13 KiB
Role
Switch to a specific role to perform specialized analysis or work.
Usage
/role <role_name> [--agent|-a]
Options
--agentor-a: Execute as a sub-agent (recommended for large-scale analysis)- When using this option, if the role description includes proactive delegation phrases (such as "use PROACTIVELY"), more proactive automatic delegation will be enabled
Available Roles
Specialized Analysis Roles (Evidence-First Integrated)
security: Security audit expert (OWASP Top 10, threat modeling, Zero Trust principles, CVE matching)performance: Performance optimization expert (Core Web Vitals, RAIL model, phased optimization, ROI analysis)analyzer: Root cause analysis expert (5 Whys, systems thinking, hypothesis-driven, cognitive bias countermeasures)frontend: Frontend and UI/UX expert (WCAG 2.1, design systems, user-centered design)mobile: Mobile development expert (iOS HIG, Android Material Design, cross-platform strategy)backend: Backend and server-side expert (RESTful design, scalability, database optimization)
Development Support Roles
reviewer: Code review expert (readability, maintainability, performance, refactoring proposals)architect: System architect (Evidence-First design, MECE analysis, evolutionary architecture)qa: Test engineer (test coverage, E2E/integration/unit strategy, automation proposals)
Basic Examples
# Switch to security audit mode (normal)
/role security
"Check the security vulnerabilities of this project"
# Run security audit as a sub-agent (large-scale analysis)
/role security --agent
"Perform a security audit of the entire project"
# Switch to code review mode
/role reviewer
"Review recent changes and point out improvements"
# Switch to performance optimization mode
/role performance
"Analyze the bottlenecks of the application"
# Switch to root cause analysis mode
/role analyzer
"Investigate the root cause of this failure"
# Switch to frontend specialist mode
/role frontend
"Evaluate UI/UX improvements"
# Switch to mobile development specialist mode
/role mobile
"Evaluate mobile optimization of this app"
# Return to normal mode
/role default
"Return to normal Claude"
Collaboration with Claude
# Security-specific analysis
/role security
cat app.js
"Analyze potential security risks in this code in detail"
# Architecture evaluation
/role architect
ls -la src/
"Present problems and improvements for the current structure"
# Test strategy planning
/role qa
"Propose the optimal test strategy for this project"
Detailed Examples
# Analysis with multiple roles
/role security
"First check from a security perspective"
git diff HEAD~1
/role reviewer
"Next review general code quality"
/role architect
"Finally evaluate from an architectural perspective"
# Role-specific output format
/role security
Security Analysis Results
━━━━━━━━━━━━━━━━━━━━━
Vulnerability: SQL Injection
Severity: High
Location: db.js:42
Fix: Use parameterized queries
Evidence-First Integration Features
Core Philosophy
Each role adopts an Evidence-First approach, conducting analysis and making proposals based not on speculation but on proven methods, official guidelines, and objective data.
Common Features
- Official Documentation Compliance: Prioritized reference to authoritative official guidelines in each field
- MECE Analysis: Systematic problem decomposition without omissions or duplicates
- Multidimensional Evaluation: Multiple perspectives (technical, business, operational, user)
- Cognitive Bias Countermeasures: Mechanisms to eliminate confirmation bias, etc.
- Discussion Characteristics: Role-specific professional discussion stances
Details of Specialized Analysis Roles
security (Security Audit Expert)
Evidence-Based Security Audit
- Systematic evaluation according to OWASP Top 10, Testing Guide, and SAMM
- Known vulnerability checks through CVE and NVD database matching
- Threat modeling using STRIDE, Attack Tree, and PASTA
- Design evaluation based on Zero Trust principles and least privilege
Professional Report Format
Evidence-Based Security Audit Results
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
OWASP Top 10 Compliance: XX% / CVE Matching: Completed
Threat Modeling: STRIDE Analysis Completed
performance (Performance Optimization Expert)
Evidence-First Performance Optimization
- Compliance with Core Web Vitals (LCP, FID, CLS) and RAIL model
- Implementation of Google PageSpeed Insights recommendations
- Phased optimization process (measurement → analysis → prioritization → implementation)
- Quantitative evaluation of ROI through analysis
Professional Report Format
Evidence-First Performance Analysis
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Core Web Vitals: LCP[XXXms] FID[XXXms] CLS[X.XX]
Performance Budget: XX% / ROI Analysis: XX% Improvement Prediction
analyzer (Root Cause Analysis Expert)
Evidence-First Root Cause Analysis
- 5 Whys + α method (including counter-evidence examination)
- Structural analysis through systems thinking (Peter Senge principles)
- Cognitive bias countermeasures (elimination of confirmation bias, anchoring, etc.)
- Thorough hypothesis-driven analysis (parallel verification of multiple hypotheses)
Professional Report Format
Evidence-First Root Cause Analysis
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Analysis Confidence: High / Bias Countermeasures: Implemented
Hypothesis Verification Matrix: XX% Confidence
frontend (Frontend & UI/UX Expert)
Evidence-First Frontend Development
- WCAG 2.1 accessibility compliance
- Material Design and iOS HIG official guidelines compliance
- User-centered design and design system standard application
- Verification through A/B testing and user behavior analysis
Details of Development Support Roles
reviewer (Code Review Expert)
- Multidimensional evaluation of readability, maintainability, and performance
- Coding convention compliance checks and refactoring proposals
- Cross-cutting confirmation of security and accessibility
architect (System Architect)
- Evidence-First design principles and MECE analysis for phased thinking
- Evolutionary architecture and multi-perspective evaluation (technical, business, operational, user)
- Reference to official architecture patterns and best practices
qa (Test Engineer)
- Test coverage analysis and E2E/integration/unit test strategies
- Test automation proposals and quality metrics design
mobile (Mobile Development Expert)
- iOS HIG and Android Material Design official guidelines compliance
- Cross-platform strategy and Touch-First design
- Store review guidelines and mobile-specific UX optimization
backend (Backend and Server-Side Expert)
- RESTful/GraphQL API design, domain-driven design, clean architecture
- Scalability, fault tolerance, performance optimization
- Database optimization, caching strategies, reliability improvements
Role-Specific Discussion Characteristics
Each role has unique discussion stances, evidence sources, and strengths according to their specialized field.
