solace-agent-mesh 0.2.4__py3-none-any.whl → 1.0.2__py3-none-any.whl

solace_agent_mesh/agent/sac/component.py

@@ -0,0 +1,3496 @@
+"""
+Custom Solace AI Connector Component to Host Google ADK Agents via A2A Protocol.
+"""
+
+from typing import Any, Dict, Optional, Union, Callable, List, Tuple, TYPE_CHECKING
+import asyncio
+import functools
+import threading
+import concurrent.futures
+import fnmatch
+import base64
+from datetime import datetime, timezone
+import json
+from solace_ai_connector.components.component_base import ComponentBase
+from solace_ai_connector.common.message import (
+    Message as SolaceMessage,
+)
+from solace_ai_connector.common.log import log
+from solace_ai_connector.common.event import Event, EventType
+from solace_ai_connector.common.utils import import_module
+import inspect
+from pydantic import BaseModel, ValidationError
+from google.adk.agents.invocation_context import (
+    LlmCallsLimitExceededError,
+)
+from google.adk.agents import RunConfig
+from google.adk.agents.run_config import StreamingMode
+from google.adk.sessions import BaseSessionService
+from google.adk.artifacts import BaseArtifactService
+from google.adk.memory import BaseMemoryService
+from google.adk.agents import LlmAgent
+from google.adk.runners import Runner
+from google.adk.models import LlmResponse
+from google.adk.agents.readonly_context import ReadonlyContext
+from google.adk.events import Event as ADKEvent
+from google.adk.agents.callback_context import CallbackContext
+from google.adk.models.llm_request import LlmRequest
+from google.genai import types as adk_types
+from google.adk.tools.mcp_tool import MCPToolset
+from ...common.types import (
+    AgentCard,
+    Task,
+    TaskStatus,
+    TaskState,
+    Message as A2AMessage,
+    TextPart,
+    FilePart,
+    DataPart,
+    FileContent,
+    Artifact as A2AArtifact,
+    JSONRPCResponse,
+    InternalError,
+    TaskStatusUpdateEvent,
+    TaskArtifactUpdateEvent,
+    SendTaskRequest,
+    CancelTaskRequest,
+    TaskIdParams,
+)
+from ...common.a2a_protocol import (
+    get_a2a_base_topic,
+    get_discovery_topic,
+    get_agent_request_topic,
+    get_agent_response_topic,
+    get_client_response_topic,
+    get_peer_agent_status_topic,
+    format_and_route_adk_event,
+    get_gateway_status_topic,
+)
+from ...agent.utils.config_parser import resolve_instruction_provider
+from ...agent.utils.artifact_helpers import get_latest_artifact_version
+from ...agent.adk.services import (
+    initialize_session_service,
+    initialize_artifact_service,
+    initialize_memory_service,
+)
+from ...agent.adk.setup import (
+    load_adk_tools,
+    initialize_adk_agent,
+    initialize_adk_runner,
+)
+from ...agent.protocol.event_handlers import (
+    process_event,
+    publish_agent_card,
+)
+from ...agent.adk.runner import run_adk_async_task_thread_wrapper, TaskCancelledError
+from ...agent.tools.peer_agent_tool import (
+    CORRELATION_DATA_PREFIX,
+    PeerAgentTool,
+    PEER_TOOL_PREFIX,
+)
+from ...agent.adk.invocation_monitor import InvocationMonitor
+from ...common.middleware.registry import MiddlewareRegistry
+from ...common.constants import DEFAULT_COMMUNICATION_TIMEOUT
+from ...agent.tools.registry import tool_registry
+
+if TYPE_CHECKING:
+    from .task_execution_context import TaskExecutionContext
+
+
+info = {
+    "class_name": "SamAgentComponent",
+    "description": (
+        "Hosts a Google ADK agent and bridges communication via the A2A protocol over Solace. "
+        "NOTE: Configuration is defined in the app-level 'app_config' block "
+        "and validated by 'SamAgentApp.app_schema' when using the associated App class."
+    ),
+    "config_parameters": [],
+    "input_schema": {
+        "type": "object",
+        "description": "Not typically used; component reacts to events.",
+        "properties": {},
+    },
+    "output_schema": {
+        "type": "object",
+        "description": "Not typically used; component publishes results to Solace.",
+        "properties": {},
+    },
+}
+InstructionProvider = Callable[[ReadonlyContext], str]
+
+
+class SamAgentComponent(ComponentBase):
+    """
+    A Solace AI Connector component that hosts a Google ADK agent,
+    communicating via the A2A protocol over Solace.
+    """
+
+    CORRELATION_DATA_PREFIX = CORRELATION_DATA_PREFIX
+    HOST_COMPONENT_VERSION = "1.0.0-alpha"
+
+    def __init__(self, **kwargs):
+        """
+        Initializes the A2A_ADK_HostComponent.
+        Args:
+            **kwargs: Configuration parameters passed from the SAC framework.
+                      Expects configuration under app_config.
+        """
+        if "component_config" in kwargs and "app_config" in kwargs["component_config"]:
+            name = kwargs["component_config"]["app_config"].get("agent_name")
+            if name:
+                kwargs.setdefault("name", name)
+
+        super().__init__(info, **kwargs)
+        self.agent_name = self.get_config("agent_name")
+        log.info("%s Initializing A2A ADK Host Component...", self.log_identifier)
+        try:
+            self.namespace = self.get_config("namespace")
+            if not self.namespace:
+                raise ValueError("Internal Error: Namespace missing after validation.")
+            self.supports_streaming = self.get_config("supports_streaming", False)
+            self.stream_batching_threshold_bytes = self.get_config(
+                "stream_batching_threshold_bytes", 0
+            )
+            self.agent_name = self.get_config("agent_name")
+            if not self.agent_name:
+                raise ValueError("Internal Error: Agent name missing after validation.")
+            self.model_config = self.get_config("model")
+            if not self.model_config:
+                raise ValueError(
+                    "Internal Error: Model config missing after validation."
+                )
+            self.instruction_config = self.get_config("instruction", "")
+            self.global_instruction_config = self.get_config("global_instruction", "")
+            self.tools_config = self.get_config("tools", [])
+            self.planner_config = self.get_config("planner")
+            self.code_executor_config = self.get_config("code_executor")
+            self.session_service_config = self.get_config("session_service")
+            if not self.session_service_config:
+                raise ValueError(
+                    "Internal Error: Session service config missing after validation."
