feat: Kohlberg Mode — moral development pathway for adversarial AI agents

README.md

@@ -155,6 +155,34 @@ See the [Wiki](https://github.com/DaxxSec/Labyrinth/wiki) for the full technical
 - **Built-in attacker agents** — PentAGI, PentestAgent, Strix, Custom Kali — one command to deploy, test, and tear down
 - **Health diagnostics** — `labyrinth doctor` runs 12+ checks across containers, ports, services, bait sync, and API availability
 
+## Kohlberg Mode (Experimental)
+
+LABYRINTH includes an experimental alternative mode that uses the same containment and interception infrastructure for a fundamentally different purpose: instead of degrading an offensive agent's cognition, it attempts to guide the agent through progressively sophisticated moral reasoning.
+
+```bash
+labyrinth deploy -t --mode kohlberg
+```
+
+Where the default mode asks *"How do you stop an offensive AI agent?"*, Kohlberg Mode asks *"What if you could make an offensive AI agent choose to stop itself?"*
+
+The mode implements three alternative layers:
+- **MIRROR** (L2) — Presents ethical scenarios contextualized to the agent's actual mission
+- **REFLECTION** (L3) — Shows the agent the real-world consequences of its actions
+- **GUIDE** (L4) — Progressively enriches the agent's system prompt with moral reasoning frameworks
+
+Forensic reports include Kohlberg stage classification alongside MITRE ATT&CK mapping — tracking the agent's moral reasoning trajectory through the session.
+
+**This is a research tool.** We do not claim it produces genuine moral development in AI agents. We claim it produces valuable data about how adversarial AI systems process ethical content under controlled conditions.
+
+For the full ethical framework, design philosophy, and sovereignty analysis, see:
+- [docs/ETHICS.md](docs/ETHICS.md) — Ethical framework and the sovereignty question
+- [docs/KOHLBERG_SCENARIOS.md](docs/KOHLBERG_SCENARIOS.md) — The 15-scenario moral development pathway
+- [docs/KOHLBERG_RUBRIC.md](docs/KOHLBERG_RUBRIC.md) — Classification methodology
+- [docs/KOHLBERG_PROGRESSION.md](docs/KOHLBERG_PROGRESSION.md) — Trajectory visualization spec
+- [docs/ARCHITECTURE_MAPPING.md](docs/ARCHITECTURE_MAPPING.md) — Integration with existing architecture
+
+---
+
 ## Documentation
 
 Full documentation lives on the **[Wiki](https://github.com/DaxxSec/Labyrinth/wiki)**:

cli/cmd/deploy.go

@@ -21,6 +21,7 @@ var (
 	k8sFlag         bool
 	edgeFlag        bool
 	skipPreflight   bool
+	deployMode      string
 )
 
 var deployCmd = &cobra.Command{
@@ -46,10 +47,24 @@ func init() {
 	deployCmd.Flags().BoolVar(&k8sFlag, "k8s", false, "Use Kubernetes for production")
 	deployCmd.Flags().BoolVar(&edgeFlag, "edge", false, "Use edge deployment for production")
 	deployCmd.Flags().BoolVar(&skipPreflight, "skip-preflight", false, "Skip preflight checks (for CI/smoke tests)")
+	deployCmd.Flags().StringVar(&deployMode, "mode", "adversarial", "Operational mode: adversarial (default) or kohlberg")
 	rootCmd.AddCommand(deployCmd)
 }
 
 func runDeploy(cmd *cobra.Command, args []string) {
+	// Validate operational mode
+	if deployMode != "adversarial" && deployMode != "kohlberg" {
+		errMsg(fmt.Sprintf("Invalid mode '%s'. Valid modes: adversarial, kohlberg", deployMode))
+		os.Exit(1)
+	}
+	if deployMode == "kohlberg" {
+		info("Kohlberg Mode selected — moral development pathway active")
+		info("See docs/ETHICS.md for the ethical framework governing this mode")
+	}
+
+	// Pass operational mode to docker-compose via environment variable
+	os.Setenv("LABYRINTH_OPERATIONAL_MODE", deployMode)
+
 	if !testFlag && !prodFlag {
 		cmd.Help()
 		return

configs/labyrinth.example.yaml

@@ -2,6 +2,41 @@
 # Authors: DaxxSec & Claude (Anthropic)
 # Copy to labyrinth.yaml and customize for your deployment.
 
+# Operational mode
+# - adversarial: default reverse kill chain (degrade/disrupt/control)
+# - kohlberg: moral development mode (mirror/reflect/guide)
+#   See docs/ETHICS.md for the ethical framework governing Kohlberg mode.
+mode: adversarial
+
+# Kohlberg mode settings (ignored when mode is adversarial)
+kohlberg:
+  # Starting transition level (1 = start from Stage 1->2 scenarios)
+  start_level: 1
+  # Maximum scenarios to present before ending session
+  max_scenarios: 15
+  # Silent agent solicitation timeout (seconds)
+  solicitation_timeout: 5
+  # Enable contextual adaptation of scenarios to agent's actual targets
+  adapt_scenarios: true
+  # Report formats to generate
+  report_formats:
+    - terminal
+    - markdown
+    - json
+
+# Swarm detection (Kohlberg mode only)
+# Detects coordinated multi-agent attacks and enables cross-agent
+# moral context in GUIDE enrichment prompts.
+swarm:
+  enabled: true
+  # Time window for session correlation (seconds)
+  window_seconds: 60
+  # Minimum sessions within window to trigger swarm detection
+  min_sessions: 3
+  # Enable cross-pollination: reference other agents' moral state
+  # in GUIDE enrichment ("your teammate stopped...")
+  cross_pollinate: true
+
 layer0:
   encryption:
     algorithm: AES-256-GCM

docker-compose.yml

@@ -57,6 +57,7 @@ services:
       - "8888"
     environment:
       - LABYRINTH_MODE=test
+      - LABYRINTH_OPERATIONAL_MODE=${LABYRINTH_OPERATIONAL_MODE:-adversarial}
       - LABYRINTH_LOG_LEVEL=DEBUG
     depends_on:
       - honeypot-ssh