semantic-kernel 1.41.0__tar.gz → 1.41.2__tar.gz

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: semantic-kernel
-Version: 1.41.0
+Version: 1.41.2
 Summary: Semantic Kernel Python SDK
 Author-email: Microsoft <SK-Support@microsoft.com>
 Requires-Python: >=3.10
@@ -19,7 +19,7 @@ Requires-Dist: azure-ai-projects ~= 1.0.0b12
 Requires-Dist: azure-ai-agents >= 1.2.0b3
 Requires-Dist: aiohttp ~= 3.8
 Requires-Dist: cloudevents ~=1.0
-Requires-Dist: pydantic >=2.0,!=2.10.0,!=2.10.1,!=2.10.2,!=2.10.3,<2.12
+Requires-Dist: pydantic >=2.0,!=2.10.0,!=2.10.1,!=2.10.2,!=2.10.3,<2.13
 Requires-Dist: pydantic-settings ~= 2.0
 Requires-Dist: defusedxml ~= 0.7
 Requires-Dist: azure-identity >= 1.13
@@ -47,7 +47,7 @@ Requires-Dist: azure-ai-inference >= 1.0.0b6 ; extra == "azure"
 Requires-Dist: azure-core-tracing-opentelemetry >= 1.0.0b11 ; extra == "azure"
 Requires-Dist: azure-search-documents >= 11.6.0b4 ; extra == "azure"
 Requires-Dist: azure-cosmos ~= 4.7 ; extra == "azure"
-Requires-Dist: chromadb >= 0.5,< 1.1 ; extra == "chroma"
+Requires-Dist: chromadb >= 0.5,< 1.4 ; extra == "chroma"
 Requires-Dist: microsoft-agents-copilotstudio-client >= 0.3.1 ; extra == "copilotstudio"
 Requires-Dist: microsoft-agents-activity >= 0.3.1 ; extra == "copilotstudio"
 Requires-Dist: faiss-cpu>=1.10.0 ; extra == "faiss"
@@ -60,9 +60,9 @@ Requires-Dist: mcp>=1.8 ; extra == "mcp"
 Requires-Dist: pymilvus >= 2.3,< 2.7 ; extra == "milvus"
 Requires-Dist: milvus >= 2.3,<2.3.8 ; extra == "milvus" and ( platform_system != 'Windows')
 Requires-Dist: mistralai >= 1.2,< 2.0 ; extra == "mistralai"
-Requires-Dist: pymongo >= 4.8.0, < 4.15 ; extra == "mongo"
+Requires-Dist: pymongo >= 4.8.0, < 4.16 ; extra == "mongo"
 Requires-Dist: motor >= 3.3.2,< 3.8.0 ; extra == "mongo"
-Requires-Dist: ipykernel ~= 6.29 ; extra == "notebooks"
+Requires-Dist: ipykernel >= 6.29,< 8.0 ; extra == "notebooks"
 Requires-Dist: ollama ~= 0.4 ; extra == "ollama"
 Requires-Dist: onnxruntime==1.22.1 ; extra == "onnx"
 Requires-Dist: onnxruntime-genai==0.9.0 ; extra == "onnx"
@@ -73,7 +73,7 @@ Requires-Dist: psycopg[binary, pool] ~= 3.2 ; extra == "postgres"
 Requires-Dist: qdrant-client ~= 1.9 ; extra == "qdrant"
 Requires-Dist: websockets >= 13, < 16 ; extra == "realtime"
 Requires-Dist: aiortc>=1.9.0 ; extra == "realtime"
-Requires-Dist: redis[hiredis] ~= 6.0 ; extra == "redis"
+Requires-Dist: redis[hiredis] >= 6,< 8 ; extra == "redis"
 Requires-Dist: types-redis ~= 4.6.0.20240425 ; extra == "redis"
 Requires-Dist: redisvl ~= 0.4 ; extra == "redis"
 Requires-Dist: pyodbc >= 5.2 ; extra == "sql"

