From 90dd87edeb45a46c93cf04fd90240836c6bf782e Mon Sep 17 00:00:00 2001
From: Claude Code <claude@vera-ai>
Date: Wed, 1 Apr 2026 16:10:24 -0500
Subject: [PATCH] fix: system prompt appends to caller's system message, empty
 = passthrough

Handle all 4 combinations of caller system message and systemprompt.md
correctly: append when both exist, passthrough when only one exists,
omit when neither exists. Fixes leading \n\n when no caller system msg.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 app/utils.py        | 19 ++++++++----
 tests/test_utils.py | 73 +++++++++++++++++++++++++++++++++++++++++++--
 2 files changed, 84 insertions(+), 8 deletions(-)

diff --git a/app/utils.py b/app/utils.py
index 2dce7b9..793b356 100644
--- a/app/utils.py
+++ b/app/utils.py
@@ -219,15 +219,22 @@ async def build_augmented_messages(incoming_messages: List[Dict]) -> List[Dict]:
     }
     
     # === LAYER 1: System Prompt ===
-    system_content = ""
+    # Caller's system message passes through; systemprompt.md appends if non-empty.
+    caller_system = ""
     for msg in incoming_messages:
         if msg.get("role") == "system":
-            system_content = msg.get("content", "")
+            caller_system = msg.get("content", "")
             break
-    
-    if system_prompt:
-        system_content += "\n\n" + system_prompt
-    
+
+    if caller_system and system_prompt:
+        system_content = caller_system + "\n\n" + system_prompt
+    elif caller_system:
+        system_content = caller_system
+    elif system_prompt:
+        system_content = system_prompt
+    else:
+        system_content = ""
+
     if system_content:
         messages.append({"role": "system", "content": system_content})
         logger.info(f"Layer 1 (system): {count_tokens(system_content)} tokens")
diff --git a/tests/test_utils.py b/tests/test_utils.py
index 0ac436e..89a9a36 100644
--- a/tests/test_utils.py
+++ b/tests/test_utils.py
@@ -1,6 +1,7 @@
 """Tests for utility functions."""
 import pytest
-from app.utils import count_tokens, truncate_by_tokens, parse_curated_turn
+from unittest.mock import AsyncMock, MagicMock, patch
+from app.utils import count_tokens, truncate_by_tokens, parse_curated_turn, build_augmented_messages
 
 
 class TestCountTokens:
@@ -326,4 +327,72 @@ class TestBuildAugmentedMessages:
             )
 
         contents = [m["content"] for m in result]
-        assert any("Old question" in c or "Old answer" in c for c in contents)
\ No newline at end of file
+        assert any("Old question" in c or "Old answer" in c for c in contents)
+
+    @pytest.mark.asyncio
+    async def test_system_prompt_appends_to_caller_system(self):
+        """systemprompt.md content appends to caller's system message."""
+        import app.utils as utils_module
+
+        mock_qdrant = self._make_qdrant_mock()
+
+        with patch.object(utils_module, "load_system_prompt", return_value="Vera memory context"), \
+             patch.object(utils_module, "get_qdrant_service", return_value=mock_qdrant):
+            incoming = [
+                {"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": "Hello"}
+            ]
+            result = await build_augmented_messages(incoming)
+
+            system_msg = result[0]
+            assert system_msg["role"] == "system"
+            assert system_msg["content"] == "You are a helpful assistant.\n\nVera memory context"
+
+    @pytest.mark.asyncio
+    async def test_empty_system_prompt_passthrough(self):
+        """When systemprompt.md is empty, only caller's system message passes through."""
+        import app.utils as utils_module
+
+        mock_qdrant = self._make_qdrant_mock()
+
+        with patch.object(utils_module, "load_system_prompt", return_value=""), \
+             patch.object(utils_module, "get_qdrant_service", return_value=mock_qdrant):
+            incoming = [
+                {"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": "Hello"}
+            ]
+            result = await build_augmented_messages(incoming)
+
+            system_msg = result[0]
+            assert system_msg["content"] == "You are a helpful assistant."
+
+    @pytest.mark.asyncio
+    async def test_no_caller_system_with_vera_prompt(self):
+        """When caller sends no system message but systemprompt.md exists, use vera prompt."""
+        import app.utils as utils_module
+
+        mock_qdrant = self._make_qdrant_mock()
+
+        with patch.object(utils_module, "load_system_prompt", return_value="Vera memory context"), \
+             patch.object(utils_module, "get_qdrant_service", return_value=mock_qdrant):
+            incoming = [{"role": "user", "content": "Hello"}]
+            result = await build_augmented_messages(incoming)
+
+            system_msg = result[0]
+            assert system_msg["role"] == "system"
+            assert system_msg["content"] == "Vera memory context"
+
+    @pytest.mark.asyncio
+    async def test_no_system_anywhere(self):
+        """When neither caller nor systemprompt.md provides system content, no system message."""
+        import app.utils as utils_module
+
+        mock_qdrant = self._make_qdrant_mock()
+
+        with patch.object(utils_module, "load_system_prompt", return_value=""), \
+             patch.object(utils_module, "get_qdrant_service", return_value=mock_qdrant):
+            incoming = [{"role": "user", "content": "Hello"}]
+            result = await build_augmented_messages(incoming)
+
+            # First message should be user, not system
+            assert result[0]["role"] == "user"
\ No newline at end of file