Roles Skill

Type 6 Guidance-Only skill providing role-based perspectives for multi-agent code review workflows.

Purpose

Provides 4 specialized review perspectives for comprehensive code review:

1. Development (code-reviewer) - Code quality, security, performance
2. QA (test-engineer) - Test coverage, edge cases, regression risks
3. Security (security-specialist) - Threat analysis, compliance, vulnerability assessment
4. Operations (devops-sre) - Deployment, monitoring, reliability

Capabilities

• Multi-Perspective Analysis: 4 parallel review roles with distinct focus areas
• Quality Frameworks: Self-critique scoring (Completeness, Actionability, Accuracy ≥8/10)
• Severity Levels: Critical/Warning/Suggestion classification
• Actionable Feedback: Specific fixes with code examples
• Skill Integration: References security-audit, coderabbit, observability skills

Quick Reference

PR Review Workflow: See resources/pr-review-workflow.md

How to Use

Natural language request (recommended):

User: "Review PR 123 from all perspectives"
User: "Multi-perspective review for https://github.com/..."
User: "Security review for PR 456"

Claude loads this skill and executes resources/pr-review-workflow.md.

Direct skill invocation:

Skill("roles")

# Execute complete workflow:
# resources/pr-review-workflow.md

# Or access individual role perspectives:
# - resources/code-reviewer.md
# - resources/test-engineer.md
# - resources/security-specialist.md
# - resources/devops-sre.md

Resources

Development Perspective (resources/code-reviewer.md):

• Code quality assessment (readability, naming, DRY)
• Security vulnerabilities (secrets, injection, XSS)
• Performance issues (algorithmic efficiency, N+1 queries)
• Maintainability (test coverage >80%, error handling, logging)
• CodeRabbit CLI integration for expert issue detection

QA Perspective (resources/test-engineer.md):

• Test coverage assessment (unit, integration, E2E)
• Edge case detection (boundary values, error paths)
• Regression risk analysis (breaking changes, compatibility)
• Test quality evaluation (assertions, mocking, flakiness)
• Test documentation (setup, fixtures, data management)

Security Perspective (resources/security-specialist.md):

• Threat modeling (STRIDE framework, attack trees)
• Vulnerability assessment (OWASP Top 10, CVE patterns)
• Secure architecture (defense in depth, zero trust)
• Authentication/authorization (OAuth, JWT, RBAC)
• Compliance mapping (SOC2, HIPAA, GDPR requirements)
• References security-audit skill for structured procedures

Operations Perspective (resources/devops-sre.md):

• Deployment readiness (migrations, rollback, feature flags)
• Observability (logging, metrics, tracing, alerting)
• Reliability (error handling, retries, circuit breakers, SLIs)
• Performance (resource usage, scaling, bottlenecks)
• Infrastructure (config, secrets, permissions)

Review Synthesis (resources/review-synthesis.md):

• Multi-agent finding aggregation (collect from 4-6 agents)
• Severity classification (Critical/Warning/Suggestion)
• Deduplication strategy (overlapping issues)
• Prioritization algorithm (implementation order)
• Action plan generation (checklist format)
• Used by /pr-review command Step 4

Review Framework

Self-Critique Quality Gate

Each role provides self-critique scoring:

Quality threshold: All scores ≥8/10

Severity Classification

Integration with Other Skills

security-audit skill: STRIDE threat modeling, OWASP checks, compliance frameworks

coderabbit skill: CodeRabbit CLI for expert issue detection (race conditions, memory leaks)

observability skill: Prometheus/Grafana patterns, SLI/SLO definitions

When to Use

Trigger phrases:

• "PR review", "code review", "multi-perspective analysis"
• "Security review", "QA review", "operations review"
• "Review from different angles", "comprehensive review"

Actions:

1. Load skill via /pr-review command (recommended)
2. Or reference individual role perspectives for focused reviews
3. Apply role frameworks from resources

Progressive Disclosure

Metadata (~60 words):

• Loaded when "role", "review", "perspective" mentioned
• Description covers 4 roles + multi-agent pattern

SKILL.md (~400 lines):

• Overview of 4 roles
• Quick reference table
• Integration with other skills

Resources (~58K total, loaded on-demand):

• code-reviewer.md (~14K) - Development perspective
• test-engineer.md (~26K) - QA perspective
• security-specialist.md (~9K) - Security perspective
• devops-sre.md (~9K) - Operations perspective

Token efficiency: Load only needed perspective(s), not all 4

Success Metrics

Skill properly established:

• ✅ Type 6 Guidance-Only (no scripts, guidance only)
• ✅ 4 role perspectives in resources
• ✅ /pr-review command references skill
• ✅ Parent standards inherited
• ✅ Progressive disclosure effective

Quality outcomes:

• ✅ All reviews ≥8/10 on self-critique dimensions
• ✅ Severity levels properly classified
• ✅ Actionable feedback with code examples
• ✅ Integration with specialized skills (security-audit, coderabbit, observability)

Notes

Type 6 Guidance-Only: No scripts, role perspective guidance only (matching pragmatic, framework, security-audit patterns)

Multi-agent pattern: PR review workflow orchestrates 4 parallel reviews using perspectives from this skill (invoked via natural language)

Relationship with agents: Skill contains guidance and workflow, Claude constructs Task calls with prompts from skill resources

Portability: Self-contained .claude/skills/roles/ folder (reusable across projects)