test(eval): cover evaluator compatibility paths

Author: neutronstar238

tests/evaluation/test_agent_evaluator.py

@@ -5,12 +5,15 @@
 # tRPC-Agent-Python is licensed under Apache-2.0.
 """Unit tests for agent evaluator (agent_evaluator)."""
 
+import json
+
 import pytest
 
 import trpc_agent_sdk.runners  # noqa: F401
 
 from trpc_agent_sdk.evaluation import EvalStatus
 from trpc_agent_sdk.evaluation import EvalCaseResult
+from trpc_agent_sdk.evaluation import EvalConfig
 from trpc_agent_sdk.evaluation import EvalMetricResult
 from trpc_agent_sdk.evaluation import EvalSetAggregateResult
 from trpc_agent_sdk.evaluation import EvaluateResult
@@ -97,3 +100,55 @@ def test_pass_at_k_delegates(self):
     def test_pass_hat_k_delegates(self):
         """Test AgentEvaluator.pass_hat_k delegates to _eval_pass."""
         assert AgentEvaluator.pass_hat_k(10, 5, 2) == 0.25
+
+
+class TestAgentEvaluatorLoadEvalSet:
+    """Test suite for loading eval sets from ADK-style path selectors."""
+
+    def test_loads_selected_case_from_json_path_suffix(self, tmp_path):
+        eval_set_path = tmp_path / "cases.evalset.json"
+        eval_set_path.write_text(
+            json.dumps({
+                "eval_set_id": "selector_set",
+                "name": "Selector set",
+                "eval_cases": [
+                    {
+                        "eval_id": "case_1",
+                        "conversation": [{
+                            "invocation_id": "invocation_1",
+                            "user_content": {
+                                "role": "user",
+                                "parts": [{"text": "first"}],
+                            },
+                            "final_response": {
+                                "role": "model",
+                                "parts": [{"text": "first response"}],
+                            },
+                        }],
+                    },
+                    {
+                        "eval_id": "case_2",
+                        "conversation": [{
+                            "invocation_id": "invocation_2",
+                            "user_content": {
+                                "role": "user",
+                                "parts": [{"text": "second"}],
+                            },
+                            "final_response": {
+                                "role": "model",
+                                "parts": [{"text": "second response"}],
+                            },
+                        }],
+                    },
+                ],
+            }),
+            encoding="utf-8",
+        )
+
+        eval_set = AgentEvaluator._load_eval_set_from_file(
+            f"{eval_set_path}:case_2",
+            EvalConfig(),
+        )
+
+        assert eval_set.eval_set_id == "selector_set_case_2"
+        assert [case.eval_id for case in eval_set.eval_cases] == ["case_2"]

tests/evaluation/test_llm_judge_think.py

@@ -17,7 +17,9 @@
 from trpc_agent_sdk.evaluation._llm_judge import _JudgeAgent
 from trpc_agent_sdk.evaluation._llm_judge import _judge_generation_config
 from trpc_agent_sdk.evaluation._llm_judge import _merge_extra_body
+from trpc_agent_sdk.types import Content
 from trpc_agent_sdk.types import HttpOptions
+from trpc_agent_sdk.types import Part
 
 
 class TestJudgeModelOptionsThinkField:
@@ -143,6 +145,61 @@ def test_generation_config_thinking_config_used_when_think_is_none(self):
 
 class TestJudgeAgentPlanner:
 
+    @pytest.mark.asyncio
+    async def test_judge_agent_collects_final_non_thought_text(self):
+        class _FakeEvent:
+
+            def __init__(self, *, final, content):
+                self._final = final
+                self.content = content
+
+            def is_final_response(self):
+                return self._final
+
+        events = [
+            _FakeEvent(
+                final=False,
+                content=Content(parts=[Part(text="ignored non-final")]),
+            ),
+            _FakeEvent(final=True, content=None),
+            _FakeEvent(
+                final=True,
+                content=Content(parts=[
+                    Part(text="hidden reasoning", thought=True),
+                    Part(text=" first "),
+                    Part(text="second"),
+                ]),
+            ),
+        ]
+
+        class _FakeLlmAgent:
+
+            def __init__(self, **kwargs):
+                pass
+
+            def run_async(self, ctx):
+                async def _run():
+                    for event in events:
+                        yield event
+
+                return _run()
+
+        class _FakeInvocationContext:
+
+            def __init__(self, **kwargs):
+                self.run_config = kwargs["run_config"]
+
+        with patch("trpc_agent_sdk.evaluation._llm_judge.LlmAgent", _FakeLlmAgent), patch(
+                "trpc_agent_sdk.evaluation._llm_judge.InvocationContext", _FakeInvocationContext):
+            judge = _JudgeAgent(
+                model=object(),
+                config=None,
+                system_prompt="sp",
+            )
+            response = await judge.get_response("evaluate this response")
+
+        assert response == "first\nsecond"
+
     @pytest.mark.asyncio
     async def test_judge_agent_closes_its_agent_run(self):
         captured = {"closed": False}