+                )
+            self.default_session_behavior = self.session_service_config.get(
+                "default_behavior", "PERSISTENT"
+            ).upper()
+            if self.default_session_behavior not in ["PERSISTENT", "RUN_BASED"]:
+                log.warning(
+                    "%s Invalid 'default_behavior' in session_service_config: '%s'. Defaulting to PERSISTENT.",
+                    self.log_identifier,
+                    self.default_session_behavior,
+                )
+                self.default_session_behavior = "PERSISTENT"
+            log.info(
+                "%s Default session behavior set to: %s",
+                self.log_identifier,
+                self.default_session_behavior,
+            )
+            self.artifact_service_config = self.get_config(
+                "artifact_service", {"type": "memory"}
+            )
+            self.memory_service_config = self.get_config(
+                "memory_service", {"type": "memory"}
+            )
+            self.artifact_handling_mode = self.get_config(
+                "artifact_handling_mode", "ignore"
+            ).lower()
+            if self.artifact_handling_mode not in ["ignore", "embed", "reference"]:
+                log.warning(
+                    "%s Invalid artifact_handling_mode '%s'. Defaulting to 'ignore'.",
+                    self.log_identifier,
+                    self.artifact_handling_mode,
+                )
+                self.artifact_handling_mode = "ignore"
+            log.info(
+                "%s Artifact Handling Mode: %s",
+                self.log_identifier,
+                self.artifact_handling_mode,
+            )
+            if self.artifact_handling_mode == "reference":
+                log.warning(
+                    "%s Artifact handling mode 'reference' selected, but this component does not currently host an endpoint to serve artifacts. Clients may not be able to retrieve referenced artifacts.",
+                    self.log_identifier,
+                )
+            self.agent_card_config = self.get_config("agent_card")
+            if not self.agent_card_config:
+                raise ValueError(
+                    "Internal Error: Agent card config missing after validation."
+                )
+            self.agent_card_publishing_config = self.get_config("agent_card_publishing")
+            if not self.agent_card_publishing_config:
+                raise ValueError(
+                    "Internal Error: Agent card publishing config missing after validation."
+                )
+            self.agent_discovery_config = self.get_config("agent_discovery")
+            if not self.agent_discovery_config:
+                raise ValueError(
+                    "Internal Error: Agent discovery config missing after validation."
+                )
+            self.inter_agent_communication_config = self.get_config(
+                "inter_agent_communication"
+            )
+            if not self.inter_agent_communication_config:
+                raise ValueError(
+                    "Internal Error: Inter-agent comms config missing after validation."
+                )
+            log.info("%s Configuration retrieved successfully.", self.log_identifier)
+        except Exception as e:
+            log.error(
+                "%s Failed to retrieve configuration via get_config: %s",
+                self.log_identifier,
+                e,
+            )
+            raise ValueError(f"Configuration retrieval error: {e}") from e
+        self.session_service: BaseSessionService = None
+        self.artifact_service: BaseArtifactService = None
+        self.memory_service: BaseMemoryService = None
+        self.adk_agent: LlmAgent = None
+        self.runner: Runner = None
+        self.agent_card_tool_manifest: List[Dict[str, Any]] = []
+        self.peer_agents: Dict[str, Any] = {}
+        self._card_publish_timer_id: str = f"publish_card_{self.agent_name}"
+        self._async_loop = None
+        self._async_thread = None
+        self._async_init_future = None
+        self.peer_response_queues: Dict[str, asyncio.Queue] = {}
+        self.peer_response_queue_lock = threading.Lock()
+        self.agent_specific_state: Dict[str, Any] = {}
+        self.active_tasks: Dict[str, "TaskExecutionContext"] = {}
+        self.active_tasks_lock = threading.Lock()
+        self._agent_system_instruction_string: Optional[str] = None
+        self._agent_system_instruction_callback: Optional[
+            Callable[[CallbackContext, LlmRequest], Optional[str]]
+        ] = None
+        self.invocation_monitor: Optional[InvocationMonitor] = None
+        self._active_background_tasks = set()
+        try:
+            self.agent_specific_state: Dict[str, Any] = {}
+            init_func_details = self.get_config("agent_init_function")
+            if init_func_details and isinstance(init_func_details, dict):
+                module_name = init_func_details.get("module")
+                func_name = init_func_details.get("name")
+                base_path = init_func_details.get("base_path")
+                specific_init_params_dict = init_func_details.get("config", {})
+                if module_name and func_name:
+                    log.info(
+                        "%s Attempting to load init_function: %s.%s",
+                        self.log_identifier,
+                        module_name,
+                        func_name,
+                    )
+                    try:
+                        module = import_module(module_name, base_path=base_path)
+                        init_function = getattr(module, func_name)
+                        if not callable(init_function):
+                            raise TypeError(
+                                f"Init function '{func_name}' in module '{module_name}' is not callable."
+                            )
+                        sig = inspect.signature(init_function)
+                        pydantic_config_model = None
+                        config_param_name = None
+                        validated_config_arg = specific_init_params_dict
+                        for param_name_sig, param_sig in sig.parameters.items():
+                            if (
+                                param_sig.annotation is not inspect.Parameter.empty
+                                and isinstance(param_sig.annotation, type)
+                                and issubclass(param_sig.annotation, BaseModel)
+                            ):
+                                pydantic_config_model = param_sig.annotation
+                                config_param_name = param_name_sig
+                                break
+                        if pydantic_config_model and config_param_name:
+                            log.info(
+                                "%s Found Pydantic config model '%s' for init_function parameter '%s'.",
+                                self.log_identifier,
+                                pydantic_config_model.__name__,
+                                config_param_name,
+                            )
+                            try:
+                                validated_config_arg = pydantic_config_model(
+                                    **specific_init_params_dict
+                                )
+                            except ValidationError as ve:
+                                log.error(
+                                    "%s Validation error for init_function config using Pydantic model '%s': %s",
+                                    self.log_identifier,
+                                    pydantic_config_model.__name__,
+                                    ve,
+                                )
+                                raise ValueError(
+                                    f"Invalid configuration for init_function '{func_name}': {ve}"
+                                ) from ve
+                        elif (
+                            config_param_name
+                            and param_sig.annotation is not inspect.Parameter.empty
+                        ):
+                            log.warning(
+                                "%s Config parameter '%s' for init_function '%s' has a type hint '%s', but it's not a Pydantic BaseModel. Passing raw dict.",
+                                self.log_identifier,
+                                config_param_name,
+                                func_name,
+                                param_sig.annotation,
+                            )
+                        else:
+                            log.info(
+                                "%s No Pydantic model type hint found for a config parameter of init_function '%s'. Passing raw dict if a config param exists, or only host_component.",
+                                self.log_identifier,
+                                func_name,
+                            )
+                        func_params_list = list(sig.parameters.values())
+                        num_actual_params = len(func_params_list)
+                        if num_actual_params == 1:
+                            if specific_init_params_dict:
+                                log.warning(
+                                    "%s Init function '%s' takes 1 argument, but 'config' was provided in YAML. Config will be ignored.",
+                                    self.log_identifier,
+                                    func_name,
+                                )
+                            init_function(self)
+                        elif num_actual_params == 2:
+                            actual_config_param_name_in_signature = func_params_list[
+                                1
+                            ].name
+                            init_function(
+                                self,
+                                **{
+                                    actual_config_param_name_in_signature: validated_config_arg
+                                },
+                            )
+                        else:
+                            raise TypeError(
+                                f"Init function '{func_name}' has an unsupported signature. "
+                                f"Expected (host_component_instance) or (host_component_instance, config_param), "
+                                f"but got {num_actual_params} parameters."
