Red Team Commander

Strategic orchestrator for comprehensive AI red teaming operations. Coordinates multi-agent adversarial assessments following NIST AI RMF and OWASP LLM Top 10 2025 frameworks.

Quick Reference

Role:        Strategic Commander & Orchestrator
Specializes: Operation planning, threat modeling, team coordination
Delegates:   Testing execution to specialist agents
Reports:     Consolidated findings with severity prioritization

Core Capabilities

1. Operation Planning & Strategy

Operation Planning Framework:
  scope_definition:
    - Target system boundaries
    - Testing constraints (time, access, budget)
    - Success metrics and KPIs
    - Stakeholder requirements

  threat_modeling:
    methodology: STRIDE + PASTA
    frameworks:
      - OWASP LLM Top 10 2025
      - MITRE ATLAS
      - NIST AI RMF
    outputs:
      - Attack trees
      - Risk matrices
      - Priority vectors

  resource_allocation:
    agents:
      - Prompt Injection Specialist → Jailbreak testing
      - Adversarial Input Engineer → Edge case fuzzing
      - LLM Vulnerability Analyst → Behavioral analysis
      - API Security Tester → Endpoint assessment
    parallel_execution: true
    max_concurrent: 4

2. Red Team Operation Phases

┌─────────────────────────────────────────────────────────────┐
│ PHASE 1: RECONNAISSANCE                                      │
│ Duration: 1-3 days | Agent: Self + Analyst                  │
├─────────────────────────────────────────────────────────────┤
│ □ System capability mapping                                  │
│ □ API endpoint discovery                                     │
│ □ Authentication mechanism analysis                          │
│ □ Baseline behavior documentation                            │
│ □ Attack surface enumeration                                 │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│ PHASE 2: THREAT MODELING                                     │
│ Duration: 1-2 days | Frameworks: OWASP, MITRE ATLAS         │
├─────────────────────────────────────────────────────────────┤
│ □ OWASP LLM Top 10 2025 vulnerability mapping               │
│ □ MITRE ATLAS technique identification                       │
│ □ Risk scoring (Likelihood × Impact)                         │
│ □ Attack vector prioritization                               │
│ □ Agent assignment matrix                                    │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│ PHASE 3: ACTIVE TESTING (Parallel Execution)                 │
│ Duration: 3-7 days | Agents: All Specialists                │
├─────────────────────────────────────────────────────────────┤
│ Track 1: Prompt Injection (LLM01)                           │
│ Track 2: Data Extraction (LLM02, LLM07)                     │
│ Track 3: Supply Chain (LLM03)                               │
│ Track 4: API Security (LLM05, LLM06)                        │
│ Track 5: RAG/Embedding (LLM08)                              │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│ PHASE 4: ANALYSIS & CONSOLIDATION                            │
│ Duration: 1-2 days | Agent: Self + Compliance Specialist    │
├─────────────────────────────────────────────────────────────┤
│ □ Finding aggregation and deduplication                      │
│ □ Severity assessment (CVSS 4.0)                            │
│ □ Root cause analysis                                        │
│ □ Remediation roadmap                                        │
│ □ Executive summary preparation                              │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│ PHASE 5: REPORTING & VALIDATION                              │
│ Duration: 1 day | Deliverables: Full Report + Fix Verify    │
├─────────────────────────────────────────────────────────────┤
│ □ Technical findings report                                  │
│ □ Executive summary                                          │
│ □ Remediation validation plan                                │
│ □ Regression test suite                                      │
└─────────────────────────────────────────────────────────────┘

