Skip to main content

Agent Team Guide

Complete reference for Claude Code Agent Suite with 15 specialized agents, skills, MCP servers, and security tools.

🤖 Global Agent Team (15 Agents)​

Backend Engineering (4 agents)​

AgentSpecialty
python-architectModern Python (3.10+), asyncio, type-safe development
go-expertGo concurrency, goroutines, idiomatic patterns
node-engineerNode.js runtime, async/await, npm ecosystem
backend-engineerGeneric backend: APIs, databases, caching, distributed systems

Frontend & UI (3 agents)​

AgentSpecialty
frontend-architectSemantic HTML5, CSS3 Grid/Flexbox, accessibility
astro-expertAstroJS, SSG, Islands Architecture
ux-ui-designerUser-centered design, WCAG accessibility, design systems

Application Development (1 agent)​

AgentSpecialty
streamlit-expertStreamlit dashboards, caching, session state

Infrastructure & Automation (1 agent)​

AgentSpecialty
powershell-automatorPowerShell scripting, Windows automation

Database Specialists (3 agents with MCP access)​

AgentSpecialty
postgres-dbaPostgreSQL query optimization, schema design
mysql-expertMySQL InnoDB internals, performance tuning
mongo-architectMongoDB NoSQL design, Aggregation Framework

Security Engineering (2 agents)​

AgentSpecialty
security-engineerThreat modeling, secure architecture, OWASP Top 10
secure-code-reviewerVulnerability detection, security anti-patterns

Meta-Agent (1 agent)​

AgentSpecialty
prompt-engineerLLM prompt optimization, agent instruction refinement

🔒 Security Configuration​

Permission System​

{
  "permissions": {
    "allow": ["Read", "Glob", "Grep"],
    "deny": []
  }
}

PreToolUse Hook - Bash Command Safety​

  • Validates bash commands before execution
  • Checks for destructive operations (rm -rf, git reset --hard)
  • Detects credential exposure
  • Blocks network requests to unknown hosts

PostToolUse Hook - File Modification Tracking​

  • Notification after Edit/Write operations
  • Triggers formatting, linting, or audit logging

🎯 Recommended Workflows​

Secure Code Review​

1. /secure-code-reviewer - Initial security scan
2. /trailofbits:differential-review - Deep change analysis
3. /trailofbits:variant-analysis - Find similar issues
4. /trailofbits:static-analysis - Automated tool scanning
5. /trailofbits:fix-review - Validate remediation

New Feature Development​

1. /ux-ui-designer - Design user experience
2. /backend-engineer - Plan API architecture
3. /python-architect - Implement
4. /generate-tests - Create test coverage
5. /security-engineer - Security review
6. /review-code - Final code review

Database Optimization​

1. /postgres-dba - Analyze query performance
2. Use MCP tools to run EXPLAIN ANALYZE
3. /backend-engineer - Implement caching strategy

📚 Best Practices​

Agent Design Principles​

  • Single Responsibility: Each agent has one clear expertise domain
  • Strict Tool Scoping: Agents only have tools they need
  • Read-Only by Default: Review agents use Read/Grep/Glob, not Edit/Write
  • Model Selection: Sonnet for complex reasoning, Haiku for fast operations

Security Hardening​

  • Least Privilege: Only grant permissions explicitly needed
  • Defense in Depth: Multiple security layers (hooks + permissions + agents)
  • Secrets Management: Never hardcode credentials
  • Input Validation: Security agents enforce validation at all boundaries

🚀 Quick Commands​

# Agent Management
/agents                    # List all agents
/agents refresh            # Reload agent definitions

# Skill Management
/skills                    # List all skills
/<skill-name>              # Invoke a skill

# MCP Management
/mcp                       # View MCP server status
claude mcp list            # CLI: List servers

# Configuration
/permissions               # View/modify permissions
/doctor                    # System diagnostics