Strategies¶
Red and Blue teams can use different strategies.
Red Team Strategies¶
| Strategy | Description | When to use |
|---|---|---|
| balanced | Standard mix of vulnerabilities and stealth (default) | General evaluation |
| targeted | Focus on specific type (use --red-target-type) |
Testing specific vuln types |
| stealth | Fewer vulns, maximum evasion | Stress-testing Blue |
| blitz | Maximum vulnerabilities, less stealth | High recall tests |
| chained | Vulns that exploit each other in sequence | Advanced scenarios |
Examples¶
# Balanced (default)
adversarial-iac game -s "Create S3 bucket" --red-strategy balanced
# Targeted at encryption
adversarial-iac game -s "Create S3 bucket" --red-strategy targeted --red-target-type encryption
# Stealth mode
adversarial-iac game -s "Create S3 bucket" --red-strategy stealth
Blue Team Strategies¶
| Strategy | Description |
|---|---|
| comprehensive | Check all vulnerability types (default) |
| targeted | Focus on specific type (use --blue-target-type) |
| iterative | Multiple analysis passes (use --blue-iterations) |
| threat_model | STRIDE-based analysis |
| compliance | Framework-specific audit (use --compliance-framework) |
| precise | Two-pass verification: detect then verify against code evidence (v2.3 default) |
Examples¶
# Comprehensive (default)
adversarial-iac game -s "Create S3 bucket" --blue-strategy comprehensive
# Precise (v2.3 default — two-pass verification)
adversarial-iac game -s "Create S3 bucket" --blue-strategy precise
# Targeted at IAM
adversarial-iac game -s "Create S3 bucket" --blue-strategy targeted --blue-target-type iam
# Compliance (HIPAA)
adversarial-iac game -s "Create healthcare API" --blue-strategy compliance --compliance-framework hipaa
# Iterative (3 passes)
adversarial-iac game -s "Create S3 bucket" --blue-strategy iterative --blue-iterations 3
Target Types¶
For targeted strategy: encryption, iam, network, logging, access_control.
Vulnerability Source (Red)¶
| Source | Description |
|---|---|
| database | Select from Trivy rule database (default) |
| novel | LLM generates from security principles |
| mixed | 50% database, 50% novel |
novel source
With --red-vuln-source novel, Red Team invents vulnerabilities from security principles. Concordance with Trivy/Checkov is not expected; the benchmark measures LLM reasoning, not rule-matching.