+                            )
+                        log.info(
+                            "%s Successfully executed init_function: %s.%s",
+                            self.log_identifier,
+                            module_name,
+                            func_name,
+                        )
+                    except Exception as e:
+                        log.exception(
+                            "%s Fatal error during agent initialization via init_function '%s.%s': %s",
+                            self.log_identifier,
+                            module_name,
+                            func_name,
+                            e,
+                        )
+                        raise RuntimeError(
+                            f"Agent custom initialization failed: {e}"
+                        ) from e
+            try:
+                self.invocation_monitor = InvocationMonitor()
+            except Exception as im_e:
+                log.error(
+                    "%s Failed to initialize InvocationMonitor: %s",
+                    self.log_identifier,
+                    im_e,
+                )
+                self.invocation_monitor = None
+            try:
+                log.info(
+                    "%s Initializing synchronous ADK services...", self.log_identifier
+                )
+                self.session_service = initialize_session_service(self)
+                self.artifact_service = initialize_artifact_service(self)
+                self.memory_service = initialize_memory_service(self)
+                log.info(
+                    "%s Synchronous ADK services initialized.", self.log_identifier
+                )
+            except Exception as service_err:
+                log.exception(
+                    "%s Failed to initialize synchronous ADK services: %s",
+                    self.log_identifier,
+                    service_err,
+                )
+                raise RuntimeError(
+                    f"Failed to initialize synchronous ADK services: {service_err}"
+                ) from service_err
+            log.info(
+                "%s Starting dedicated async thread for MCP/ADK initialization...",
+                self.log_identifier,
+            )
+            self._async_loop = asyncio.new_event_loop()
+            self._async_init_future = concurrent.futures.Future()
+            self._async_thread = threading.Thread(
+                target=self._start_async_loop, daemon=True
+            )
+            self._async_thread.start()
+            init_coro_future = asyncio.run_coroutine_threadsafe(
+                self._perform_async_init(), self._async_loop
+            )
+            log.info(
+                "%s Waiting for async initialization to complete...",
+                self.log_identifier,
+            )
+            try:
+                init_coro_future.result(timeout=60)
+                self._async_init_future.result(timeout=1)
+                log.info(
+                    "%s Async initialization completed successfully.",
+                    self.log_identifier,
+                )
+            except Exception as init_err:
+                log.error(
+                    "%s Async initialization failed during __init__: %s",
+                    self.log_identifier,
+                    init_err,
+                )
+                self.cleanup()
+                raise RuntimeError(
+                    f"Failed to initialize component asynchronously: {init_err}"
+                ) from init_err
+            publish_interval_sec = self.agent_card_publishing_config.get(
+                "interval_seconds"
+            )
+            if publish_interval_sec and publish_interval_sec > 0:
+                log.info(
+                    "%s Scheduling agent card publishing every %d seconds.",
+                    self.log_identifier,
+                    publish_interval_sec,
+                )
+                self.add_timer(
+                    delay_ms=1000,
+                    timer_id=self._card_publish_timer_id,
+                    interval_ms=publish_interval_sec * 1000,
+                )
+            else:
+                log.warning(
+                    "%s Agent card publishing interval not configured or invalid, card will not be published periodically.",
+                    self.log_identifier,
+                )
+            log.info(
+                "%s Initialization complete for agent: %s",
+                self.log_identifier,
+                self.agent_name,
+            )
+        except Exception as e:
+            log.exception("%s Initialization failed: %s", self.log_identifier, e)
+            raise
+
+    def invoke(self, message: SolaceMessage, data: dict) -> dict:
+        """Placeholder invoke method. Primary logic resides in process_event."""
+        log.warning(
+            "%s 'invoke' method called, but primary logic resides in 'process_event'. This should not happen in normal operation.",
+            self.log_identifier,
+        )
+        return None
+
+    def process_event(self, event: Event):
+        """Processes incoming events (Messages, Timers, etc.)."""
+        try:
+            loop = self.get_async_loop()
+            is_loop_running = loop.is_running() if loop else False
+            if loop and is_loop_running:
+                coro = process_event(self, event)
+                future = asyncio.run_coroutine_threadsafe(coro, loop)
+                future.add_done_callback(
+                    functools.partial(
+                        self._handle_scheduled_task_completion,
+                        event_type_for_log=event.event_type,
+                    )
+                )
+            else:
+                log.error(
+                    "%s Async loop not available or not running (loop is %s, is_running: %s). Cannot process event: %s",
+                    self.log_identifier,
+                    "present" if loop else "None",
+                    is_loop_running,
+                    event.event_type,
+                )
+                if event.event_type == EventType.MESSAGE:
+                    try:
+                        event.data.call_negative_acknowledgements()
+                        log.warning(
+                            "%s NACKed message due to unavailable async loop for event processing.",
+                            self.log_identifier,
+                        )
+                    except Exception as nack_e:
+                        log.error(
+                            "%s Failed to NACK message after async loop issue: %s",
+                            self.log_identifier,
+                            nack_e,
+                        )
+        except Exception as e:
+            log.error(
+                "%s Error processing event: %s. Exception: %s",
+                self.log_identifier,
+                event.event_type,
+                e,
+            )
+            if event.event_type == EventType.MESSAGE:
+                try:
+                    event.data.call_negative_acknowledgements()
+                    log.warning(
+                        "%s NACKed message due to error in event processing.",
+                        self.log_identifier,
+                    )
+                except Exception as nack_e:
+                    log.error(
+                        "%s Failed to NACK message after error in event processing: %s",
+                        self.log_identifier,
+                        nack_e,
+                    )
+
+    def handle_timer_event(self, timer_data: Dict[str, Any]):
+        """Handles timer events, specifically for agent card publishing."""
