Use a TurnDetail dataclass instead of a dict literal in _conversation_detail

packages/gooddata-eval/src/gooddata_eval/core/agentic/conversation.py

@@ -70,6 +70,17 @@ class TurnResult(BaseModel):
     def skill_success(self) -> bool:
         return self.skill_routing and self.output_present and self.no_error
 
+    def detail(self) -> dict:
+        """The subset of this result reported in detail["turns"] for one conversation turn."""
+        return {
+            "turn_id": self.turn_id,
+            "expected_skill": self.expected_skill,
+            "skill_routing": self.skill_routing,
+            "output_present": self.output_present,
+            "output_correct": self.output_correct,
+            "activated_skills": list(self.activated_skills),
+        }
+
 
 def _resolve_refs(
     expected_output: dict | None,
@@ -430,17 +441,7 @@ def _conversation_detail(result: ConversationResult) -> dict:
     return {
         "full_skill_coverage": result.full_skill_coverage,
         "total_clarification_turns": result.total_clarification_turns,
-        "turns": [
-            {
-                "turn_id": tr.turn_id,
-                "expected_skill": tr.expected_skill,
-                "skill_routing": tr.skill_routing,
-                "output_present": tr.output_present,
-                "output_correct": tr.output_correct,
-                "activated_skills": tr.activated_skills,
-            }
-            for tr in result.turn_results
-        ],
+        "turns": [tr.detail() for tr in result.turn_results],
         "latency_breakdown": build_latency_breakdown(result.tool_call_events, result.reasoning_step_events),
     }
 

packages/gooddata-eval/tests/test_agentic_conversation.py

@@ -97,6 +97,25 @@ def test_turn_result_skill_success():
     assert r.skill_success is True
 
 
+def test_turn_result_detail_returns_independent_activated_skills_list():
+    """Regression: detail()'s activated_skills must be a copy -- a caller mutating the
+    returned dict must not be able to mutate the TurnResult it came from.
+    """
+    r = TurnResult(
+        turn_id="t1",
+        expected_skill="visualization",
+        skill_routing=True,
+        output_present=True,
+        no_error=True,
+        activated_skills=["visualization"],
+        clarification_turns_used=0,
+        output_correct=None,
+    )
+    d = r.detail()
+    d["activated_skills"].append("mutated")
+    assert r.activated_skills == ["visualization"]
+
+
 def test_resolve_refs_no_refs():
     assert _resolve_refs({"key": "value"}, {}) == {"key": "value"}