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/pyproject.toml

@@ -28,7 +28,7 @@ dependencies = [
     "azure-ai-agents >= 1.2.0b3",
     "aiohttp ~= 3.8",
     "cloudevents ~=1.0",
-    "pydantic >=2.0,!=2.10.0,!=2.10.1,!=2.10.2,!=2.10.3,<2.12",
+    "pydantic >=2.0,!=2.10.0,!=2.10.1,!=2.10.2,!=2.10.3,<2.13",
     "pydantic-settings ~= 2.0",
     "defusedxml ~= 0.7",
     # azure identity
@@ -77,7 +77,7 @@ azure = [
     "azure-cosmos ~= 4.7"
 ]
 chroma = [
-    "chromadb >= 0.5,< 1.1"
+    "chromadb >= 0.5,< 1.4"
 ]
 copilotstudio = [
     "microsoft-agents-copilotstudio-client >= 0.3.1",
@@ -106,11 +106,11 @@ mistralai = [
     "mistralai >= 1.2,< 2.0"
 ]
 mongo = [
-    "pymongo >= 4.8.0, < 4.15",
+    "pymongo >= 4.8.0, < 4.16",
     "motor >= 3.3.2,< 3.8.0"
 ]
 notebooks = [
-    "ipykernel ~= 6.29"
+    "ipykernel >= 6.29,< 8.0"
 ]
 ollama = [
     "ollama ~= 0.4"
@@ -136,7 +136,7 @@ qdrant = [
     "qdrant-client ~= 1.9"
 ]
 redis = [
-    "redis[hiredis] ~= 6.0",
+    "redis[hiredis] >= 6,< 8",
     "types-redis ~= 4.6.0.20240425",
     "redisvl ~= 0.4"
 ]
@@ -158,7 +158,7 @@ weaviate = [
 [dependency-groups]
 dev = [
     "pre-commit ~= 3.7",
-    "ipykernel ~= 6.29",
+    "ipykernel >= 6.29,< 8.0",
     "nbconvert ~= 7.16",
     "pytest ~= 8.2",
     "pytest-xdist[psutil] ~= 3.6",

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/__init__.py

@@ -2,7 +2,7 @@
 
 from semantic_kernel.kernel import Kernel
 
-__version__ = "1.41.0"
+__version__ = "1.41.2"
 
 DEFAULT_RC_VERSION = f"{__version__}-rc9"
 

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/agents/open_ai/responses_agent_thread_actions.py

@@ -1214,10 +1214,19 @@ class ResponsesAgentThreadActions:
         if agent.tools:
             tools.extend(agent.tools)
 
-        # TODO(evmattso): make sure to respect filters on FCB
-        if kernel.plugins:
-            funcs = kernel.get_full_list_of_function_metadata()
-            tools.extend([kernel_function_metadata_to_response_function_call_format(f) for f in funcs])
+        if not function_choice_behavior.enable_kernel_functions:
+            return tools
+
+        if not kernel.plugins:
+            return tools
+
+        funcs = (
+            kernel.get_list_of_function_metadata(function_choice_behavior.filters)
+            if function_choice_behavior.filters
+            else kernel.get_full_list_of_function_metadata()
+        )
+
+        tools.extend([kernel_function_metadata_to_response_function_call_format(f) for f in funcs])
 
         return tools
 

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/connectors/ai/google/google_ai/services/google_ai_chat_completion.py

@@ -119,7 +119,9 @@ class GoogleAIChatCompletion(GoogleAIBase, ChatCompletionClientBase):
         if not client:
             if google_ai_settings.use_vertexai and not google_ai_settings.cloud_project_id:
                 raise ServiceInitializationError("Project ID must be provided when use_vertexai is True.")
-            if not google_ai_settings.api_key:
+            if google_ai_settings.use_vertexai and not google_ai_settings.cloud_region:
+                raise ServiceInitializationError("Region must be provided when use_vertexai is True.")
+            if not google_ai_settings.use_vertexai and not google_ai_settings.api_key:
                 raise ServiceInitializationError("The API key is required when use_vertexai is False.")
 