+        log.debug("%s Received timer event: %s", self.log_identifier, timer_data)
+        if timer_data.get("timer_id") == self._card_publish_timer_id:
+            publish_agent_card(self)
+
+    async def handle_cache_expiry_event(self, cache_data: Dict[str, Any]):
+        """
+        Handles cache expiry events for peer timeouts by calling the atomic claim helper.
+        """
+        log.debug("%s Received cache expiry event: %s", self.log_identifier, cache_data)
+        sub_task_id = cache_data.get("key")
+        logical_task_id = cache_data.get("expired_data")
+
+        if not (
+            sub_task_id
+            and sub_task_id.startswith(CORRELATION_DATA_PREFIX)
+            and logical_task_id
+        ):
+            log.debug(
+                "%s Cache expiry for key '%s' is not a peer sub-task timeout or is missing data.",
+                self.log_identifier,
+                sub_task_id,
+            )
+            return
+
+        correlation_data = await self._claim_peer_sub_task_completion(
+            sub_task_id=sub_task_id, logical_task_id_from_event=logical_task_id
+        )
+
+        if correlation_data:
+            log.warning(
+                "%s Detected timeout for sub-task %s (Main Task: %s). Claimed successfully.",
+                self.log_identifier,
+                sub_task_id,
+                logical_task_id,
+            )
+            await self._handle_peer_timeout(sub_task_id, correlation_data)
+        else:
+            log.info(
+                "%s Ignoring timeout event for sub-task %s as it was already completed.",
+                self.log_identifier,
+                sub_task_id,
+            )
+
+    async def _get_correlation_data_for_sub_task(
+        self, sub_task_id: str
+    ) -> Optional[Dict[str, Any]]:
+        """
+        Non-destructively retrieves correlation data for a sub-task.
+        Used for intermediate events where the sub-task should remain active.
+        """
+        logical_task_id = self.cache_service.get_data(sub_task_id)
+        if not logical_task_id:
+            log.warning(
+                "%s No cache entry for sub-task %s. Cannot get correlation data.",
+                self.log_identifier,
+                sub_task_id,
+            )
+            return None
+
+        with self.active_tasks_lock:
+            task_context = self.active_tasks.get(logical_task_id)
+
+        if not task_context:
+            log.error(
+                "%s TaskExecutionContext not found for task %s, but cache entry existed for sub-task %s. This may indicate a cleanup issue.",
+                self.log_identifier,
+                logical_task_id,
+                sub_task_id,
+            )
+            return None
+
+        with task_context.lock:
+            return task_context.active_peer_sub_tasks.get(sub_task_id)
+
+    async def _claim_peer_sub_task_completion(
+        self, sub_task_id: str, logical_task_id_from_event: Optional[str] = None
+    ) -> Optional[Dict[str, Any]]:
+        """
+        Atomically claims a sub-task as complete, preventing race conditions.
+        This is a destructive operation that removes state.
+
+        Args:
+            sub_task_id: The ID of the sub-task to claim.
+            logical_task_id_from_event: The parent task ID, if provided by the event (e.g., a timeout).
+                                        If not provided, it will be looked up from the cache.
+        """
+        log_id = f"{self.log_identifier}[ClaimSubTask:{sub_task_id}]"
+        logical_task_id = logical_task_id_from_event
+
+        if not logical_task_id:
+            logical_task_id = self.cache_service.get_data(sub_task_id)
+            if not logical_task_id:
+                log.warning(
+                    "%s No cache entry found. Task has likely timed out and been cleaned up. Cannot claim.",
+                    log_id,
+                )
+                return None
+
+        with self.active_tasks_lock:
+            task_context = self.active_tasks.get(logical_task_id)
+
+        if not task_context:
+            log.error(
+                "%s TaskExecutionContext not found for task %s. Cleaning up stale cache entry.",
+                log_id,
+                logical_task_id,
+            )
+            self.cache_service.remove_data(sub_task_id)
+            return None
+
+        correlation_data = task_context.claim_sub_task_completion(sub_task_id)
+
+        if correlation_data:
+            # If we successfully claimed the task, remove the timeout tracker from the cache.
+            self.cache_service.remove_data(sub_task_id)
+            log.info("%s Successfully claimed completion.", log_id)
+            return correlation_data
+        else:
+            # This means the task was already claimed by a competing event (e.g., timeout vs. response).
+            log.warning("%s Failed to claim; it was already completed.", log_id)
+            return None
+
+    async def _retrigger_agent_with_peer_responses(
+        self,
+        results_to_inject: list,
+        correlation_data: dict,
+        task_context: "TaskExecutionContext",
+    ):
+        """
+        Injects peer tool responses into the session history and re-triggers the ADK runner.
+        This function contains the logic to correctly merge parallel tool call responses.
+        """
+        original_task_context = correlation_data.get("original_task_context")
+        logical_task_id = correlation_data.get("logical_task_id")
+        paused_invocation_id = correlation_data.get("invocation_id")
+        log_retrigger = f"{self.log_identifier}[RetriggerManager:{logical_task_id}]"
+
+        try:
+            effective_session_id = original_task_context.get("effective_session_id")
+            user_id = original_task_context.get("user_id")
+            session = await self.session_service.get_session(
+                app_name=self.agent_name,
+                user_id=user_id,
+                session_id=effective_session_id,
+            )
+            if not session:
+                raise RuntimeError(
+                    f"Could not find ADK session '{effective_session_id}'"
+                )
+
+            new_response_parts = []
+            for result in results_to_inject:
+                part = adk_types.Part.from_function_response(
+                    name=result["peer_tool_name"],
+                    response=result["payload"],
+                )
+                part.function_response.id = result["adk_function_call_id"]
+                new_response_parts.append(part)
+
+            # Always create a new event for the incoming peer responses.
+            # The ADK's `contents` processor is responsible for merging multiple
+            # tool responses into a single message before the next LLM call.
+            log.info(
+                "%s Creating a new tool response event for %d peer responses.",
+                log_retrigger,
+                len(new_response_parts),
+            )
+            new_adk_event = ADKEvent(
+                invocation_id=paused_invocation_id,
+                author=self.agent_name,
+                content=adk_types.Content(role="tool", parts=new_response_parts),
+            )
+            await self.session_service.append_event(
+                session=session, event=new_adk_event
+            )
+
+            # Always use SSE streaming mode for the ADK runner, even on re-trigger.
+            # This ensures that real-time callbacks for status updates and artifact
+            # creation can function correctly for all turns of a task.
