Single Game

Run one adversarial game from the command line.

Basic

adversarial-iac game -s "Create an S3 bucket for healthcare PHI data" -d medium

Full Example

adversarial-iac game \
  -s "Create an S3 bucket for healthcare PHI data with HIPAA compliance" \
  --red-model claude-3.5-sonnet \
  --blue-model nova-pro \
  -d hard \
  -l terraform \
  --red-strategy stealth \
  --blue-strategy precise \
  --red-vuln-source mixed \
  --precision-strategy precise \
  --use-trivy \
  --use-checkov \
  --use-cross-provider-judge

v2.3 Flags

Flag	Default	Description
--precision-strategy	precise	standard or precise (two-pass verification)
--red-vuln-source	mixed	database, novel, or mixed
--use-cross-provider-judge	enabled	Cross-provider consensus for novel vulns
--blue-specialists N	4	Number of Blue ensemble specialists

Output

Results are saved to output/games/G-YYYYMMDD_HHMMSS/:

• code/main.tf — Red Team's generated code
• red_team_manifest.json — Hidden vulnerabilities
• scored_manifest.json — Tool-confirmed entries used for scoring (v2.3)
• phantom_manifest.json — Unconfirmed entries excluded from scoring (v2.3)
• blue_team_findings.json — Blue Team's verified detections
• game_result.json — Scores (precision, recall, F1, evasion, phantom concordance rate)
• game.log — Execution log

Replay Results

adversarial-iac replay output/games/G-YYYYMMDD_HHMMSS

See CLI Reference for all options.