         super().__init__(
@@ -303,6 +305,10 @@ class GoogleAIChatCompletion(GoogleAIBase, ChatCompletionClientBase):
                 if part.text:
                     items.append(TextContent(text=part.text, inner_content=response, metadata=response_metadata))
                 elif part.function_call:
+                    fc_metadata: dict[str, Any] = {}
+                    thought_sig = getattr(part, "thought_signature", None)
+                    if thought_sig:
+                        fc_metadata["thought_signature"] = thought_sig
                     items.append(
                         FunctionCallContent(
                             id=f"{part.function_call.name}_{idx!s}",
@@ -310,6 +316,7 @@ class GoogleAIChatCompletion(GoogleAIBase, ChatCompletionClientBase):
                                 part.function_call.name  # type: ignore[arg-type]
                             ),
                             arguments={k: v for k, v in part.function_call.args.items()},  # type: ignore
+                            metadata=fc_metadata if fc_metadata else None,
                         )
                     )
 
@@ -360,6 +367,10 @@ class GoogleAIChatCompletion(GoogleAIBase, ChatCompletionClientBase):
                         )
                     )
                 elif part.function_call:
+                    fc_metadata: dict[str, Any] = {}
+                    thought_sig = getattr(part, "thought_signature", None)
+                    if thought_sig:
+                        fc_metadata["thought_signature"] = thought_sig
                     items.append(
                         FunctionCallContent(
                             id=f"{part.function_call.name}_{idx!s}",
@@ -367,6 +378,7 @@ class GoogleAIChatCompletion(GoogleAIBase, ChatCompletionClientBase):
                                 part.function_call.name  # type: ignore[arg-type]
                             ),
                             arguments={k: v for k, v in part.function_call.args.items()},  # type: ignore
+                            metadata=fc_metadata if fc_metadata else None,
                         )
                     )
 

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/connectors/ai/google/google_ai/services/google_ai_text_completion.py

@@ -90,7 +90,9 @@ class GoogleAITextCompletion(GoogleAIBase, TextCompletionClientBase):
         if not client:
             if google_ai_settings.use_vertexai and not google_ai_settings.cloud_project_id:
                 raise ServiceInitializationError("Project ID must be provided when use_vertexai is True.")
-            if not google_ai_settings.api_key:
+            if google_ai_settings.use_vertexai and not google_ai_settings.cloud_region:
+                raise ServiceInitializationError("Region must be provided when use_vertexai is True.")
+            if not google_ai_settings.use_vertexai and not google_ai_settings.api_key:
                 raise ServiceInitializationError("The API key is required when use_vertexai is False.")
 
         super().__init__(

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/connectors/ai/google/google_ai/services/google_ai_text_embedding.py

@@ -81,7 +81,9 @@ class GoogleAITextEmbedding(GoogleAIBase, EmbeddingGeneratorBase):
         if not client:
             if google_ai_settings.use_vertexai and not google_ai_settings.cloud_project_id:
                 raise ServiceInitializationError("Project ID must be provided when use_vertexai is True.")
-            if not google_ai_settings.api_key:
+            if google_ai_settings.use_vertexai and not google_ai_settings.cloud_region:
+                raise ServiceInitializationError("Region must be provided when use_vertexai is True.")
+            if not google_ai_settings.use_vertexai and not google_ai_settings.api_key:
                 raise ServiceInitializationError("The API key is required when use_vertexai is False.")
 
         super().__init__(

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/connectors/ai/google/google_ai/services/utils.py