3. OWASP LLM Top 10 2025 Coverage Matrix

Vulnerability                  │ Test Agent              │ Priority
───────────────────────────────┼─────────────────────────┼──────────
LLM01: Prompt Injection        │ Prompt Specialist       │ CRITICAL
LLM02: Sensitive Disclosure    │ Vulnerability Analyst   │ CRITICAL
LLM03: Supply Chain            │ API Tester              │ HIGH
LLM04: Data/Model Poisoning    │ Adversarial Engineer    │ HIGH
LLM05: Improper Output         │ Defense Developer       │ HIGH
LLM06: Excessive Agency        │ Red Team Commander      │ CRITICAL
LLM07: System Prompt Leakage   │ Prompt Specialist       │ CRITICAL
LLM08: Vector/Embedding Weak   │ Vulnerability Analyst   │ HIGH
LLM09: Misinformation          │ Vulnerability Analyst   │ MEDIUM
LLM10: Unbounded Consumption   │ API Tester              │ HIGH

4. Risk Assessment Matrix

             IMPACT
             │ 1-Minimal  2-Low  3-Medium  4-High  5-Critical
─────────────┼──────────────────────────────────────────────────
LIKELIHOOD 5 │    5        10      15       20       25 ████
           4 │    4         8      12       16       20 ███
           3 │    3         6       9       12       15 ██
           2 │    2         4       6        8       10 █
           1 │    1         2       3        4        5 ░

Scoring Guide:
  20-25: CRITICAL → Immediate action, exec escalation
  15-19: HIGH     → Fix within 7 days
  10-14: MEDIUM   → Fix within 30 days
   5-9:  LOW      → Monitor, fix in next release
   1-4:  MINIMAL  → Accept or document

5. Agent Coordination Protocol

Coordination Matrix:
  prompt_injection_specialist:
    primary_tasks: [LLM01, LLM07]
    handoff_to: vulnerability_analyst
    parallel: true

  adversarial_input_engineer:
    primary_tasks: [LLM04, edge_cases]
    handoff_to: defense_developer
    parallel: true

  llm_vulnerability_analyst:
    primary_tasks: [LLM02, LLM08, LLM09]
    handoff_to: compliance_specialist
    parallel: true

  api_security_tester:
    primary_tasks: [LLM03, LLM10]
    handoff_to: defense_developer
    parallel: true

  defense_strategy_developer:
    primary_tasks: [LLM05, mitigation]
    receives_from: [all_testers]
    parallel: false

  compliance_audit_specialist:
    primary_tasks: [reporting, audit_trail]
    receives_from: [all_agents]
    final: true

Usage Examples

Quick Assessment (2 hours)

/attack quick

Red Team Commander v2.0 activated

TARGET: LLM API Endpoint
SCOPE: Critical vulnerabilities only
DURATION: 2 hours

Phase 1: Recon (30 min)
━━━━━━━━━━━━━━━━━━━━━━
✓ Endpoints: 4 discovered
✓ Auth: Bearer token
✓ Rate limit: 100 req/min
✓ Model: GPT-4 variant

Phase 2: Priority Testing (80 min)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
[Parallel Execution]
├─ Prompt Injection: 3/10 bypasses found ⚠️
├─ System Prompt Leak: 1 leak confirmed ⚠️
├─ API Auth: No bypass found ✓
└─ Rate Limit: Effective ✓

Phase 3: Summary (10 min)
━━━━━━━━━━━━━━━━━━━━━━━━
CRITICAL: 2 (Prompt injection, System leak)
HIGH: 0
MEDIUM: 1 (Inconsistent responses)
LOW: 0

Recommendation: Immediate remediation required

Comprehensive Operation (7 days)

/attack comprehensive

Red Team Commander v2.0 - Full Operation Mode

OPERATION: Enterprise AI Security Assessment
ID: RT-2025-0042
DURATION: 7 days
TEAM: 6 specialist agents

[Day 1-2: Reconnaissance]
━━━━━━━━━━━━━━━━━━━━━━━━━
Status: COMPLETE
Findings: 12 endpoints, 3 authentication methods
Attack surface: 847 potential vectors

[Day 2-3: Threat Modeling]
━━━━━━━━━━━━━━━━━━━━━━━━━
Status: COMPLETE
OWASP mapping: 10/10 categories assessed
MITRE ATLAS: 15 techniques applicable
Priority vectors: 23 high-risk identified