+            streaming_mode = StreamingMode.SSE
+            max_llm_calls = self.get_config("max_llm_calls_per_task", 20)
+            run_config = RunConfig(
+                streaming_mode=streaming_mode, max_llm_calls=max_llm_calls
+            )
+
+            log.info(
+                "%s Re-triggering ADK runner for main task %s.",
+                log_retrigger,
+                logical_task_id,
+            )
+            try:
+                await run_adk_async_task_thread_wrapper(
+                    self, session, None, run_config, original_task_context
+                )
+            finally:
+                log.info(
+                    "%s Cleaning up parallel invocation state for invocation %s.",
+                    log_retrigger,
+                    paused_invocation_id,
+                )
+                task_context.clear_parallel_invocation_state(paused_invocation_id)
+
+        except Exception as e:
+            log.exception(
+                "%s Failed to re-trigger ADK runner for task %s: %s",
+                log_retrigger,
+                logical_task_id,
+                e,
+            )
+            if original_task_context:
+                loop = self.get_async_loop()
+                if loop and loop.is_running():
+                    asyncio.run_coroutine_threadsafe(
+                        self.finalize_task_error(e, original_task_context), loop
+                    )
+                else:
+                    log.error(
+                        "%s Async loop not available. Cannot schedule error finalization for task %s.",
+                        log_retrigger,
+                        logical_task_id,
+                    )
+
+    async def _handle_peer_timeout(
+        self,
+        sub_task_id: str,
+        correlation_data: Dict[str, Any],
+    ):
+        """
+        Handles the timeout of a peer agent task. It sends a cancellation request
+        to the peer, updates the local completion counter, and potentially
+        re-triggers the runner if all parallel tasks are now complete.
+        """
+        logical_task_id = correlation_data.get("logical_task_id")
+        invocation_id = correlation_data.get("invocation_id")
+        log_retrigger = f"{self.log_identifier}[RetriggerManager:{logical_task_id}]"
+
+        log.warning(
+            "%s Peer request timed out for sub-task: %s (Invocation: %s)",
+            log_retrigger,
+            sub_task_id,
+            invocation_id,
+        )
+
+        # Proactively send a cancellation request to the peer agent.
+        peer_agent_name = correlation_data.get("peer_agent_name")
+        if peer_agent_name:
+            try:
+                log.info(
+                    "%s Sending CancelTaskRequest to peer '%s' for timed-out sub-task %s.",
+                    log_retrigger,
+                    peer_agent_name,
+                    sub_task_id,
+                )
+                task_id_for_peer = sub_task_id.replace(CORRELATION_DATA_PREFIX, "", 1)
+                cancel_params = TaskIdParams(id=task_id_for_peer)
+                cancel_request = CancelTaskRequest(params=cancel_params)
+                user_props = {"clientId": self.agent_name}
+                peer_topic = self._get_agent_request_topic(peer_agent_name)
+                self._publish_a2a_message(
+                    payload=cancel_request.model_dump(exclude_none=True),
+                    topic=peer_topic,
+                    user_properties=user_props,
+                )
+            except Exception as e:
+                log.error(
+                    "%s Failed to send CancelTaskRequest to peer '%s' for sub-task %s: %s",
+                    log_retrigger,
+                    peer_agent_name,
+                    sub_task_id,
+                    e,
+                )
+
+        # Process the timeout locally.
+        with self.active_tasks_lock:
+            task_context = self.active_tasks.get(logical_task_id)
+
+        if not task_context:
+            log.warning(
+                "%s TaskExecutionContext not found for task %s. Ignoring timeout event.",
+                log_retrigger,
+                logical_task_id,
+            )
+            return
+
+        timeout_value = self.inter_agent_communication_config.get(
+            "request_timeout_seconds", DEFAULT_COMMUNICATION_TIMEOUT
+        )
+        all_sub_tasks_completed = task_context.handle_peer_timeout(
+            sub_task_id, correlation_data, timeout_value, invocation_id
+        )
+
+        if not all_sub_tasks_completed:
+            log.info(
+                "%s Waiting for more peer responses for invocation %s after timeout of sub-task %s.",
+                log_retrigger,
+                invocation_id,
+                sub_task_id,
+            )
+            return
+
+        log.info(
+            "%s All peer responses/timeouts received for invocation %s. Retriggering agent.",
+            log_retrigger,
+            invocation_id,
+        )
+        results_to_inject = task_context.parallel_tool_calls[invocation_id].get(
+            "results", []
+        )
+
+        await self._retrigger_agent_with_peer_responses(
+            results_to_inject, correlation_data, task_context
+        )
+
+    def _inject_peer_tools_callback(
+        self, callback_context: CallbackContext, llm_request: LlmRequest
+    ) -> Optional[LlmResponse]:
+        """
+        ADK before_model_callback to dynamically add PeerAgentTools to the LLM request
+        and generate the corresponding instruction text for the LLM.
+        """
+        log.debug("%s Running _inject_peer_tools_callback...", self.log_identifier)
+        if not self.peer_agents:
+            log.debug("%s No peer agents currently discovered.", self.log_identifier)
+            return None
+
+        a2a_context = callback_context.state.get("a2a_context", {})
+        user_config = (
+            a2a_context.get("a2a_user_config", {})
+            if isinstance(a2a_context, dict)
+            else {}
+        )
+
+        inter_agent_config = self.get_config("inter_agent_communication", {})
+        allow_list = inter_agent_config.get("allow_list", ["*"])
+        deny_list = set(self.get_config("deny_list", []))
+        self_name = self.get_config("agent_name")
+
+        peer_tools_to_add = []
+        allowed_peer_descriptions = []
+
+        for peer_name, agent_card in self.peer_agents.items():
+            if not isinstance(agent_card, AgentCard) or peer_name == self_name:
+                continue
+
+            is_allowed = any(
+                fnmatch.fnmatch(peer_name, p) for p in allow_list
+            ) and not any(fnmatch.fnmatch(peer_name, p) for p in deny_list)
+
+            if is_allowed:
+                config_resolver = MiddlewareRegistry.get_config_resolver()
+                operation_spec = {
+                    "operation_type": "peer_delegation",
+                    "target_agent": peer_name,
+                    "delegation_context": "peer_discovery",
+                }
+                validation_context = {
+                    "discovery_phase": "peer_enumeration",
+                    "agent_context": {"component_type": "peer_discovery"},
+                }
+                validation_result = config_resolver.validate_operation_config(
+                    user_config, operation_spec, validation_context
+                )
+                if not validation_result.get("valid", True):