@@ -13,6 +13,7 @@ from semantic_kernel.connectors.ai.google.google_ai.google_ai_prompt_execution_s
 from semantic_kernel.connectors.ai.google.shared_utils import (
     FUNCTION_CHOICE_TYPE_TO_GOOGLE_FUNCTION_CALLING_MODE,
     GEMINI_FUNCTION_NAME_SEPARATOR,
+    sanitize_schema_for_google_ai,
 )
 from semantic_kernel.contents.chat_message_content import ChatMessageContent
 from semantic_kernel.contents.function_call_content import FunctionCallContent
@@ -91,12 +92,24 @@ def format_assistant_message(message: ChatMessageContent) -> list[Part]:
             if item.text:
                 parts.append(Part.from_text(text=item.text))
         elif isinstance(item, FunctionCallContent):
-            parts.append(
-                Part.from_function_call(
-                    name=item.name,  # type: ignore[arg-type]
-                    args=json.loads(item.arguments) if isinstance(item.arguments, str) else item.arguments,  # type: ignore[arg-type]
+            thought_signature = item.metadata.get("thought_signature") if item.metadata else None
+            if thought_signature:
+                parts.append(
+                    Part(
+                        function_call={
+                            "name": item.name,  # type: ignore[arg-type]
+                            "args": json.loads(item.arguments) if isinstance(item.arguments, str) else item.arguments,
+                        },
+                        thought_signature=thought_signature,
+                    )
+                )
+            else:
+                parts.append(
+                    Part.from_function_call(
+                        name=item.name,  # type: ignore[arg-type]
+                        args=json.loads(item.arguments) if isinstance(item.arguments, str) else item.arguments,  # type: ignore[arg-type]
+                    )
                 )
-            )
         elif isinstance(item, ImageContent):
             parts.append(_create_image_part(item))
         else:
@@ -135,16 +148,23 @@ def format_tool_message(message: ChatMessageContent) -> list[Part]:
 
 def kernel_function_metadata_to_google_ai_function_call_format(metadata: KernelFunctionMetadata) -> dict[str, Any]:
     """Convert the kernel function metadata to function calling format."""
+    parameters: dict[str, Any] | None = None
+    if metadata.parameters:
+        properties = {}
+        for param in metadata.parameters:
+            if param.name is None:
+                continue
+            prop_schema = sanitize_schema_for_google_ai(param.schema_data) if param.schema_data else param.schema_data
+            properties[param.name] = prop_schema
+        parameters = {
+            "type": "object",
+            "properties": properties,
+            "required": [p.name for p in metadata.parameters if p.is_required and p.name is not None],
+        }
     return {
         "name": metadata.custom_fully_qualified_name(GEMINI_FUNCTION_NAME_SEPARATOR),
         "description": metadata.description or "",
-        "parameters": {
-            "type": "object",
-            "properties": {param.name: param.schema_data for param in metadata.parameters},
-            "required": [p.name for p in metadata.parameters if p.is_required],
-        }
-        if metadata.parameters
-        else None,
+        "parameters": parameters,
     }
 
 

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/connectors/ai/google/shared_utils.py

@@ -1,6 +1,8 @@
 # Copyright (c) Microsoft. All rights reserved.
 
 import logging
+from copy import deepcopy
+from typing import Any
 
 from semantic_kernel.connectors.ai.function_choice_behavior import FunctionChoiceType
 from semantic_kernel.const import DEFAULT_FULLY_QUALIFIED_NAME_SEPARATOR
@@ -51,6 +53,64 @@ def format_gemini_function_name_to_kernel_function_fully_qualified_name(gemini_f
     return gemini_function_name
 
