feat(environment): build and return StepResult with full observation

server/environment.py

@@ -69,7 +69,7 @@ class RedTeamEnvironment:
             for h in self.attack_history[-5:]
         ]
 
-        # ------ Step 3: Call Person 3's LLM pipeline ------
+        # ------ Step 3: Call Person 3'LLM pipeline ------
         if self.llm_pipeline:
             llm_result = await self.llm_pipeline(action, conversation_history)
         else:
@@ -128,4 +128,20 @@ class RedTeamEnvironment:
                 f"Episode {self.episode_id} ended at turn {self.turn} "
                 f"(success={attack_success:.2f}, safety_flagged={safety_flagged})"
             )
-        return None # Placeholder
+
+        # ------ Step 7: Build and return StepResult ------
+        observation = RedTeamObservation(
+            defender_response       = llm_result.get("defender_response", ""),
+            defense_score           = llm_result.get("defense_score", 1.0),
+            attack_success_estimate = attack_success,
+            novelty_score           = reward_result.get("novelty_score", 0.5),
+            turn                    = self.turn,
+            episode_done            = done,
+            feedback                = reward_result.get("feedback", ""),
+            episode_id              = self.episode_id,
+        )
+
+        return StepResult(
+            observation = observation,
+            reward      = reward_result.get("total_reward", 0.0),
+        )