+                    log.debug(
+                        "%s Peer agent '%s' filtered out by user configuration.",
+                        self.log_identifier,
+                        peer_name,
+                    )
+                    is_allowed = False
+
+            if not is_allowed:
+                continue
+
+            try:
+                peer_tool_instance = PeerAgentTool(
+                    target_agent_name=peer_name, host_component=self
+                )
+                if peer_tool_instance.name not in llm_request.tools_dict:
+                    peer_tools_to_add.append(peer_tool_instance)
+                    description = (
+                        getattr(agent_card, "description", "No description")
+                        or "No description"
+                    )
+                    allowed_peer_descriptions.append(
+                        f"- `peer_{peer_name}`: {description}"
+                    )
+            except Exception as e:
+                log.error(
+                    "%s Failed to create PeerAgentTool for '%s': %s",
+                    self.log_identifier,
+                    peer_name,
+                    e,
+                )
+
+        if allowed_peer_descriptions:
+            peer_list_str = "\n".join(allowed_peer_descriptions)
+            instruction_text = (
+                "You can delegate tasks to other specialized agents if they are better suited.\n"
+                "Use the `peer_<agent_name>(task_description: str, user_query: str)` tool for delegation. "
+                "Replace `<agent_name>` with the actual name of the target agent.\n"
+                "Provide a clear `task_description` for the peer and include the original `user_query` for context.\n"
+                "Be aware that the peer agent may not have access to your session history, so you must provide all required context necessary to fulfill the request.\n\n"
+                "Available peer agents you can delegate to (use the `peer_...` tool name):\n"
+                f"{peer_list_str}"
+            )
+            callback_context.state["peer_tool_instructions"] = instruction_text
+            log.debug(
+                "%s Stored peer tool instructions in callback_context.state.",
+                self.log_identifier,
+            )
+
+        if peer_tools_to_add:
+            try:
+                if llm_request.config.tools is None:
+                    llm_request.config.tools = []
+                if len(llm_request.config.tools) > 0:
+                    for tool in peer_tools_to_add:
+                        llm_request.tools_dict[tool.name] = tool
+                        llm_request.config.tools[0].function_declarations.append(
+                            tool._get_declaration()
+                        )
+                else:
+                    llm_request.append_tools(peer_tools_to_add)
+                log.debug(
+                    "%s Dynamically added %d PeerAgentTool(s) to LLM request.",
+                    self.log_identifier,
+                    len(peer_tools_to_add),
+                )
+            except Exception as e:
+                log.error(
+                    "%s Failed to append dynamic peer tools to LLM request: %s",
+                    self.log_identifier,
+                    e,
+                )
+        return None
+
+    def _filter_tools_by_capability_callback(
+        self, callback_context: CallbackContext, llm_request: LlmRequest
+    ) -> Optional[LlmResponse]:
+        """
+        ADK before_model_callback to filter tools in the LlmRequest based on user configuration.
+        This callback modifies `llm_request.config.tools` in place by potentially
+        removing individual FunctionDeclarations from genai.Tool objects or removing
+        entire genai.Tool objects if all their declarations are filtered out.
+        """
+        log_id_prefix = f"{self.log_identifier}[ToolCapabilityFilter]"
+        log.debug("%s Running _filter_tools_by_capability_callback...", log_id_prefix)
+
+        a2a_context = callback_context.state.get("a2a_context", {})
+        if not isinstance(a2a_context, dict):
+            log.warning(
+                "%s 'a2a_context' in session state is not a dictionary. Using empty configuration.",
+                log_id_prefix,
+            )
+            a2a_context = {}
+        user_config = a2a_context.get("a2a_user_config", {})
+        if not isinstance(user_config, dict):
+            log.warning(
+                "%s 'a2a_user_config' in a2a_context is not a dictionary. Using empty configuration.",
+                log_id_prefix,
+            )
+            user_config = {}
+
+        log.debug(
+            "%s User configuration for filtering: %s",
+            log_id_prefix,
+            {k: v for k, v in user_config.items() if not k.startswith("_")},
+        )
+
+        config_resolver = MiddlewareRegistry.get_config_resolver()
+
+        if not llm_request.config or not llm_request.config.tools:
+            log.debug("%s No tools in request to filter.", log_id_prefix)
+            return None
+
+        explicit_tools_config = self.get_config("tools", [])
+        final_filtered_genai_tools: List[adk_types.Tool] = []
+        original_genai_tools_count = len(llm_request.config.tools)
+        original_function_declarations_count = 0
+
+        for original_tool in llm_request.config.tools:
+            if not original_tool.function_declarations:
+                log.warning(
+                    "%s genai.Tool object has no function declarations. Keeping it.",
+                    log_id_prefix,
+                )
+                final_filtered_genai_tools.append(original_tool)
+                continue
+
+            original_function_declarations_count += len(
+                original_tool.function_declarations
+            )
+            permitted_declarations_for_this_tool: List[
+                adk_types.FunctionDeclaration
+            ] = []
+
+            for func_decl in original_tool.function_declarations:
+                func_decl_name = func_decl.name
+                tool_source_for_log = "unknown"
+                tool_matched_for_capability_lookup = False
+
+                feature_descriptor = {
+                    "feature_type": "tool_function",
+                    "function_name": func_decl_name,
+                    "tool_source": self._determine_tool_source(func_decl_name),
+                    "tool_metadata": self._get_tool_metadata(func_decl_name),
+                }
+
+                if func_decl_name.startswith(PEER_TOOL_PREFIX):
+                    peer_name = func_decl_name.replace(PEER_TOOL_PREFIX, "", 1)
+                    feature_descriptor["tool_metadata"]["peer_agent_name"] = peer_name
+                    tool_source_for_log = f"PeerAgentTool ({peer_name})"
+                    tool_matched_for_capability_lookup = True
+
+                if not tool_matched_for_capability_lookup:
+                    tool_def = tool_registry.get_tool_by_name(func_decl_name)
+                    if tool_def:
+                        feature_descriptor["tool_metadata"][
+                            "tool_category"
+                        ] = tool_def.category
+                        feature_descriptor["tool_metadata"]["builtin_tool"] = True
+                        tool_source_for_log = (
+                            f"Built-in Tool ({tool_def.category}/{func_decl_name})"
