/deep-web-research

allowed-tools: Task, WebSearch, WebFetch, Write, Bash(gdate:*), Bash(pwd:*), Bash(fd:*), Bash(rg:*), Bash(git:*)
name: "Deep Web Research"
description: "Comprehensive parallel web research with multi-agent analysis and synthesis"
author: "wcygan"
tags: ["analyze","research"]
version: "1.0.0"
created_at: "2025-07-14T00:00:00Z"
updated_at: "2025-07-14T00:00:00Z"

Context

• Session ID: !`gdate +%s%N`
• Current directory: !`pwd`
• Project type: !`fd -e json -e toml -e xml . | rg "(deno\.json|package\.json|Cargo\.toml|pom\.xml|requirements\.txt)" | head -3 || echo "No project files detected"`
• Technology stack: !`fd -e json . | head -3 | xargs -I {} rg "(react|vue|angular|typescript|python|rust|go|java|kubernetes|docker)" {} | head -5 || echo "No stack detected"`
• Git context: !`git status --porcelain | head -3 2>/dev/null || echo "Not a git repository"`
• Current branch: !`git branch --show-current 2>/dev/null || echo "No git repository"`
• Research query: $ARGUMENTS

Your Task

Conduct comprehensive web research using parallel sub-agents for thorough coverage and analysis.

Think deeply about the optimal research strategy for this specific query, considering:

• Technical complexity and scope
• Time-sensitive vs. foundational information needs
• Target audience (beginner, intermediate, expert)
• Project context and practical application requirements

STEP 1: Research Strategy Initialization

• Parse research query: $ARGUMENTS
• Initialize session state: /tmp/deep-web-research-$SESSION_ID.json
• Determine research complexity (simple/moderate/complex)
• Generate research angles and perspectives
• Create agent delegation strategy

STEP 2: Multi-Perspective Query Generation

• Generate 5-7 search query variations:
• Current best practices and standards
• Real-world implementation experiences
• Recent trends and innovations
• Comparative analysis and alternatives
• Security and performance considerations
• Common pitfalls and lessons learned
• Enterprise vs. small-scale approaches

STEP 3: Parallel Research Execution

Deploy 5 research agents in parallel:

**Agent 1: Authoritative Sources Research**

• Task: "Research official documentation, standards, and authoritative best practices for: $ARGUMENTS"
• Focus: Official docs, RFC standards, vendor documentation
• Sources: 2-3 authoritative sources
• Output: /tmp/research-auth-$SESSION_ID.json

**Agent 2: Community Insights Research**

• Task: "Research community experiences, forums, and practical implementations for: $ARGUMENTS"
• Focus: Stack Overflow, Reddit, GitHub discussions, practitioner blogs
• Sources: 2-3 community sources
• Output: /tmp/research-community-$SESSION_ID.json

**Agent 3: Innovation and Trends Research**

• Task: "Research recent developments, emerging trends, and cutting-edge approaches for: $ARGUMENTS"
• Focus: Recent conference talks, innovation blogs, research papers
• Sources: 2-3 trend sources
• Output: /tmp/research-trends-$SESSION_ID.json

**Agent 4: Comparative Analysis Research**

• Task: "Research alternatives, comparisons, and decision criteria for: $ARGUMENTS"
• Focus: Technology comparisons, decision frameworks, trade-off analyses
• Sources: 2-3 comparative sources
• Output: /tmp/research-comparison-$SESSION_ID.json

**Agent 5: Risk and Implementation Research**

• Task: "Research risks, challenges, implementation considerations, and mitigation strategies for: $ARGUMENTS"
• Focus: Case studies, failure analyses, implementation guides
• Sources: 2-3 implementation sources
• Output: /tmp/research-risks-$SESSION_ID.json

STEP 4: Research Coordination and Validation

WAIT for all 5 agents to complete research phases.

FOR EACH agent result:

• Validate source quality and recency
• Extract key findings and insights
• Identify conflicting information
• Score source authority and relevance
• Update session state with findings

STEP 5: Cross-Reference Analysis

Think harder about potential biases in the sources and how to synthesize conflicting information:

• Identify consensus vs. debate areas
• Validate claims across multiple sources
• Resolve conflicting recommendations
• Assess information recency and relevance
• Weight findings by source authority

STEP 6: Comprehensive Report Generation

Create detailed research report: /tmp/deep-research-report-$SESSION_ID.md

Report structure:

# Deep Research Report: [Query]

## Executive Summary

- Key findings (3-5 bullet points)
- Primary recommendations
- Critical considerations

## Detailed Findings

### Current Best Practices

[Synthesized from Agent 1]

### Community Insights

[Synthesized from Agent 2]

### Emerging Trends

[Synthesized from Agent 3]

### Comparative Analysis

[Synthesized from Agent 4]

### Implementation Considerations

[Synthesized from Agent 5]

## Recommendations

### For [Project Context]

- Specific actionable recommendations
- Implementation priorities
- Risk mitigation strategies

## Further Research

- Recommended deep-dive topics
- Additional resources
- Follow-up questions

## Sources

[Comprehensive source list with quality ratings]

STEP 7: Context-Aware Recommendations

IF project context detected:

• Tailor recommendations to detected technology stack
• Provide specific implementation guidance
• Include project-relevant examples
• Suggest integration approaches

ELSE:

• Provide general best practices
• Include multiple implementation options
• Focus on foundational concepts

STEP 8: Quality Validation and Output

TRY:

• Validate report completeness and accuracy
• Ensure all research angles covered
• Verify source citations and links
• Generate executive summary
• Save final report to session directory

CATCH (incomplete_research):

• Identify missing research areas
• Launch additional targeted research agents
• Supplement findings with focused searches
• Update report with additional insights

CATCH (conflicting_information):

• Document conflicting viewpoints
• Research additional sourc