stories and recommendations

4. DeepWiki:
   - Compare GitHub activity and community
   - Analyze technical implementations

Step 3: MCP Tool Usage

Exa Search (Neural/Semantic)

Use the Exa MCP tools for semantic search:

• exa_search: For finding relevant documents with neural search
• exa_find_similar: For finding documents similar to a good source
• exa_get_contents: For extracting full content from URLs

Example query patterns:

exa_search: query="[topic] comprehensive guide", num_results=10, type="auto"
exa_search: query="[topic]", category="research paper", num_results=5
exa_find_similar: url="[good_source_url]", num_results=5

Brave Search (General Web)

Use the Brave Search MCP tool:

• brave_web_search: For general web search with freshness options

Example:

brave_web_search: query="[topic] 2024", count=10, freshness="pw" (past week)
brave_web_search: query="[topic] news", count=5, freshness="pd" (past day)

DeepWiki (GitHub/Technical Docs)

Use the DeepWiki MCP tool:

• read_wiki_structure: Get structure of a GitHub repo’s documentation
• read_wiki_contents: Read specific documentation pages
• ask_question: Ask questions about a repository

Example:

read_wiki_structure: repoPath="facebook/react"
read_wiki_contents: repoPath="vercel/next.js", filePath="README.md"
ask_question: repoPath="[owner/repo]", question="[specific question about implementation]"

HackerNews (Community Discussions)

Use the HN MCP tool:

• hn_search: Search HackerNews stories and comments
• hn_get_story: Get full story with comments

Example:

hn_search: query="[topic]", tags="story", numericFilters="points>50"
hn_get_story: id=[story_id]

Step 4: Synthesize Findings

After gathering sources, synthesize into coherent findings:

1. Identify Key Themes

   • What concepts appear across multiple sources?
   • What are the main approaches/solutions discussed?
   • What controversies or debates exist?

2. Validate Information

   • Cross-reference facts across sources
   • Note where sources disagree
   • Identify most authoritative sources

3. Extract Actionable Insights

   • What decisions can be made from this research?
   • What are the clear recommendations?
   • What requires further investigation?

4. Organize by Relevance

   • Lead with most important findings
   • Group related information
   • Separate facts from opinions

Step 5: Generate Report

Based on requested format:

Quick Summary Format

## Quick Summary: [Topic]

**Research Date:** [Date]
**Mode:** [technical/market/academic/competitive]
**Sources Consulted:** [count]

[2-3 paragraph synthesis of key findings]

**Key Points:**
- [Most important finding 1]
- [Most important finding 2]
- [Most important finding 3]

**Top Sources:**
1. [Source 1 title](url) - [why relevant]
2. [Source 2 title](url) - [why relevant]

Deep Dive Format

# Deep Research Report: [Topic]

**Research Date:** [Date]
**Mode:** [Mode]
**Sources Analyzed:** [count]

---

## Executive Summary

[3-4 paragraph overview of key findings, conclusions, and recommendations]

---

## Background & Context

[What is this topic? Why does it matter? What's the current state?]

---

## Key Findings

### Finding 1: [Theme/Topic]

[Detailed explanation with citations]

**Evidence:**
- [Source 1] states that... [1]
- According to [Source 2]... [2]
- Community sentiment on HN suggests... [3]

### Finding 2: [Theme/Topic]

[Detailed explanation with citations]

### Finding 3: [Theme/Topic]

[Detailed explanation with citations]

---

## Analysis & Insights

### Consensus Points
[What do most sources agree on?]

### Areas of Debate
[Where do sources disagree? What are the different camps?]

### Gaps in Available Information
[What couldn't be determined? What needs more research?]

---

## Practical Implications

### Recommendations
1. [Actionable recommendation 1]
2. [Actionable recommendation 2]
3. [Actionable recommendation 3]

### Next Steps for Further Research
- [Specific area to explore]
- [Question that remains unanswered]

---

## Sources & Citations

### Primary Sources (High Authority)
1. [Title](URL) - [Brief description]
2. [Title](URL) - [Brief description]

### Supporting Sources
3. [Title](URL) - [Brief description]
4. [Title](URL) - [Brief description]

### Community Discussions
5. [HN Discussion Title](URL) - [Key insight from discussion]

---

## Appendix: Raw Source Notes

<details>
<summary>Click to expand source excerpts</summary>

### From [Source 1]
> [Key quote or excerpt]

### From [Source 2]
> [Key quote or excerpt]

</details>

Annotated Bibliography Format

# Annotated Bibliography: [Topic]

**Research Date:** [Date]
**Total Sources:** [count]

---

## Essential Reading (Must-Read)

### 1. [Title](URL)
**Type:** [Documentation/Paper/Article/Discussion]
**Authority:** [High/Medium/Low]
**Published:** [Date if known]