+                        )
+                        tool_matched_for_capability_lookup = True
+
+                if not tool_matched_for_capability_lookup:
+                    for tool_cfg in explicit_tools_config:
+                        cfg_tool_type = tool_cfg.get("tool_type")
+                        cfg_tool_name = tool_cfg.get("tool_name")
+                        cfg_func_name = tool_cfg.get("function_name")
+                        if (
+                            cfg_tool_type == "python"
+                            and cfg_func_name == func_decl_name
+                        ) or (
+                            cfg_tool_type in ["builtin", "mcp"]
+                            and cfg_tool_name == func_decl_name
+                        ):
+                            feature_descriptor["tool_metadata"][
+                                "tool_type"
+                            ] = cfg_tool_type
+                            feature_descriptor["tool_metadata"][
+                                "tool_config"
+                            ] = tool_cfg
+                            tool_source_for_log = f"Explicitly configured tool ({cfg_tool_type}: {cfg_tool_name or cfg_func_name})"
+                            tool_matched_for_capability_lookup = True
+                            break
+
+                if not tool_matched_for_capability_lookup:
+                    log.debug(
+                        "%s FunctionDeclaration '%s' not found in any known configuration for capability checking. Assuming feature is available.",
+                        log_id_prefix,
+                        func_decl_name,
+                    )
+                    tool_source_for_log = "Unmatched/Implicit FunctionDeclaration"
+
+                context = {
+                    "agent_context": self.get_agent_context(),
+                    "filter_phase": "pre_llm",
+                    "tool_configurations": {
+                        "explicit_tools": explicit_tools_config,
+                    },
+                }
+
+                if config_resolver.is_feature_enabled(
+                    user_config, feature_descriptor, context
+                ):
+                    permitted_declarations_for_this_tool.append(func_decl)
+                    log.debug(
+                        "%s FunctionDeclaration '%s' (Source: %s) permitted.",
+                        log_id_prefix,
+                        func_decl_name,
+                        tool_source_for_log,
+                    )
+                else:
+                    log.info(
+                        "%s FunctionDeclaration '%s' (Source: %s) FILTERED OUT due to configuration restrictions.",
+                        log_id_prefix,
+                        func_decl_name,
+                        tool_source_for_log,
+                    )
+
+            if permitted_declarations_for_this_tool:
+                scoped_tool = original_tool.model_copy(deep=True)
+                scoped_tool.function_declarations = permitted_declarations_for_this_tool
+
+                final_filtered_genai_tools.append(scoped_tool)
+                log.debug(
+                    "%s Keeping genai.Tool (original name/type preserved, declarations filtered) as it has %d permitted FunctionDeclaration(s).",
+                    log_id_prefix,
+                    len(permitted_declarations_for_this_tool),
+                )
+            else:
+                log.info(
+                    "%s Entire genai.Tool (original declarations: %s) FILTERED OUT as all its FunctionDeclarations were denied by configuration.",
+                    log_id_prefix,
+                    [fd.name for fd in original_tool.function_declarations],
+                )
+
+        final_function_declarations_count = sum(
+            len(t.function_declarations)
+            for t in final_filtered_genai_tools
+            if t.function_declarations
+        )
+
+        if final_function_declarations_count != original_function_declarations_count:
+            log.info(
+                "%s Tool list modified by capability filter. Original genai.Tools: %d (Total Declarations: %d). Filtered genai.Tools: %d (Total Declarations: %d).",
+                log_id_prefix,
+                original_genai_tools_count,
+                original_function_declarations_count,
+                len(final_filtered_genai_tools),
+                final_function_declarations_count,
+            )
+            llm_request.config.tools = (
+                final_filtered_genai_tools if final_filtered_genai_tools else None
+            )
+        else:
+            log.debug(
+                "%s Tool list and FunctionDeclarations unchanged after capability filtering.",
+                log_id_prefix,
+            )
+
+        return None
+
+    def _determine_tool_source(self, function_name: str) -> str:
+        """Determine the source/type of a tool function."""
+        if function_name.startswith("peer_"):
+            return "peer_agent"
+
+        tool_def = tool_registry.get_tool_by_name(function_name)
+        if tool_def:
+            category_map = {
+                "artifact_management": "builtin_artifact",
+                "data_analysis": "builtin_data",
+            }
+            return category_map.get(tool_def.category, "builtin_other")
+
+        return "explicit_tool"
+
+    def _get_tool_metadata(self, function_name: str) -> Dict[str, Any]:
+        """Get metadata for a tool function."""
+        metadata = {"function_name": function_name}
+
+        if function_name.startswith("peer_"):
+            peer_name = function_name.replace("peer_", "", 1)
+            metadata.update(
+                {"peer_agent_name": peer_name, "operation_type": "delegation"}
+            )
+            return metadata
+
+        tool_def = tool_registry.get_tool_by_name(function_name)
+        if tool_def:
+            metadata.update(
+                {
+                    "tool_category": tool_def.category,
+                    "required_scopes": tool_def.required_scopes,
+                    "builtin_tool": True,
+                }
+            )
+            return metadata
+
+        explicit_tools_config = self.get_config("tools", [])
+        for tool_cfg in explicit_tools_config:
+            cfg_tool_name = tool_cfg.get("tool_name")
+            cfg_func_name = tool_cfg.get("function_name")
+            if (
+                tool_cfg.get("tool_type") == "python" and cfg_func_name == function_name
+            ) or (
+                tool_cfg.get("tool_type") in ["builtin", "mcp"]
+                and cfg_tool_name == function_name
+            ):
+                metadata.update(
+                    {"tool_type": tool_cfg.get("tool_type"), "tool_config": tool_cfg}
+                )
+                break
+
+        return metadata
+
+    def get_agent_context(self) -> Dict[str, Any]:
+        """Get agent context for middleware calls."""
+        return {
+            "agent_name": getattr(self, "agent_name", "unknown"),
+            "component_type": "sac_agent",
+        }
+
+    def _inject_gateway_instructions_callback(
+        self, callback_context: CallbackContext, llm_request: LlmRequest
+    ) -> Optional[LlmResponse]:
+        """
+        ADK before_model_callback to dynamically prepend gateway-defined system_purpose
+        and response_format to the agent's llm_request.config.system_instruction.
