Skip to main content

Claude Code

Claude Code is Anthropic's official CLI coding assistant. It can be configured to work through the University of Florida's AI LiteLLM proxy, allowing you to use AI models provided by NaviGator Toolkit.

Installing Claude Code​

Claude Code can be installed via npm:

npm install -g @anthropic-ai/claude-code

Alternatively, see the official installation guide for other methods.

claude code application

Configure Claude Code to use NaviGator Toolkit​

To configure Claude Code to use NaviGator Toolkit, you need to set the following environment variables in your shell:

export ANTHROPIC_BASE_URL="https://api.ai.it.ufl.edu"
export ANTHROPIC_AUTH_TOKEN="YOUR_API_KEY"
export DISABLE_PROMPT_CACHING=1
export ANTHROPIC_DEFAULT_SONNET_MODEL="claude-sonnet-4.6"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="claude-haiku-4.5"
export ANTHROPIC_DEFAULT_OPUS_MODEL="claude-opus-4.6"

Replace YOUR_API_KEY with your actual UF AI virtual key.

These environment variables tell Claude Code where to send requests and how to authenticate, map Claude Code's internal model roles to the proxy's model names, and disable prompt caching (which the UF proxy does not support).

To make these settings permanent, add them to your shell's configuration file (~/.zshrc on macOS, ~/.bashrc on Linux).

How to Use Claude Code​

Claude Code is a powerful terminal-based AI coding assistant that brings AI assistance directly to your command line. Unlike IDE-integrated tools, Claude Code is designed for developers who prefer terminal workflows and need AI assistance that integrates seamlessly with their existing command-line tools and scripts.

Key Features & Capabilities​

  • Interactive TUI Interface: Terminal user interface for seamless AI interaction
  • Session Management: Continue previous conversations with --continue or specific --session
  • Multiple Execution Modes: Run with direct messages or interactive sessions
  • Model Selection: Choose from various AI models with -m, --model flag
  • Agent Support: Use specialized agents for different tasks
  • GitHub Integration: Built-in GitHub agent for repository operations
  • Headless Server Mode: Run Claude Code as a server for remote access

Common Use Cases & Examples​

1. Code Review & Analysis​

Example: "Review this Python module for best practices and potential issues"

claude
# In the interactive TUI:
# "Analyze the user_auth.py file for security vulnerabilities and code quality issues"

Or use direct message mode:

claude run "Review the user_auth.py file for security issues and suggest improvements"

Claude Code will:

  • Examine your code structure and patterns
  • Identify potential security issues
  • Suggest improvements for readability and maintainability
  • Generate detailed review comments

2. Debugging & Problem Solving​

Example: "Help me debug why my tests are failing"

claude
# In the interactive session:
# "Run the test suite and help me understand what's failing and why"

Or for quick debugging:

claude run "My React tests are failing with 'Cannot read property of undefined'. Help me debug this."

Claude Code can:

  • Execute your test commands and analyze output
  • Identify root causes of test failures
  • Suggest specific fixes for failing tests
  • Help trace through complex debugging scenarios

3. Documentation Generation​

Example: "Generate comprehensive API documentation for my Express.js routes"

claude
# In the TUI:
# "Create API documentation for all routes in the routes/ directory"

Or as a direct command:

claude run "Generate OpenAPI documentation for my Express.js API endpoints"

Claude Code will:

  • Analyze your route definitions and middleware
  • Generate OpenAPI/Swagger specifications
  • Create markdown documentation with examples
  • Add inline code comments where needed

4. Feature Implementation​

Example: "Add authentication middleware to my Node.js API"

claude
# Interactive session for complex implementations:
# "Implement JWT-based authentication middleware with proper error handling"

Claude Code can:

  • Create new middleware files
  • Update existing route configurations
  • Add necessary dependencies to package.json
  • Implement proper error handling and validation

5. GitHub Integration & Repository Management​

Example: Working with GitHub repositories

claude github
# Access GitHub-specific agent for repository operations

This allows you to:

  • Create and manage pull requests
  • Analyze repository structure and issues
  • Automate common GitHub workflows
  • Review code changes across branches

Terminal Workflow Examples​

Quick Code Analysis​

# Start Claude Code and analyze a specific file
claude
> "Analyze the performance bottlenecks in database.py and suggest optimizations"

Batch File Processing​

# Process multiple files with a single command
claude
> "Add proper error handling to all Python files in the src/ directory"

Integration with Git Workflows​

# Review changes before committing
claude run "Review the files I've changed in this commit for any issues"

Or use interactive mode for detailed review:

claude --continue
# "Please review my recent changes and suggest improvements before I commit"

Best Practices for Terminal-Based AI Coding​

Effective Command Usage​

  • Be Specific: Use precise descriptions of what you want to accomplish
  • Provide Context: Mention file paths, project structure, and constraints
  • Choose the Right Mode: Use interactive mode for complex tasks, direct mode for quick queries

Integration with Existing Workflows​

  • Combine with Git: Use Claude Code for pre-commit reviews and analysis
  • Script Integration: Incorporate claude run commands into your build scripts
  • Session Continuity: Use --continue to maintain context across work sessions

Managing Large Projects​

  • Focus on Specific Areas: Target specific directories or file patterns in your prompts
  • Use Session Management: Continue previous sessions to maintain project context
  • Leverage GitHub Agent: Use claude github for repository-wide operations

Getting Started Tips​

  1. Start with Simple Queries: Begin with claude run for quick tasks to understand capabilities
  2. Try Interactive Mode: Use the TUI for complex, multi-step tasks
  3. Use Session Management: Continue sessions to build on previous conversations
  4. Explore Agents: Try claude github for repository-specific tasks

Example Terminal Session​

$ claude
Claude Code AI Agent
> "My React app isn't rendering the user list component. Help me debug this."

[Claude Code analyzes your project structure]
[Identifies potential issues in component files]
[Suggests running specific commands to diagnose]

> "Run npm test -- --verbose UserList.test.js"
[Executes test and analyzes output]
[Provides specific recommendations for fixes]

> "The issue is in the useEffect dependency array. Let me fix that."
[Makes targeted changes to resolve the issue]
[Verifies the fix works correctly]

This terminal-first approach makes Claude Code particularly powerful for developers who live in the command line and want AI assistance that fits naturally into their existing workflows.

References

  1. Claude Code https://docs.anthropic.com/en/docs/claude-code/overview
  2. Claude Code Docs https://docs.anthropic.com/en/docs/claude-code/
  3. Github Repository https://github.com/anthropics/claude-code