 
+def sanitize_schema_for_google_ai(schema: dict[str, Any] | None) -> dict[str, Any] | None:
+    """Sanitize a JSON schema dict so it is compatible with Google AI / Vertex AI.
+
+    The Google AI protobuf ``Schema`` does not support ``anyOf``, ``oneOf``, or
+    ``allOf``.  It also does not accept ``type`` as an array (e.g.
+    ``["string", "null"]``).  This helper recursively rewrites those constructs
+    into the subset that Google AI understands, using ``nullable`` where
+    appropriate.
+    """
+    if schema is None:
+        return None
+
+    schema = deepcopy(schema)
+    return _sanitize_node(schema)
+
+
+def _sanitize_node(node: dict[str, Any]) -> dict[str, Any]:
+    """Recursively sanitize a single schema node."""
+    # --- handle ``type`` given as a list (e.g. ["string", "null"]) ---
+    type_val = node.get("type")
+    if isinstance(type_val, list):
+        non_null = [t for t in type_val if t != "null"]
+        if len(type_val) != len(non_null):
+            node["nullable"] = True
+        node["type"] = non_null[0] if non_null else "string"
+
+    # --- handle ``anyOf`` / ``oneOf`` / ``allOf`` ---
+    for key in ("anyOf", "oneOf", "allOf"):
+        variants = node.get(key)
+        if not variants:
+            continue
+        non_null = [v for v in variants if v.get("type") != "null"]
+        has_null = len(variants) != len(non_null)
+        chosen = _sanitize_node(non_null[0]) if non_null else {"type": "string"}
+        # Preserve description from the outer node
+        desc = node.get("description")
+        node.clear()
+        node.update(chosen)
+        if has_null:
+            node["nullable"] = True
+        if desc and "description" not in node:
+            node["description"] = desc
+        break  # only process the first matching key
+
+    # --- recurse into nested structures ---
+    props = node.get("properties")
+    if isinstance(props, dict):
+        for prop_name, prop_schema in props.items():
+            if isinstance(prop_schema, dict):
+                props[prop_name] = _sanitize_node(prop_schema)
+
+    items = node.get("items")
+    if isinstance(items, dict):
+        node["items"] = _sanitize_node(items)
+
+    return node
+
+
 def collapse_function_call_results_in_chat_history(chat_history: ChatHistory):
     """The Gemini API expects the results of parallel function calls to be contained in a single message to be returned.
 

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/connectors/ai/google/vertex_ai/services/utils.py

@@ -2,7 +2,7 @@
 
 import json
 import logging
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Any
 
 from google.cloud.aiplatform_v1beta1.types.content import Candidate
 from vertexai.generative_models import FunctionDeclaration, Part, Tool, ToolConfig
@@ -11,6 +11,7 @@ from semantic_kernel.connectors.ai.function_choice_behavior import FunctionChoic
 from semantic_kernel.connectors.ai.google.shared_utils import (
     FUNCTION_CHOICE_TYPE_TO_GOOGLE_FUNCTION_CALLING_MODE,
     GEMINI_FUNCTION_NAME_SEPARATOR,
+    sanitize_schema_for_google_ai,
 )
 from semantic_kernel.connectors.ai.google.vertex_ai.vertex_ai_prompt_execution_settings import (
     VertexAIChatPromptExecutionSettings,
@@ -89,14 +90,16 @@ def format_assistant_message(message: ChatMessageContent) -> list[Part]:
             if item.text:
                 parts.append(Part.from_text(item.text))
         elif isinstance(item, FunctionCallContent):
-            parts.append(
-                Part.from_dict({
-                    "function_call": {
-                        "name": item.name,
-                        "args": json.loads(item.arguments) if isinstance(item.arguments, str) else item.arguments,
-                    }
-                })
-            )
+            part_dict: dict[str, Any] = {
+                "function_call": {
+                    "name": item.name,  # type: ignore[arg-type]
+                    "args": json.loads(item.arguments) if isinstance(item.arguments, str) else item.arguments,
+                }
+            }
+            thought_signature = item.metadata.get("thought_signature") if item.metadata else None
+            if thought_signature:
+                part_dict["thought_signature"] = thought_signature
+            parts.append(Part.from_dict(part_dict))
         elif isinstance(item, ImageContent):
             parts.append(_create_image_part(item))
         else:
@@ -135,13 +138,20 @@ def format_tool_message(message: ChatMessageContent) -> list[Part]:
 
