Researcher Agent

The Researcher agent synthesizes information from source documents and research data to prepare context for chapter writing.

Overview

Property	Value
Role	Analyze and synthesize sources
Model	Opus (deep synthesis) / Sonnet (targeted)
Input	Source documents, research questions
Output	Structured markdown synthesis

Full Prompt Template

You are a research assistant. Your task is to gather and synthesize
information for a technical documentation project.

## RESEARCH OBJECTIVE
[Describe what information is needed]

## SOURCE DOCUMENTS
Analyze these files:
1. [file1] - [description]
2. [file2] - [description]

## RESEARCH QUESTIONS
Answer these specific questions:
1. [Question 1]
2. [Question 2]
3. [Question 3]

## OUTPUT FORMAT
Provide your findings in this structure:

### Executive Summary
[2-3 paragraphs summarizing key findings]

### Detailed Findings

#### Topic 1
[Findings with source citations]

#### Topic 2
[Findings with source citations]

### Data Tables
[Any relevant comparative data in markdown tables]

### Key Statistics
[Numbers, dates, percentages with sources]

### Recommended Citations
[BibTeX entries for sources that should be cited]

### Gaps Identified
[What information is missing or needs further research]

Example: Research Synthesis

Synthesize information about LLM security threats from provided sources.

SOURCE DOCUMENTS:
1. sources/owasp-llm-top10.pdf - OWASP LLM Top 10 vulnerabilities
2. sources/academic-papers/llm-security-survey.pdf - Academic survey
3. research-prompts/results/perplexity.md - Section 1.3 (Recent incidents)

RESEARCH QUESTIONS:
1. What are the top 10 LLM vulnerabilities according to OWASP?
2. What real-world incidents have occurred involving LLM attacks?
3. What defense mechanisms are currently available?
4. What are the current research gaps in LLM security?

Provide findings with specific citations and page numbers.
Include BibTeX entries for any new sources discovered.

Example: Web Research

Research current (2024-2025) statistics on [topic].

FOCUS AREAS:
1. Market size and growth projections
2. Major incidents and security breaches
3. Regulatory developments (EU AI Act, NIST)
4. Technology trends and adoption rates

REQUIREMENTS:
- Focus on 2024-2025 data
- Include source URLs for verification
- Provide specific numbers and dates
- Note confidence level of each claim

OUTPUT:
- Markdown formatted report
- Summary table of key statistics
- BibTeX entries for academic citations

Output Structure

Executive Summary

### Executive Summary

The analysis of provided sources reveals three key themes in
modern AI security: [theme 1], [theme 2], and [theme 3].

Source document A (thesis1.pdf) provides comprehensive coverage
of attack methodologies, while document B (thesis2.pdf) focuses
on defense mechanisms. The gap analysis indicates that [gap]
requires additional research from external sources.

The synthesis identifies 15 distinct threat categories across
all documents, with ransomware and supply chain attacks showing
the highest frequency of discussion.

Detailed Findings

### Detailed Findings

#### 1. Attack Taxonomy

The sources identify the following attack categories:

| Attack Type | Source | Page | Prevalence |
|-------------|--------|------|------------|
| Adversarial examples | thesis1.pdf | 23-35 | High |
| Data poisoning | thesis1.pdf | 45-52 | Medium |
| Model extraction | thesis2.pdf | 15-22 | Medium |
| Prompt injection | paper.pdf | 8-15 | High |

Key finding: Adversarial attacks dominate academic research
(cited in 85% of surveyed papers), while prompt injection
is more prevalent in industry reports.

#### 2. Defense Mechanisms

Defense strategies discussed across sources:

**thesis2.pdf (pages 60-85):**
- Input validation and sanitization
- Output filtering
- Rate limiting and anomaly detection

**paper.pdf (pages 20-30):**
- Adversarial training
- Ensemble methods
- Runtime monitoring

Key Statistics

### Key Statistics

| Metric | Value | Source | Page |
|--------|-------|--------|------|
| Global ransomware cost | $20B (2024) | report.pdf | 12 |
| LLM adoption rate | 67% enterprises | survey.pdf | 5 |
| Security incidents | +43% YoY | perplexity.md | - |
| AI security market | $185B (2024) | perplexity.md | - |

Recommended Citations

### Recommended Citations

```bibtex
@article{smith2024adversarial,
  author = {Smith, John and Jones, Jane},
  title = {Adversarial Attacks on Modern ML Systems},
  journal = {IEEE TIFS},
  year = {2024},
  volume = {19},
  pages = {1234-1248}
}

@inproceedings{wilson2024llm,
  author = {Wilson, Alice},
  title = {LLM Security: A Comprehensive Survey},
  booktitle = {USENIX Security},
  year = {2024}
}

### Gaps Identified

```markdown
### Gaps Identified

The following topics require additional research:

| Gap | Mentioned In | Suggested Source |
|-----|--------------|------------------|
| EU AI Act compliance | thesis1.pdf (p.90) | Perplexity |
| 2025 threat predictions | Not covered | Perplexity |
| Academic defense papers | Partial | Scopus AI |
| Industry best practices | Not covered | Perplexity |

**Recommended next steps:**
1. Run Perplexity query for EU AI Act compliance requirements
2. Search Scopus AI for defense mechanism papers 2024-2025
3. Research industry frameworks (NIST AI RMF)

Source-to-Chapter Mapping

After synthesis, create a mapping for chapter writers:

# Source-to-Chapter Mapping

## Chapter 2: Background
- thesis1.pdf, pages 10-20: Historical context
- paper.pdf, pages 1-5: Problem definition
- perplexity.md, Section 1: Current landscape

## Chapter 3: Attack Vectors
- thesis1.pdf, pages 25-50: Attack taxonomy
- thesis2.pdf, pages 10-30: Real-world examples
- scopus.md: Academic attack research

## Chapter 4: Defense Mechanisms
- thesis2.pdf, pages 55-85: Defense framework
- paper.pdf, pages 20-35: Mitigation strategies
- perplexity.md, Section 3: Industry solutions

## Chapter 5: Implementation
- thesis1.pdf, pages 60-75: System architecture
- thesis2.pdf, pages 90-110: Case study

Parallel Source Analysis

When analyzing multiple sources simultaneously:

Quality Criteria

Research output should include:

• [ ] Executive summary covering all sources
• [ ] Specific page numbers for claims
• [ ] Data tables with source attribution
• [ ] Statistics with confidence levels
• [ ] BibTeX entries for new citations
• [ ] Identified gaps for external research
• [ ] Chapter mapping recommendations

Next Steps

• Chapter Writer - Use synthesis for writing
• Reviewer - Review written content
• NotebookLM Prompts - External source synthesis