Advanced Claude Code Features

Unlock Claude Code's full potential with these advanced features.

Planning Mode

Planning mode lets Claude design an implementation approach before writing code.

Entering Plan Mode

Plan how to implement user authentication with JWT tokens

Or explicitly:

/plan Add a caching layer to the API

What Happens in Planning Mode

Claude explores the codebase
Identifies relevant files and patterns
Designs an implementation approach
Presents the plan for your approval
Only proceeds after confirmation

When to Use Planning

Complex multi-file changes
Architectural decisions
New feature implementations
Unclear requirements

Extended Thinking

Extended thinking allows Claude to reason more deeply about complex problems.

Enabling Extended Thinking

// ~/.claude/settings.json
{
  "model": {
    "enableExtendedThinking": true,
    "thinkingBudget": 10000
  }
}

Best Use Cases

Complex debugging
Architecture decisions
Security analysis
Performance optimization

Background Tasks

Run long-running operations without blocking your session.

Starting a Background Task

Run the full test suite in the background

Checking Task Status

What's the status of my background tasks?

Use Cases

Long test suites
Build processes
Data migrations
Large refactoring operations

Subagents

Claude can spawn specialized agents for specific tasks.

How Subagents Work

Claude (main) → spawns → Explorer Agent (codebase analysis)
                      → spawns → Planner Agent (architecture)
                      → spawns → Reviewer Agent (code review)

Subagent Types

| Agent | Purpose | |-------|---------| | Explore | Codebase analysis and search | | Plan | Architecture and implementation design | | Code Review | Quality and security analysis | | Feature Dev | Guided feature development |

Triggering Subagents

Most subagents activate automatically based on context, but you can request them:

Explore how error handling works in this codebase

Checkpoints and Undo

Creating Checkpoints

Claude automatically creates checkpoints before significant changes.

Reverting Changes

Undo the last set of changes

Viewing History

Show me the recent changes you've made

Context Management

Summarization

For long sessions, Claude automatically summarizes older context to maintain focus.

Manual Context Control

/clear          # Clear conversation history
/compact        # Summarize current context

Adding Context

Reference specific files:

Read src/auth/middleware.ts and explain the authentication flow

Performance Optimization

Model Selection

{
  "model": {
    "default": "claude-sonnet-4-20250514",
    "planning": "claude-opus-4-20250514",
    "quickTasks": "claude-haiku-3-5-20241022"
  }
}

Token Management

{
  "limits": {
    "maxTokensPerTurn": 4096,
    "maxContextTokens": 100000
  }
}

CLI Flags

Useful Flags

# Start with specific model
claude --model claude-opus-4-20250514

# Start in a specific directory
claude --cwd /path/to/project

# Resume last session
claude --resume

# Start with a task
claude "Fix the failing tests"

Environment Variables

| Variable | Description | |----------|-------------| | ANTHROPIC_API_KEY | API authentication | | CLAUDE_CODE_DEBUG | Enable debug logging | | CLAUDE_CODE_MODEL | Default model override |

Debugging

Enable Debug Mode

CLAUDE_CODE_DEBUG=1 claude

View Logs

tail -f ~/.claude/logs/claude-code.log

Common Issues

| Issue | Solution | |-------|----------| | Slow responses | Check model selection, reduce context | | Tool failures | Check permissions, verify paths | | Memory issues | Use /compact or /clear |

Best Practices

Use planning for complex tasks: Better results with upfront design
Leverage background tasks: Don't wait for long operations
Manage context actively: Clear/compact when sessions get long
Choose the right model: Balance speed and capability
Enable extended thinking: For truly complex problems