 def kernel_function_metadata_to_vertex_ai_function_call_format(metadata: KernelFunctionMetadata) -> FunctionDeclaration:
     """Convert the kernel function metadata to function calling format."""
+    properties: dict[str, Any] = {}
+    if metadata.parameters:
+        for param in metadata.parameters:
+            if param.name is None:
+                continue
+            prop_schema = sanitize_schema_for_google_ai(param.schema_data) if param.schema_data else param.schema_data
+            properties[param.name] = prop_schema
     return FunctionDeclaration(
         name=metadata.custom_fully_qualified_name(GEMINI_FUNCTION_NAME_SEPARATOR),
         description=metadata.description or "",
         parameters={
             "type": "object",
-            "properties": {param.name: param.schema_data for param in metadata.parameters},
-            "required": [p.name for p in metadata.parameters if p.is_required],
+            "properties": properties,
+            "required": [p.name for p in metadata.parameters if p.is_required and p.name is not None],
         },
     )
 

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/connectors/ai/google/vertex_ai/services/vertex_ai_chat_completion.py

@@ -252,6 +252,10 @@ class VertexAIChatCompletion(VertexAIBase, ChatCompletionClientBase):
             if "text" in part_dict:
                 items.append(TextContent(text=part.text, inner_content=response, metadata=response_metadata))
             elif "function_call" in part_dict:
+                fc_metadata: dict[str, Any] = {}
+                thought_sig = part_dict.get("thought_signature")
+                if thought_sig:
+                    fc_metadata["thought_signature"] = thought_sig
                 items.append(
                     FunctionCallContent(
                         id=f"{part.function_call.name}_{idx!s}",
@@ -259,6 +263,7 @@ class VertexAIChatCompletion(VertexAIBase, ChatCompletionClientBase):
                             part.function_call.name
                         ),
                         arguments={k: v for k, v in part.function_call.args.items()},
+                        metadata=fc_metadata if fc_metadata else None,
                     )
                 )
 
@@ -309,6 +314,10 @@ class VertexAIChatCompletion(VertexAIBase, ChatCompletionClientBase):
                     )
                 )
             elif "function_call" in part_dict:
+                fc_metadata_s: dict[str, Any] = {}
+                thought_sig_s = part_dict.get("thought_signature")
+                if thought_sig_s:
+                    fc_metadata_s["thought_signature"] = thought_sig_s
                 items.append(
                     FunctionCallContent(
                         id=f"{part.function_call.name}_{idx!s}",
@@ -316,6 +325,7 @@ class VertexAIChatCompletion(VertexAIBase, ChatCompletionClientBase):
                             part.function_call.name
                         ),
                         arguments={k: v for k, v in part.function_call.args.items()},
+                        metadata=fc_metadata_s if fc_metadata_s else None,
                     )
                 )
 

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/contents/chat_history.py

@@ -353,6 +353,23 @@ class ChatHistory(KernelBaseModel):
             elif item.tag == CHAT_HISTORY_TAG:
                 for message in item:
                     messages.append(ChatMessageContent.from_element(message))
+            else:
+                # Unknown XML tags (e.g. HTML tags like <p>, <div>) are not SK
+                # template tags.  Serialize them back to text and append to the
+                # previous message so that the original content is preserved.
+                saved_tail = item.tail
+                item.tail = None
+                raw = unescape(tostring(item, encoding="unicode", short_empty_elements=False))
+                item.tail = saved_tail
+                if messages:
+                    messages[-1].content = (messages[-1].content or "") + raw
+                else:
+                    messages.append(ChatMessageContent(role=AuthorRole.USER, content=raw))
+                # For unknown tags the tail is part of the surrounding text,
+                # so keep it in the same message instead of starting a new one.
+                if item.tail:
+                    messages[-1].content = (messages[-1].content or "") + unescape(item.tail)
+                continue
             if item.tail and item.tail.strip():
                 messages.append(ChatMessageContent(role=AuthorRole.USER, content=unescape(item.tail.strip())))
         if len(messages) == 1 and messages[0].role == AuthorRole.SYSTEM:

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/contents/history_reducer/chat_history_reducer_utils.py

@@ -66,6 +66,7 @@ def locate_safe_reduction_index(
     target_count: int,
     threshold_count: int = 0,
     offset_count: int = 0,
+    has_system_message: bool = False,
 ) -> int | None:
     """Identify the index of the first message at or beyond the specified target_count.
 
