Modes Overview

Behavioral Modes

Modes adjust Claude’s communication style, output format, and problem-solving approach. Choose the right mode for your task to optimize efficiency and results.

Quick Reference

Mode	Best For	Token Usage	Output Style
Default	General development tasks	Standard	Balanced explanation + code
Brainstorm	Design, ideation, exploration	Higher	Questions, alternatives, comparisons
Token-Efficient	High-volume, simple tasks	30-70% less	Minimal prose, code-focused
Deep Research	Investigation, audits	Higher	Structured reports, citations
Implementation	Executing clear plans	Lower	Code blocks, minimal explanation
Review	Code review, QA	Standard	Categorized findings, actionable feedback
Orchestration	Complex multi-task coordination	Variable	Task delegation, progress tracking

Mode Selection Guide

By Development Phase

graph LR
    A[New Feature] --> B[Brainstorm Mode]
    B --> C[Plan Created]
    C --> D[Implementation Mode]
    D --> E[Code Complete]
    E --> F[Review Mode]
    F --> G[Ship]

By Task Type

Task	Recommended Mode	Alternative
”How should I structure this?”	Brainstorm	Default
”Implement this feature”	Implementation	Default
”Fix this bug”	Default	Deep Research (if complex)
“Review this PR”	Review	Default + —persona=security
”Research this approach”	Deep Research	Brainstorm
”Generate 10 similar files”	Token-Efficient	Implementation
”Coordinate multi-file refactor”	Orchestration	Default

By Team Experience

Experience Level	Recommended Modes
Junior developers	Default, Brainstorm (for learning)
Senior developers	Token-Efficient, Implementation (for speed)
Mixed teams	Default, Review (for knowledge sharing)

Switching Modes

Session-Wide Mode Change

/mode brainstorm              # All subsequent commands use brainstorm mode
/mode token-efficient         # Switch to token-efficient for the session
/mode default                 # Return to standard behavior

Single Command Override

/plan --mode=brainstorm "new feature"
/fix --format=concise "error message"
/review --mode=review --persona=security src/auth/

Mode Combinations

Some modes work well together via flags:

# Deep research with concise output
/research --mode=deep-research --format=concise [topic]

# Implementation with checkpoints
/execute-plan --mode=implementation --checkpoint [file]

# Brainstorm with saved output
/brainstorm --mode=brainstorm --save=plans/design.md [topic]

Understanding Mode Behavior

Each mode modifies three key aspects:

1. Communication Style

• Verbose modes (Brainstorm, Deep Research): More questions, explanations, alternatives
• Concise modes (Token-Efficient, Implementation): Direct answers, minimal prose
• Balanced modes (Default, Review): Clear but not excessive

2. Problem-Solving Approach

• Exploratory (Brainstorm, Deep Research): Consider many options, delay decisions
• Decisive (Implementation, Token-Efficient): Make reasonable defaults, execute
• Critical (Review): Question assumptions, find issues

3. Output Format

• Structured reports (Deep Research, Review): Categorized, formatted findings
• Code-focused (Implementation, Token-Efficient): Minimal text, maximum code
• Discussion-oriented (Brainstorm, Default): Balanced code and explanation

Mode Feature Matrix

Feature	Default	Brainstorm	Token-Efficient	Deep Research	Implementation	Review	Orchestration
Ask clarifying questions	Sometimes	Frequently	Rarely	Sometimes	Rarely	Sometimes	Sometimes
Present alternatives	Sometimes	Always	No	Sometimes	No	Sometimes	No
Detailed explanations	Yes	Yes	No	Yes	No	Yes	Moderate
Code comments	Standard	Verbose	Minimal	Standard	Minimal	Standard	Standard
Progress updates	No	No	No	Yes	Yes	No	Yes
Confidence levels	No	No	No	Yes	No	No	No
File citations	No	No	No	Yes	No	Yes	No
Task parallelization	No	No	No	No	No	No	Yes

Best Practices

Start Broad, Then Focus

# 1. Explore options
/brainstorm "authentication approach"

# 2. Plan implementation
/plan --mode=default "implement OAuth2"

# 3. Execute efficiently
/execute-plan --mode=implementation plan.md

# 4. Review thoroughly
/review --mode=review src/auth/

Match Mode to Iteration

• First iteration: Use Brainstorm or Default to explore
• Subsequent iterations: Use Implementation or Token-Efficient for speed
• Final iteration: Use Review to catch issues

Know When to Switch

• Stuck on implementation? → Switch to Brainstorm
• Plan is clear? → Switch to Implementation
• Need to save costs? → Switch to Token-Efficient
• Complex bug? → Switch to Deep Research

Common Patterns

Feature Development

/mode brainstorm
/brainstorm "user profile feature"
# Explore options, make decisions

/mode implementation
/feature "user profile with avatar upload"
# Execute the plan

/mode review
/review src/features/profile/
# Quality check before merge

Bug Investigation

/mode deep-research
/fix --depth=5 "memory leak in worker process"
# Thorough investigation

/mode implementation
# Apply the fix

/mode review
/review --persona=performance src/workers/
# Verify fix and performance

Batch Operations

/mode token-efficient
/test --format=ultra src/services/*.ts
# Generate tests for all services efficiently

Custom Mode Configuration

Advanced users can customize mode behavior in .claude/modes/. Each mode is defined in a markdown file that specifies:

• Communication patterns
• Output formats
• Problem-solving approaches
• Activation conditions

See individual mode pages for detailed behavior specifications.

Related Documentation

• Commands Reference - Using modes with commands
• Flags & Options - Mode-related flags
• Token Optimization - Cost-saving strategies