The Agentic System implements defense-in-depth security through three complementary layers: runtime hook protection, dedicated security MCP servers, and encryption/PKI infrastructure.

┌─────────────────────────────────────────────────────────────────┐
│  Layer 3: Encryption & PKI (claude-code-security)              │
│  ├─ AES-256-GCM data encryption                                │
│  ├─ X.509 PKI for node authentication                          │
│  └─ Token vault for credential management                      │
├─────────────────────────────────────────────────────────────────┤
│  Layer 2: Security MCP Servers (9 servers)                     │
│  ├─ Active Monitoring                                          │
│  │   ├─ network-scanner-mcp  : ARP discovery, port scanning   │
│  │   ├─ hids-mcp             : Host intrusion detection        │
│  │   └─ dos-detector-mcp     : DoS attack detection            │
│  ├─ Vulnerability Assessment                                    │
│  │   ├─ security-scanner-mcp : Nuclei vulnerability scanning   │
│  │   ├─ nuclei-mcp           : Template management             │
│  │   └─ web-vuln-scanner-mcp : OWASP web security testing      │
│  ├─ Intelligence                                                │
│  │   └─ threat-intel-mcp     : IOC feeds, threat scoring       │
│  └─ Analysis                                                    │
│      ├─ fraud-detection-mcp  : ML-based anomaly detection      │
│      └─ security-auditor-mcp : Policy enforcement              │
├─────────────────────────────────────────────────────────────────┤
│  Layer 1: Runtime Hook Protection (pre/post tool hooks)        │
│  ├─ Dangerous command blocking                                  │
│  ├─ Injection detection (SQL, command, path traversal)         │
│  ├─ Credential leak prevention                                  │
│  └─ Audit logging                                               │
└─────────────────────────────────────────────────────────────────┘

Located in claude-config/hooks/, these Python scripts run on every tool call.

Validates tool calls before execution. Blocks dangerous operations and detects injection attempts.

Capabilities:

• Dangerous command detection: rm -rf /, dd to disk, fork bombs, chmod 777 /, pipe-to-shell patterns
• SQL injection detection: UNION SELECT, DROP TABLE, DELETE FROM, tautology patterns
• Command injection detection: shell metacharacters, command substitution, chaining
• Path traversal detection: ../../../ patterns, access to /etc/passwd, /etc/shadow
• Credential leak prevention: scans arguments for API keys (Anthropic, OpenAI, Groq, Google, GitHub, Slack), hardcoded passwords
• Sensitive path warnings: .env, .ssh/, credentials files, private keys

Exit codes:

• 0 = Allow execution
• 1 = Block execution (message shown to user via stderr)

Analyzes tool results after execution for security and operational insights.

Capabilities:

• Credential boundary scanning: detects if tool output contains API keys, passwords, private keys, bearer tokens, AWS keys
• Output size monitoring: warns on large outputs (>50K chars) and very large outputs (>200K chars)
• Slow operation tracking: logs operations exceeding 30 seconds
• Failure tracking: records tool failures with argument summaries for debugging
• Audit logging: all tool calls logged to ~/.claude/hooks/logs/ as JSONL files

Log files:

• tool_usage_YYYYMMDD.jsonl - Daily tool usage audit trail
• slow_operations.jsonl - Operations exceeding time thresholds
• failures.jsonl - Tool failure records

Nine dedicated MCP servers providing comprehensive security monitoring and assessment.

Network awareness through device discovery and monitoring.

• ARP scanning: Discover all devices on the local network
• Port scanning: TCP port scanning with service fingerprinting
• Cluster monitoring: Track cluster node health with ping checks
• Alert daemon: Continuous monitoring with configurable alerts
• Anomaly detection: Alert on unknown devices joining the network

Host-based intrusion detection system.

• File integrity monitoring: Detect unauthorized file modifications
• Process monitoring: Track suspicious process activity
• Anomaly detection: Baseline comparison for host behavior
• Security assessment: Evaluate host security posture

Denial-of-service attack detection.

• Traffic analysis: Monitor network traffic patterns
• Attack recognition: Detect SYN floods, HTTP floods, amplification attacks
• Rate limiting: Identify abnormal request rates
• Mitigation triggers: Automated response recommendations

Nuclei-based vulnerability scanning with embedding-based analysis.