**Summary:** [2-3 sentences on what this source covers]

**Key Contributions:**
- [What unique insight does this provide?]
- [What questions does it answer?]

**Relevance to Topic:** [Why this matters for understanding [topic]]

---

### 2. [Title](URL)
[Same structure]

---

## Supplementary Sources (Good to Know)

### 3. [Title](URL)
[Abbreviated annotation]

---

## Community Perspectives

### [HN Thread Title](URL)
**Discussion Quality:** [High/Medium]
**Key Voices:** [Notable commenters and their expertise]

**Best Comments:**
- [Username]: "[Key quote]"
- [Username]: "[Key quote]"

**Consensus View:** [What did the community generally agree on?]

Key Takeaways Format

## Key Takeaways: [Topic]

**Research Date:** [Date]
**Confidence Level:** [High/Medium/Low based on source quality]

### The Bottom Line
[1-2 sentence summary of the most important conclusion]

---

### Critical Facts
- [Fact 1 with source]
- [Fact 2 with source]
- [Fact 3 with source]

### Main Recommendations
- [Recommendation 1]
- [Recommendation 2]
- [Recommendation 3]

### Watch Out For
- [Common pitfall or misconception]
- [Risk or consideration]

### Best Resources for More
1. [Most recommended source](url)
2. [Second best](url)

---

**Time to read sources for full understanding:** ~[X] hours

Step 6: Save to Knowledge Base (If Requested)

Save the research to the items table:

SUPABASE_URL="https://mlzbjnjkopuzoiobinpz.supabase.co"
SUPABASE_KEY="$SUPABASE_SERVICE_KEY"

curl -X POST "$SUPABASE_URL/rest/v1/items" \
  -H "apikey: $SUPABASE_KEY" \
  -H "Authorization: Bearer $SUPABASE_KEY" \
  -H "Content-Type: application/json" \
  -H "Prefer: return=representation" \
  -d '{
    "telegram_user_id": [USER_ID],
    "summary": "[Topic] - [Mode] research report",
    "content": "[Full research report in markdown]",
    "category": "research",
    "status": "active",
    "tags": ["research", "[mode]", "[topic-tags]"],
    "metadata": {
      "research_type": "deep-research",
      "mode": "[mode]",
      "format": "[format]",
      "sources_count": [count],
      "sources": [
        {"title": "...", "url": "...", "type": "..."},
        ...
      ],
      "research_date": "[ISO timestamp]",
      "can_continue": true
    }
  }'

Step 7: Support Continuation

For --continue flag:

1. Query previous research from items table:

curl -s "$SUPABASE_URL/rest/v1/items?category=eq.research&summary=ilike.*[topic]*&telegram_user_id=eq.[USER_ID]&order=created_at.desc&limit=1" \
  -H "apikey: $SUPABASE_KEY"

2. Load previous findings into context
3. Search for new/updated information
4. Merge with previous research
5. Update the saved item or create new version

Examples

Example 1: Technical Research

User: /deep-research WebSocket scaling patterns

Process:

1. Mode: technical (inferred from topic)
2. Exa: Search for WebSocket scaling documentation, architecture patterns
3. DeepWiki: Find Socket.io, ws library implementations
4. HackerNews: Search for real-world scaling stories
5. Brave: Search for recent articles on WebSocket at scale

Output: Deep dive report covering:

• Connection management strategies
• Horizontal scaling approaches (Redis pub/sub, etc.)
• Real-world case studies from HN discussions
• Library comparisons with code examples

Example 2: Market Research

User: /deep-research --mode market "AI code assistant market 2024"

Process:

1. Mode: market (explicit)
2. Brave: Search for market analysis, funding news, industry trends
3. Exa: Find market research reports
4. HackerNews: Community sentiment on various tools

Output: Market analysis covering:

• Market size and growth projections
• Key players and funding rounds
• Competitive landscape
• Community adoption and preferences

Example 3: Academic Research

User: /deep-research --mode academic --format bibliography "attention mechanisms in transformers"

