deep-research

cameronsjo's avatarfrom cameronsjo

Enables automatic deep research mode with comprehensive analysis when trigger phrases are detected

⭐1stars🔀0forks📁View on GitHub🕐Updated Dec 22, 2025
[Web Research]

When & Why to Use This Skill

The Deep Research skill for Claude automates comprehensive investigation workflows by detecting natural language triggers. It empowers the AI to perform extensive web searches, deploy specialized agents for codebase exploration, and synthesize multi-source data into cited, high-quality reports, ensuring no detail is overlooked during complex tasks.

Use Cases

  • Technical Deep Dives: Automatically trigger exhaustive analysis of complex codebase structures, such as authentication flows or architectural logic, using specialized exploration agents.
  • Comprehensive Web Investigation: Conduct thorough market or technical research by forcing the AI to use WebSearch extensively to ground responses in current, real-world data.
  • Systematic Problem Solving: Activate a methodical 'thinking' mode to debug persistent build failures or resolve intricate architectural bottlenecks through step-by-step analysis.
  • Evidence-Based Reporting: Generate detailed summaries that synthesize information from various sources with full citations, perfect for professional documentation and knowledge sharing.
namedeep-research
descriptionEnables automatic deep research mode with comprehensive analysis when trigger phrases are detected

Deep Research Skill

Enables automatic deep research mode when trigger phrases are detected in user prompts.

Overview

This skill provides a UserPromptSubmit hook that detects natural language cues indicating the user wants thorough, comprehensive research. When triggered, it injects instructions for Claude to use enhanced research capabilities.

When This Skill Activates

The hook automatically detects these trigger phrases (case-insensitive):

  • "deep dive"
  • "use your noodle"
  • "pull out the stops"
  • "dig deep"
  • "really research"
  • "thorough investigation"
  • "comprehensive analysis"
  • "leave no stone unturned"

What Deep Research Mode Does

When activated, Claude will:

  1. Use WebSearch extensively - Ground responses with current information from the web
  2. Spawn Explore agents - Thoroughly investigate the codebase with specialized agents
  3. Think systematically - Work through the problem methodically before responding
  4. Synthesize multiple sources - Combine findings from various sources into a cohesive answer
  5. Provide citations - Include sources for web-based findings

Installation

Hook Configuration

Add to your .claude/settings.json:

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "command": "~/.claude/skills/deep-research/scripts/detect-research-mode.sh"
      }
    ]
  }
}

Manual Setup

  1. Copy the skill to your Claude skills directory:

    cp -r registry/skills/deep-research ~/.claude/skills/

  2. Make the hook script executable:

    chmod +x ~/.claude/skills/deep-research/scripts/detect-research-mode.sh

  3. Add the hook configuration above to your settings

Usage Examples

Automatic Activation

Simply use trigger phrases naturally in your prompts:

"I need to deep dive into how authentication works in this codebase"
"Use your noodle on this one - why is the build failing?"
"Let's pull out the stops and figure out the best caching strategy"

What Happens

When a trigger phrase is detected, you'll see output like:

Deep research mode: Use WebSearch for current info, spawn Explore agents for codebase investigation, think systematically, synthesize findings from multiple sources.

This instructs Claude to use enhanced research capabilities for your question.

Customization

Adding Trigger Phrases

Edit scripts/detect-research-mode.sh and add patterns to the grep regex:

if echo "$input" | grep -qiE "(your-phrase|another-phrase)"; then

Modifying Research Instructions

Edit the output message in the script to customize what research behaviors are triggered.

Scripts

  • scripts/detect-research-mode.sh - UserPromptSubmit hook that detects trigger phrases
  • research-tools plugin - Commands for structured research workflows
  • comprehensive-researcher agent - Agent specialized in multi-source research