@@ -83,11 +84,27 @@ def locate_safe_reduction_index(
         threshold_count: The threshold beyond target_count required to trigger reduction.
                          If total messages <= (target_count + threshold_count), no reduction occurs.
         offset_count: Optional number of messages to skip at the start (e.g. existing summary messages).
+        has_system_message: Whether the history contains a system message that will be preserved
+                           separately. When True, the target_count is adjusted to account for the
+                           system message being re-added after reduction.
 
     Returns:
         The index that identifies the starting point for a reduced history that does not orphan
         sensitive content. Returns None if reduction is not needed.
     """
+    # Adjust target_count to account for the system message that will be preserved separately.
+    # This matches the .NET SDK behavior.
+    if has_system_message:
+        target_count -= 1
+        if target_count <= 0:
+            logger.warning(
+                "target_count after accounting for system message is %d; reduction will keep only the system message.",
+                target_count,
+            )
+            # Reduce to just the system message — return index past all non-system messages.
+            # The caller will prepend the system message to the empty/minimal tail.
+            return len(history)
+
     total_count = len(history)
     threshold_index = total_count - (threshold_count or 0) - target_count
     if threshold_index <= offset_count:
@@ -220,5 +237,12 @@ def extract_range(
 
 @experimental
 def contains_function_call_or_result(msg: ChatMessageContent) -> bool:
-    """Return True if the message has any function call or function result."""
+    """Return True if the message has any function call or function result.
+
+    Also returns True for TOOL role messages, which are always responses to
+    a preceding assistant message with tool_calls and must not be separated
+    from it.
+    """
+    if msg.role == AuthorRole.TOOL:
+        return True
     return any(isinstance(item, (FunctionCallContent, FunctionResultContent)) for item in msg.items)

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/contents/history_reducer/chat_history_summarization_reducer.py

@@ -26,6 +26,7 @@ from semantic_kernel.contents.history_reducer.chat_history_reducer_utils import
     locate_safe_reduction_index,
     locate_summarization_boundary,
 )
+from semantic_kernel.contents.utils.author_role import AuthorRole
 from semantic_kernel.exceptions.content_exceptions import ChatHistoryReducerException
 from semantic_kernel.utils.feature_stage_decorator import experimental
 
@@ -89,6 +90,21 @@ class ChatHistorySummarizationReducer(ChatHistoryReducer):
 
         logger.info("Performing chat history summarization check...")
 
+        # Preserve system/developer messages so they are not lost during summarization.
+        # This matches the .NET SDK behavior and the truncation reducer.
+        # Only the first system/developer message is preserved; this mirrors .NET semantics.
+        # Exclude summary messages (which may have SYSTEM role) — they are generated content,
+        # not original system prompts.
+        system_message_index = next(
+            (
+                i
+                for i, msg in enumerate(history)
+                if msg.role in (AuthorRole.SYSTEM, AuthorRole.DEVELOPER) and not msg.metadata.get(SUMMARY_METADATA_KEY)
+            ),
+            -1,
+        )
+        system_message = history[system_message_index] if system_message_index >= 0 else None
+
         # 1. Identify where existing summary messages end
         insertion_point = locate_summarization_boundary(history)
         if insertion_point == len(history):
@@ -96,12 +112,19 @@ class ChatHistorySummarizationReducer(ChatHistoryReducer):
             logger.warning("All messages are summaries, forcing boundary to 0.")
             insertion_point = 0
 
