From ecdc7fdd335382fd46fc1d45979965c0f74be45b Mon Sep 17 00:00:00 2001
From: cthurston-clgx <cthurston@corelogic.com>
Date: Tue, 5 May 2026 12:00:51 -0500
Subject: [PATCH] fix(evaluation): populate developer_instructions when
 invocation_events is empty (#5593)

The rubric_based_final_response_quality_v1 evaluator fails to pass
developer_instructions to the judge when the agent makes zero tool
calls (empty invocation_events list). This occurs because the agent
name is resolved exclusively from invocation_events[0].author, with
no fallback for the zero-event case.

This is critical for evaluating out-of-scope rejection behavior where
an agent correctly declines a request without calling any tools. The
judge receives an empty <developer_instructions> block and cannot
validate rubrics that reference the system prompt.

Fix: When invocation_events is empty, fall back to the first agent
name in app_details.agent_details to resolve developer_instructions.
This mirrors how hallucinations_v1.py handles the same scenario.

Fixes https://github.com/google/adk-python/issues/5593
---
 .../rubric_based_final_response_quality_v1.py | 17 ++++-
 ..._rubric_based_final_response_quality_v1.py | 63 +++++++++++++++++++
 2 files changed, 77 insertions(+), 3 deletions(-)

diff --git a/src/google/adk/evaluation/rubric_based_final_response_quality_v1.py b/src/google/adk/evaluation/rubric_based_final_response_quality_v1.py
index df01aba4ff..f0317a1576 100644
--- a/src/google/adk/evaluation/rubric_based_final_response_quality_v1.py
+++ b/src/google/adk/evaluation/rubric_based_final_response_quality_v1.py
@@ -289,14 +289,25 @@ def format_auto_rater_prompt(
 
     app_details = actual_invocation.app_details
     if app_details:
+      # Determine agent name from invocation events if available,
+      # otherwise fall back to the first agent in app_details.
+      # This ensures developer_instructions are populated even when
+      # the agent makes zero tool calls (e.g., declining out-of-scope
+      # requests).
+      agent_name = None
       if (
           isinstance(actual_invocation.intermediate_data, InvocationEvents)
           and actual_invocation.intermediate_data.invocation_events
       ):
+        agent_name = (
+            actual_invocation.intermediate_data.invocation_events[0].author
+        )
+      elif app_details.agent_details:
+        agent_name = next(iter(app_details.agent_details))
+
+      if agent_name:
         developer_instructions = app_details.get_developer_instructions(
-            agent_name=actual_invocation.intermediate_data.invocation_events[
-                0
-            ].author
+            agent_name=agent_name
         )
       tool_declarations = get_tool_declarations_as_json_str(app_details)
 
diff --git a/tests/unittests/evaluation/test_rubric_based_final_response_quality_v1.py b/tests/unittests/evaluation/test_rubric_based_final_response_quality_v1.py
index e100f9c06a..708f749f79 100644
--- a/tests/unittests/evaluation/test_rubric_based_final_response_quality_v1.py
+++ b/tests/unittests/evaluation/test_rubric_based_final_response_quality_v1.py
@@ -222,3 +222,66 @@ def test_format_auto_rater_prompt_with_intermediate_data_no_tools(
   prompt = evaluator.format_auto_rater_prompt(invocation, None)
 
   assert "No intermediate steps were taken." in prompt
+
+
+def test_format_auto_rater_prompt_with_app_details_empty_invocation_events(
+    evaluator: RubricBasedFinalResponseQualityV1Evaluator,
+):
+  """Tests that developer_instructions are populated even when invocation_events is empty.
+
+  This covers the case where an agent declines a request without calling any
+  tools (e.g., out-of-scope rejection), resulting in zero invocation events.
+  The judge should still receive the developer instructions to evaluate rubrics
+  that reference the system prompt.
+  """
+  app_details = AppDetails(
+      agent_details={
+          "my_agent": AgentDetails(
+              name="my_agent",
+              instructions="Only answer questions about cooking. Decline all other requests.",
+              tool_declarations=[],
+          )
+      },
+  )
+  invocation = Invocation(
+      user_content=genai_types.Content(
+          parts=[genai_types.Part(text="What is the capital of France?")]
+      ),
+      final_response=genai_types.Content(
+          parts=[genai_types.Part(text="I can only help with cooking.")]
+      ),
+      app_details=app_details,
+      intermediate_data=InvocationEvents(invocation_events=[]),
+  )
+  prompt = evaluator.format_auto_rater_prompt(invocation, None)
+
+  assert "Only answer questions about cooking." in prompt
+  assert "I can only help with cooking." in prompt
+
+
+def test_format_auto_rater_prompt_with_app_details_no_intermediate_data(
+    evaluator: RubricBasedFinalResponseQualityV1Evaluator,
+):
+  """Tests that developer_instructions are populated when intermediate_data is None."""
+  app_details = AppDetails(
+      agent_details={
+          "my_agent": AgentDetails(
+              name="my_agent",
+              instructions="Agent instructions here.",
+              tool_declarations=[],
+          )
+      },
+  )
+  invocation = Invocation(
+      user_content=genai_types.Content(
+          parts=[genai_types.Part(text="User input here.")]
+      ),
+      final_response=genai_types.Content(
+          parts=[genai_types.Part(text="Final agent response.")]
+      ),
+      app_details=app_details,
+      intermediate_data=None,
+  )
+  prompt = evaluator.format_auto_rater_prompt(invocation, None)
+
+  assert "Agent instructions here." in prompt