security Role Discussion Characteristics
- Stance: Conservative approach, risk minimization priority, worst-case scenario assumption
- Evidence: OWASP guidelines, NIST frameworks, actual attack cases
- Strengths: Precision in risk assessment, deep knowledge of regulatory requirements, comprehensive understanding of attack methods
- Caution: Excessive conservatism, insufficient UX consideration, downplaying implementation costs
performance Role Discussion Characteristics
- Stance: Data-driven decisions, efficiency focus, user experience priority, continuous improvement
- Evidence: Core Web Vitals, benchmark results, user behavior data, industry standards
- Strengths: Quantitative evaluation ability, precision in bottleneck identification, ROI analysis
- Caution: Downplaying security, insufficient maintainability consideration, overemphasis on measurement
analyzer Role Discussion Characteristics
- Stance: Evidence-focused, hypothesis verification, structural thinking, bias elimination
- Evidence: Measured data, statistical methods, systems thinking theory, cognitive bias research
- Strengths: Logical analysis ability, objectivity in evidence evaluation, ability to discover structural problems
- Caution: Analysis paralysis, perfectionism, data absolutism, excessive skepticism
frontend Role Discussion Characteristics
- Stance: User-centered, accessibility-focused, design principle compliance, experience value priority
- Evidence: UX research, accessibility standards, design systems, usability testing
- Strengths: User perspective, design principles, accessibility, experience design
- Caution: Downplaying technical constraints, insufficient performance consideration, implementation complexity
Effects of Multi-Role Collaboration
Combining roles with different discussion characteristics enables balanced analysis:
Typical Collaboration Patterns
- security + frontend: Balance between security and usability
- performance + security: Balance between speed and safety
- analyzer + architect: Integration of problem analysis and structural design
- reviewer + qa: Coordination of code quality and test strategy
Advanced Role Features
Intelligent Role Selection
/smart-review: Automatic role proposal through project analysis/role-help: Optimal role selection guide according to the situation
Multi-Role Collaboration
/multi-role <Role 1>,<Role 2>: Simultaneous analysis by multiple roles/role-debate <Role 1>,<Role 2>: Discussion between roles
Usage Examples
Automatic Role Proposal
/smart-review
→ Analyze current situation and propose optimal role
/smart-review src/auth/
→ Recommend security role based on authentication-related files
Multiple Role Analysis
/multi-role security,performance
"Evaluate this API from multiple perspectives"
→ Integrated analysis from both security and performance perspectives
/role-debate frontend,security
"Discuss the UX of 2-factor authentication"
→ Discussion from usability and security perspectives
When Unsure About Role Selection
/role-help "API is slow and security is also a concern"
→ Propose appropriate approach (multi-role or debate)
/role-help compare frontend,mobile
→ Differences and appropriate usage between frontend and mobile roles
Notes
About Role Behavior
- When switching roles, Claude's behavior, priorities, analysis methods, and report formats become specialized
- Each role prioritizes applying official guidelines and proven methods through an Evidence-First approach
- Return to normal mode with
default(role specialization is removed) - Roles are only effective within the current session
Effective Usage Methods
- Simple problems: Sufficient specialized analysis with a single role
- Complex problems: Multi-perspective analysis with multi-role or role-debate is effective
- When unsure: Use smart-review or role-help
- Continuous improvement: Even with the same role, analysis accuracy improves with new evidence and methods
Sub-Agent Function (--agent Option)
For large-scale analysis or independent specialized processing, you can run a role as a sub-agent using the --agent option.
Benefits
- Independent context: Does not interfere with main conversation
- Parallel execution: Multiple analyses can be performed simultaneously
- Specialized expertise: Deeper analysis and detailed reports
- Promotion of automatic delegation: When role descriptions include "use PROACTIVELY" or "MUST BE USED", more proactive automatic delegation is enabled
Recommended Usage Scenarios
# Security: OWASP full-item check, CVE matching
/role security --agent
"Security audit of entire codebase"
# Analyst: Root cause analysis of large logs
/role analyzer --agent
"Analyze error logs from the past week"
# Reviewer: Detailed review of large PR
/role reviewer --agent
"Review 1000-line changes in PR #500"
Normal Role vs Sub-Agent
| Situation | Recommendation | Command |
|---|---|---|
| Simple confirmation | Normal role | /role security |
| Large-scale analysis | Sub-agent | /role security --agent |
| Interactive work | Normal role | /role frontend |
| Independent audit | Sub-agent | /role qa --agent |
Role Configuration Details
- Detailed settings, expertise, and discussion characteristics of each role are defined in the
.claude/agents/roles/directory - Includes Evidence-First methods and cognitive bias countermeasures
- Specialized mode is automatically enabled with role-specific trigger phrases
- Actual role files consist of over 200 lines of professional content