Process:

1. Mode: academic
2. Format: bibliography
3. Exa: Semantic search for papers on attention mechanisms
4. Brave: Search for surveys and reviews

Output: Annotated bibliography with:

• Original attention paper citation
• Key follow-up research
• Survey papers
• Implementation resources

Example 4: Competitive Analysis

User: /deep-research --mode competitive "Supabase vs Firebase vs PlanetScale"

Process:

1. Mode: competitive
2. Brave: Search for comparisons, pricing, reviews
3. DeepWiki: Analyze GitHub repos, documentation quality
4. HackerNews: User migration stories, recommendations
5. Exa: Find technical deep-dives on each

Output: Comparison report covering:

• Feature matrix
• Pricing comparison
• Community sentiment
• Migration considerations
• Recommendations by use case

Example 5: Quick Research

User: /deep-research --format quick "Bun vs Node.js"

Output:

## Quick Summary: Bun vs Node.js

**Research Date:** [Date]
**Mode:** technical
**Sources Consulted:** 12

Bun is a new JavaScript runtime that aims to be a drop-in replacement for Node.js with significantly better performance. Benchmarks consistently show Bun 2-4x faster for most operations, particularly for bundling and package installation. However, Node.js has a much more mature ecosystem with better production stability and tooling support.

The community is cautiously optimistic about Bun. HackerNews discussions reveal many developers using it for development but sticking with Node.js for production. Key concerns include compatibility issues with some npm packages and the smaller community for troubleshooting.

**Key Points:**
- Bun is 2-4x faster in benchmarks but Node.js is more battle-tested
- Package compatibility is ~95% but edge cases exist
- Recommended: Use Bun for dev/scripts, Node.js for production (for now)

**Top Sources:**
1. [Bun Documentation](https://bun.sh) - Official benchmarks and compatibility info
2. [HN Discussion: Bun 1.0](https://news.ycombinator.com/...) - Real-world experiences

Best Practices

Research Quality

• Always cross-reference claims across multiple sources
• Note the date of sources – prefer recent for fast-moving topics
• Distinguish between facts and opinions
• Cite specific sources for key claims

Source Prioritization

• Official documentation > Blog posts > Forum discussions
• Recent sources > Older sources (for tech topics)
• High-engagement HN discussions often surface real issues
• Academic papers for foundational understanding

Time Management

• Quick format: 5-10 minutes of research
• Deep format: 15-30 minutes of research
• Academic mode: May require 30+ minutes for thorough coverage

Saving Research

• Save extensive research to avoid re-doing work
• Use tags for easy retrieval
• Enable continuation for ongoing research projects

Error Handling

No results from a source:

• Note the gap in the report
• Suggest why this might be (niche topic, new technology, etc.)
• Recommend alternative research approaches

Conflicting information:

• Present both perspectives
• Note which sources are more authoritative
• Highlight where community consensus differs from official docs

Rate limits on MCP tools:

• Prioritize highest-value searches
• Cache results for continuation
• Inform user if research is incomplete

User Interest Profile (Prioritize These Topics)

When conducting research for the user, apply extra depth and attention to these interest areas:

AI & Development (High Priority):

• Claude Code, MCP servers, AI agents, LLMs, AI skills/tooling
• Autonomous AI experiments, AI vs AI competitions
• Agentic workflows, tool use patterns

Crypto & DeFi (High Priority):

• Polymarket, prediction markets, Kalshi
• Hyperliquid, MEV, on-chain analytics
• DeFi protocols, yield strategies, liquidity
• Ethereum ecosystem, Solana developments

Emerging Tech:

• Quantum computing breakthroughs and applications
• Edge compute, distributed systems
• Hardware innovations, IoT, WiFi protocols
• WebAuthn, passkeys, authentication systems

Specialized Domains:

• Legal tech innovations
• Tax tech and compliance automation
• Blockchain infrastructure and scalability

Research Prioritization Rules:

• For technical mode on these topics: Go deeper, find more sources
• For market mode: Include prediction market angles where relevant
• Cross-domain: Flag interesting intersections (AI+crypto, quantum+security, etc.)
• Source preference: Prioritize Substack authors, crypto researchers, AI labs
• HackerNews: Search for discussions from practitioners in these areas

Related Skills

• content-research-writer: For creating content based on research
• summarize: For quick URL summaries
• note: For quick capture of research findings
• market-analyst: For prediction market analysis