Claude Code Heavy

July 18, 2025 · View on GitHub

A powerful multi-agent research system using Claude Code to deliver comprehensive, multi-perspective analysis through intelligent orchestration. Inspired by Grok's heavy mode and make-it-heavy.

🚀 What is Claude Code Heavy?

Claude Code Heavy gives Claude full control to orchestrate parallel research:

  • 🧠 Intelligent Planning: Claude analyzes your query and creates the optimal research strategy
  • 🔀 Parallel Research: Claude decides how many agents to deploy (2-8 available)
  • 🎯 Dynamic Adaptation: Questions and roles are tailored to your specific query
  • 🔄 Smart Synthesis: Claude combines all findings into unified insights
  • 🚙 Interactive Mode: Simple setup with powerful results

🎯 Quick Start

# Clone the repo
git clone https://github.com/yourusername/claude-code-heavy
cd claude-code-heavy

# Make script executable
chmod +x ccheavy.sh

# Run interactively (recommended)
./ccheavy.sh

# Or run with command line args
./ccheavy.sh "How do we solve the global water supply issue?"

🤔 Why This Approach?

Claude is the Intelligence: Instead of hard-coding research patterns, we let Claude:

  • Analyze your query and determine the best approach
  • Decide how many research assistants are needed
  • Create specific, targeted research questions
  • Adapt the research strategy in real-time

Pre-built Infrastructure: The script simply:

  • Creates 8 research workspaces (worktrees)
  • Sets up output directories
  • Launches Claude with full orchestration control

📖 How It Works

graph TD
    A[Run ./ccheavy.sh] --> B[Setup Workspaces]
    B --> C[Claude Analyzes Query]
    C --> D[Creates Research Plan]
    D --> E[Decides Assistant Count]
    E --> F[Generates Questions]
    F --> G[Parallel Research Phase]
    G --> H1[RA-1 Research]
    G --> H2[RA-2 Research]
    G --> H3[RA-3 Research]
    G --> H4[... RA-N Research]
    H1 --> I[Synthesis]
    H2 --> I
    H3 --> I
    H4 --> I
    I --> J[Final Analysis]

Process Flow

  1. You provide a query: Any research question or topic
  2. Script prepares environment: Creates 8 workspaces, output directories
  3. Claude takes over:
    • Analyzes complexity and scope
    • Plans optimal research strategy
    • Creates targeted questions
    • Orchestrates parallel execution
    • Synthesizes comprehensive results

🛠️ Installation

Prerequisites

  • Claude Code installed (npm install -g @anthropic-ai/claude-code)
  • Claude Pro or Teams subscription (strongly recommended for best performance)
  • Git with worktree support (git 2.7+)
  • Unix-like environment (Mac/Linux/WSL)

Setup

# Clone and setup
git clone https://github.com/gtrusler/claude-code-heavy
cd claude-code-heavy
./setup.sh

🎮 Usage

./ccheavy.sh
# Follow the prompts:
# - Enter your research question
# - Choose output format (markdown/text)
# - Decide on dangerous mode (default: N)
# - Confirm to proceed (default: Y)
# - Optionally auto-launch Claude Code (default: Y)

Command Line Mode

# Basic usage
./ccheavy.sh "What are the latest advances in quantum computing?"

# With text output
./ccheavy.sh "Explain Docker containers" text

# YOLO mode
./ccheavy.sh "Analyze my codebase" markdown --dangerous

Output Structure

outputs/
└── 2025-07-18-solve-global-water-crisis/
    ├── orchestration-prompt.md    # Initial prompt for Claude
    ├── research-plan.md          # Claude's research strategy
    ├── assistants/
    │   ├── ra-1-findings.md      # Assistant 1's research
    │   ├── ra-2-findings.md      # Assistant 2's research
    │   └── ...                   # Up to 8 assistants
    └── final-analysis.md         # Synthesized results

📋 Example Outputs

See the examples/ directory for complete research outputs:

1. Austin's 5-Year Outlook

  • Query: "What does the outlook for the next 5 years look like for Austin, TX?"
  • Assistants: 6 (Claude's decision based on complexity)
  • Total Output: ~11,500 words
  • Key Insights: Economic growth, tech hub evolution, infrastructure challenges

2. AI's Impact on Healthcare

  • Query: "How will AI impact healthcare?"
  • Assistants: 5 (covering different healthcare domains)
  • Focus Areas: Diagnostics, drug discovery, patient care, ethics, implementation
  • Comprehensive analysis of transformation across the industry

Quick examples of queries Claude handles well:

  • Scientific: "How do mRNA vaccines work?"
  • Business: "Analyze the electric vehicle market"
  • Policy: "Evaluate universal basic income proposals"
  • Technical: "Compare Rust vs Go for systems programming"
  • Historical: "What led to the fall of the Roman Empire?"
  • Local: "What's the 5-year outlook for Austin, TX?"

🔧 Configuration

The script is intentionally minimal. Claude handles:

  • Determining optimal assistant count
  • Creating research questions
  • Assigning roles and perspectives
  • Adapting to query complexity

🚀 Advanced Features

Security Options

  • Option to use --dangerously-skip-permissions flag
  • Default is secure mode (recommended)
  • Only enable for trusted research tasks

Parallel Execution

  • Claude orchestrates parallel tool calls
  • Multiple web searches execute simultaneously
  • Faster research completion

Smart Folder Naming

  • Queries converted to readable folder names automatically
  • Date prefixing for easy chronological sorting
  • Example: "How will AI impact health care?" → 2025-07-18-how-will-ai-impact-health-care

🏗️ Architecture

Why Git Worktrees?

  • Each assistant gets isolated workspace
  • Parallel execution without conflicts
  • Easy context switching for Claude
  • Git tracks all research history

Why Let Claude Decide?

  • Every query is unique
  • Claude understands context better
  • Dynamic adaptation beats static patterns
  • Optimal resource allocation

🐛 Troubleshooting

"Command not found: claude"

Make sure Claude Code is installed: npm install -g @anthropic-ai/claude-code

"Git worktree error"

# Update git
brew upgrade git  # Mac
sudo apt-get update && sudo apt-get upgrade git  # Linux

Research seems slow

  • Normal research takes 5-10 minutes
  • Complex queries may take up to 15-20 minutes
  • Check if web searches are working

🤝 Contributing

  1. Fork the repository
  2. Create your feature branch
  3. Test with various query types
  4. Submit a pull request

📄 License

MIT License with Commercial Attribution Requirement

For products with 100K+ users: Please include attribution to Graydon Trusler and mention the "Claude Code Heavy" framework in your documentation or credits.

See LICENSE file for full details.

🙏 Acknowledgments

  • Inspired by make-it-heavy by Pietro Schirano
  • Inspired by Grok Heavy mode
  • Built on Claude Code by Anthropic