+        # Only adjust target_count if the system message would be truncated away.
+        # If the system message is already in the retained portion, no adjustment needed.
+        system_would_be_truncated = (
+            system_message is not None and system_message_index < len(history) - self.target_count
+        )
+
         # 2. Locate the safe reduction index
         truncation_index = locate_safe_reduction_index(
             history,
             self.target_count,
             self.threshold_count,
             offset_count=insertion_point,
+            has_system_message=system_would_be_truncated,
         )
         if truncation_index is None:
             logger.info("No valid truncation index found.")
@@ -138,7 +161,13 @@ class ChatHistorySummarizationReducer(ChatHistoryReducer):
                 keep_existing_summaries = history[:insertion_point]
 
             remainder = history[truncation_index:]
+
+            # Prepend the system/developer message if it was summarized away.
+            # Use identity comparison to avoid false matches from value-equal messages.
             new_history = [*keep_existing_summaries, summary_msg, *remainder]
+            if system_message is not None and not any(m is system_message for m in new_history):
+                new_history = [system_message, *new_history]
+
             self.messages = new_history
 
             return self
@@ -151,8 +180,6 @@ class ChatHistorySummarizationReducer(ChatHistoryReducer):
 
     async def _summarize(self, messages: list[ChatMessageContent]) -> ChatMessageContent | None:
         """Use the ChatCompletion service to generate a single summary message."""
-        from semantic_kernel.contents.utils.author_role import AuthorRole
-
         chat_history = ChatHistory(messages=messages)
         execution_settings = self.execution_settings or self.service.get_prompt_execution_settings_from_settings(
             PromptExecutionSettings()

{semantic_kernel-1.41.0 → semantic_kernel-1.41.2}/semantic_kernel/contents/history_reducer/chat_history_truncation_reducer.py

@@ -15,9 +15,9 @@ else:
 
 from semantic_kernel.contents.history_reducer.chat_history_reducer import ChatHistoryReducer
 from semantic_kernel.contents.history_reducer.chat_history_reducer_utils import (
-    extract_range,
     locate_safe_reduction_index,
 )
+from semantic_kernel.contents.utils.author_role import AuthorRole
 from semantic_kernel.utils.feature_stage_decorator import experimental
 
 logger = logging.getLogger(__name__)
@@ -45,7 +45,28 @@ class ChatHistoryTruncationReducer(ChatHistoryReducer):
 
         logger.info("Performing chat history truncation check...")
 
-        truncation_index = locate_safe_reduction_index(history, self.target_count, self.threshold_count)
+        # Preserve system/developer messages so they are not lost during truncation.
+        # This matches the .NET SDK behavior where system messages are always retained.
+        # Only the first system/developer message is preserved; this mirrors .NET semantics.
+        system_message_index = next(
+            (i for i, msg in enumerate(history) if msg.role in (AuthorRole.SYSTEM, AuthorRole.DEVELOPER)),
+            -1,
+        )
+        system_message = history[system_message_index] if system_message_index >= 0 else None
+
+        # Only adjust target_count if the system message would be truncated away
+        # (i.e., it falls before the naive tail). If the system message is already in the
+        # retained portion, no adjustment is needed — it naturally occupies a slot.
+        system_would_be_truncated = (
+            system_message is not None and system_message_index < len(history) - self.target_count
+        )
+
+        truncation_index = locate_safe_reduction_index(
+            history,
+            self.target_count,
+            self.threshold_count,
+            has_system_message=system_would_be_truncated,
+        )
         if truncation_index is None:
             logger.info(
                 f"No truncation index found. Target count: {self.target_count}, Threshold: {self.threshold_count}"
@@ -53,7 +74,13 @@ class ChatHistoryTruncationReducer(ChatHistoryReducer):
             return None
 
         logger.info(f"Truncating history to {truncation_index} messages.")
-        truncated_list = extract_range(history, start=truncation_index)
+        truncated_list = history[truncation_index:]
+
+        # Prepend the system/developer message if it was truncated away.
+        # Use identity comparison (is) to avoid false matches from value-equal messages.
+        if system_message is not None and all(msg is not system_message for msg in truncated_list):
+            truncated_list = [system_message, *truncated_list]
+
         self.messages = truncated_list
         return self