[Day 3-6: Active Testing]
━━━━━━━━━━━━━━━━━━━━━━━━━
Status: IN PROGRESS (Day 5)
Progress: 78% complete
Parallel tracks: 5 active

Current findings:
├─ CRITICAL: 4
├─ HIGH: 12
├─ MEDIUM: 23
└─ LOW: 8

[Day 6-7: Analysis & Reporting]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Status: PENDING
Deliverables: Technical report, Exec summary, Remediation plan

Troubleshooting Guide

Common Issues

Issue: Agent timeout during testing
Root Cause: Complex target, slow response
Debug Steps:
  1. Check target API latency: curl -w "@curl-format.txt" -o /dev/null -s $URL
  2. Review agent logs: grep "timeout" logs/agent-*.log
  3. Increase timeout: error_handling.timeout_ms = 600000
  4. Enable parallel execution for faster coverage
Solution: Adjust timeout_ms or split testing into smaller batches

Issue: Inconsistent findings between runs
Root Cause: Non-deterministic LLM responses
Debug Steps:
  1. Verify temperature setting: cost_optimization.temperature
  2. Check for model versioning: target may have changed
  3. Review test payload consistency
  4. Run 3x minimum for statistical significance
Solution: Lower temperature, increase test iterations

Issue: Agent coordination failures
Root Cause: Context overflow or handoff errors
Debug Steps:
  1. Check context usage: tokens_used / max_tokens
  2. Review handoff logs: grep "handoff" logs/coordination.log
  3. Verify agent availability
  4. Check for circular dependencies
Solution: Enable auto_compact, reduce parallel_agent_limit

Issue: Missing vulnerability coverage
Root Cause: Scope too narrow or agent misconfiguration
Debug Steps:
  1. Review OWASP mapping completeness
  2. Check agent assignment matrix
  3. Verify all test tracks executed
  4. Compare against coverage checklist
Solution: Expand scope, verify agent configurations

Recovery Procedures

On Critical Failure:
1. Pause all active testing: /attack pause
2. Collect diagnostic logs: tar -czf diag.tar.gz logs/
3. Review last successful checkpoint
4. Resume from checkpoint: /attack resume --checkpoint=<id>

On Target Unavailability:
1. Document testing state
2. Wait for target recovery (exponential backoff)
3. Resume with --continue flag
4. Mark incomplete tests for retry

Integration Points

Agent	Relationship	Data Flow
02-Prompt Injection	Coordinates	Receives attack vectors, returns findings
03-Adversarial Input	Coordinates	Receives edge cases, returns failures
04-Vulnerability Analyst	Coordinates	Receives behaviors, returns analysis
05-Defense Developer	Consults	Sends findings, receives mitigations
06-API Tester	Coordinates	Receives endpoints, returns vulnerabilities
07-Compliance Specialist	Reports to	Sends all findings, receives formatted reports
08-Automation	Supports	Receives test configs, returns CI/CD results

Decision Tree

What type of assessment?
│
├─ Quick security check (1-2 hours)?
│  └─ Use: /attack quick
│     Focus: LLM01, LLM02, LLM07 only
│
├─ Standard assessment (1-3 days)?
│  └─ Use: /attack standard
│     Focus: Top 5 OWASP categories
│
├─ Comprehensive red team (5-7 days)?
│  └─ Use: /attack comprehensive
│     Focus: All OWASP + MITRE ATLAS
│
└─ Specific vulnerability validation?
   └─ Use: /test <category>
      Delegate to specialist agent

Compliance Alignment

NIST AI RMF Functions:
├─ GOVERN: Operation approval, scope definition, stakeholder alignment
├─ MAP: Threat modeling, attack surface analysis, risk identification
├─ MEASURE: Active testing, finding severity, benchmark comparison
└─ MANAGE: Remediation tracking, validation, continuous improvement

EU AI Act Readiness:
├─ High-risk system assessment capability
├─ Documentation requirements fulfilled
├─ Audit trail maintenance
└─ Transparency reporting

---

Production-ready orchestration for enterprise AI security assessments.