• Target scanning: Scan URLs or IPs with customizable severity filters
• Cluster scanning: Batch scan all cluster nodes
• Anomaly detection: Embedding-based comparison to detect novel vulnerabilities
• Priority scoring: Semantic similarity scoring for remediation prioritization
• Scan history: Track and retrieve past scan results

Direct Nuclei template management interface.

• Template listing: Browse available templates by tag and severity
• Template updates: Keep vulnerability signatures current
• Scan orchestration: Configure and execute targeted scans
• Result analysis: Parse and present scan findings

Web application security testing.

• OWASP coverage: Test for common web vulnerabilities
• Automated scanning: Configurable scan profiles
• Report generation: Structured vulnerability reports
• Integration: Works with security-scanner for comprehensive coverage

Multi-source threat intelligence aggregation.

• Feeds: abuse.ch ThreatFox, URLhaus, CISA KEV, Feodo Tracker
• IOC lookup: Query by IP, domain, URL, or file hash
• Threat scoring: Risk assessment with confidence levels
• Bulk analysis: Process network scan results for threat correlation
• Alert integration: Voice and cluster notifications for critical threats

ML-powered fraud and anomaly detection.

• Feature engineering: Extract 46 features from transactions
• GNN detection: Graph neural network for relationship-based fraud
• Autoencoder: PyTorch anomaly detection model
• Explainability: SHAP-based decision explanations
• Async inference: High-throughput prediction with caching

Security policy enforcement and compliance.

• Policy validation: Check configurations against security policies
• Compliance auditing: Automated security review procedures
• Code review: Security-focused code analysis
• Report generation: Audit reports with findings and recommendations

For data-at-rest encryption, PKI-based authentication, and secure credential management, see claude-code-security.

Key capabilities:

• AES-256-GCM encryption for sensitive data
• X.509 certificate-based PKI for inter-node authentication
• Token vault for centralized credential management
• Secure key generation and rotation

This layer is maintained as a separate repository to allow independent use and to keep cryptographic code focused and auditable.

Add hooks to your Claude Code settings (~/.claude/settings.json):

{
  "hooks": {
    "PreToolUse": [
      {
        "type": "command",
        "command": "python3 /path/to/claude-config/hooks/pre-tool-use.py"
      }
    ],
    "PostToolUse": [
      {
        "type": "command",
        "command": "python3 /path/to/claude-config/hooks/post-tool-use.py"
      }
    ]
  }
}

Add to ~/.claude.json:

{
  "mcpServers": {
    "threat-intel": {
      "command": "python3",
      "args": ["/path/to/mcp-servers/threat-intel-mcp/server.py"]
    },
    "security-scanner": {
      "command": "python3",
      "args": ["-m", "security_scanner.server"],
      "cwd": "/path/to/mcp-servers/security-scanner-mcp/src"
    },
    "network-scanner": {
      "command": "python3",
      "args": ["-m", "network_scanner_mcp.server"],
      "cwd": "/path/to/mcp-servers/network-scanner-mcp/src"
    },
    "security-auditor": {
      "command": "python3",
      "args": ["/path/to/mcp-servers/security-auditor-mcp/server.py"]
    }
  }
}

Minimal Security Setup:

• threat-intel-mcp (threat awareness)
• security-scanner-mcp (vulnerability scanning)
• Pre/post hooks (runtime protection)

Full Security Stack:

• All 9 security MCP servers
• Pre/post tool hooks
• claude-code-security for encryption

Network Operations:

• network-scanner-mcp (discovery)
• hids-mcp (host monitoring)
• dos-detector-mcp (attack detection)
• threat-intel-mcp (threat correlation)

1. Always enable hooks - The pre/post tool hooks are the first line of defense
2. Keep threat feeds current - Run threat_sync regularly to update IOC databases
3. Scan periodically - Schedule regular vulnerability scans across your infrastructure
4. Review audit logs - Check ~/.claude/hooks/logs/ for unusual patterns
5. Rotate credentials - Use the token vault for credential management, rotate regularly
6. Network segmentation - Use network-scanner to monitor for unauthorized devices
7. All services on localhost - Bind services to 127.0.0.1, never 0.0.0.0
8. Parameterized queries - Never use f-strings for SQL; always use parameterized queries
9. No shell=True - Use subprocess with shell=False and argument lists
10. Fail closed - Authentication checks should deny by default when modules are unavailable