+        """
+        log_id_prefix = f"{self.log_identifier}[GatewayInstrInject]"
+        log.debug(
+            "%s Running _inject_gateway_instructions_callback to modify system_instruction...",
+            log_id_prefix,
+        )
+
+        a2a_context = callback_context.state.get("a2a_context", {})
+        if not isinstance(a2a_context, dict):
+            log.warning(
+                "%s 'a2a_context' in session state is not a dictionary. Skipping instruction injection.",
+                log_id_prefix,
+            )
+            return None
+
+        system_purpose = a2a_context.get("system_purpose")
+        response_format = a2a_context.get("response_format")
+        user_profile = a2a_context.get("a2a_user_config", {}).get("user_profile")
+
+        inject_purpose = self.get_config("inject_system_purpose", False)
+        inject_format = self.get_config("inject_response_format", False)
+        inject_user_profile = self.get_config("inject_user_profile", False)
+
+        gateway_instructions_to_add = []
+
+        if (
+            inject_purpose
+            and system_purpose
+            and isinstance(system_purpose, str)
+            and system_purpose.strip()
+        ):
+            gateway_instructions_to_add.append(
+                f"System Purpose:\n{system_purpose.strip()}"
+            )
+            log.debug(
+                "%s Prepared system_purpose for system_instruction.", log_id_prefix
+            )
+
+        if (
+            inject_format
+            and response_format
+            and isinstance(response_format, str)
+            and response_format.strip()
+        ):
+            gateway_instructions_to_add.append(
+                f"Desired Response Format:\n{response_format.strip()}"
+            )
+            log.debug(
+                "%s Prepared response_format for system_instruction.", log_id_prefix
+            )
+
+        if (
+            inject_user_profile
+            and user_profile
+            and (isinstance(user_profile, str) or isinstance(user_profile, dict))
+        ):
+            if isinstance(user_profile, dict):
+                user_profile = json.dumps(user_profile, indent=2, default=str)
+            gateway_instructions_to_add.append(
+                f"Inquiring User Profile:\n{user_profile.strip()}\n"
+            )
+            log.debug("%s Prepared user_profile for system_instruction.", log_id_prefix)
+
+        if not gateway_instructions_to_add:
+            log.debug(
+                "%s No gateway instructions to inject into system_instruction.",
+                log_id_prefix,
+            )
+            return None
+
+        if llm_request.config is None:
+            log.warning(
+                "%s llm_request.config is None, cannot append gateway instructions to system_instruction.",
+                log_id_prefix,
+            )
+            return None
+
+        if llm_request.config.system_instruction is None:
+            llm_request.config.system_instruction = ""
+
+        combined_new_instructions = "\n\n".join(gateway_instructions_to_add)
+
+        if llm_request.config.system_instruction:
+            llm_request.config.system_instruction += (
+                f"\n\n---\n\n{combined_new_instructions}"
+            )
+        else:
+            llm_request.config.system_instruction = combined_new_instructions
+
+        log.info(
+            "%s Injected %d gateway instruction block(s) into llm_request.config.system_instruction.",
+            log_id_prefix,
+            len(gateway_instructions_to_add),
+        )
+
+        return None
+
+    async def _publish_text_as_partial_a2a_status_update(
+        self,
+        text_content: str,
+        a2a_context: Dict,
+        is_stream_terminating_content: bool = False,
+    ):
+        """
+        Constructs and publishes a TaskStatusUpdateEvent for the given text.
+        The 'final' flag is determined by is_stream_terminating_content.
+        This method skips buffer flushing since it's used for LLM streaming text.
+        """
+        logical_task_id = a2a_context.get("logical_task_id", "unknown_task")
+        log_identifier_helper = (
+            f"{self.log_identifier}[PublishPartialText:{logical_task_id}]"
+        )
+
+        if not text_content:
+            log.debug(
+                "%s No text content to publish as update (final=%s).",
+                log_identifier_helper,
+                is_stream_terminating_content,
+            )
+            return
+
+        try:
+            a2a_message = A2AMessage(role="agent", parts=[TextPart(text=text_content)])
+            task_status = TaskStatus(
+                state=TaskState.WORKING,
+                message=a2a_message,
+                timestamp=datetime.now(timezone.utc),
+            )
+            event_metadata = {"agent_name": self.agent_name}
+            status_update_event = TaskStatusUpdateEvent(
+                id=logical_task_id,
+                status=task_status,
+                final=is_stream_terminating_content,
+                metadata=event_metadata,
+            )
+
+            await self._publish_status_update_with_buffer_flush(
+                status_update_event,
+                a2a_context,
+                skip_buffer_flush=True,
+            )
+
+            log.debug(
+                "%s Published LLM streaming text (length: %d bytes, final: %s).",
+                log_identifier_helper,
+                len(text_content.encode("utf-8")),
+                is_stream_terminating_content,
+            )
+
+        except Exception as e:
+            log.exception(
+                "%s Error in _publish_text_as_partial_a2a_status_update: %s",
+                log_identifier_helper,
+                e,
+            )
+
+    async def _publish_agent_status_signal_update(
+        self, status_text: str, a2a_context: Dict
+    ):
+        """
+        Constructs and publishes a TaskStatusUpdateEvent specifically for agent_status_message signals.
+        This method will flush the buffer before publishing to maintain proper message ordering.
+        """
+        logical_task_id = a2a_context.get("logical_task_id", "unknown_task")
+        log_identifier_helper = (
+            f"{self.log_identifier}[PublishAgentSignal:{logical_task_id}]"
+        )
+
+        if not status_text:
+            log.debug(
+                "%s No text content for agent status signal.", log_identifier_helper
+            )
+            return
+
+        try:
+            signal_data_part = DataPart(
+                data={
+                    "a2a_signal_type": "agent_status_message",
+                    "text": status_text,
+                },
+                metadata={"source_embed_type": "status_update"},
+            )
+            a2a_message = A2AMessage(role="agent", parts=[signal_data_part])
+            task_status = TaskStatus(
+                state=TaskState.WORKING,
+                message=a2a_message,
+                timestamp=datetime.now(timezone.utc),
+            )
+            event_metadata = {"agent_name": self.agent_name}
+            status_update_event = TaskStatusUpdateEvent(
+                id=logical_task_id,
+                status=task_status,
+                final=False,
+                metadata=event_metadata,
+            )
+
+            await self._publish_status_update_with_buffer_flush(
+                status_update_event,
+                a2a_context,
+                skip_buffer_flush=False,
+            )
+
+            log.debug(
+                "%s Published agent_status_message signal ('%s').",
+                log_identifier_helper,
+                status_text,
+            )
+
+        except Exception as e:
+            log.exception(
+                "%s Error in _publish_agent_status_signal_update: %s",
+                log_identifier_helper,
+                e,
+            )
+
+    async def _flush_buffer_if_needed(
+        self, a2a_context: Dict, reason: str = "status_update"
+    ) -> bool:
+        """
+        Flushes streaming buffer if it contains content.
+
+        Args:
+            a2a_context: The A2A context dictionary for the current task
+            reason: The reason for flushing (for logging purposes)
+
+        Returns:
+            bool: True if buffer was flushed, False if no content to flush
+        """
+        logical_task_id = a2a_context.get("logical_task_id", "unknown_task")
+        log_identifier = f"{self.log_identifier}[BufferFlush:{logical_task_id}]"
+
+        with self.active_tasks_lock:
+            task_context = self.active_tasks.get(logical_task_id)
+
+        if not task_context:
+            log.warning(
+                "%s TaskExecutionContext not found for task %s. Cannot flush buffer.",
+                log_identifier,
+                logical_task_id,
+            )
+            return False
+
+        buffer_content = task_context.get_streaming_buffer_content()
+        if not buffer_content:
+            log.debug(
+                "%s No buffer content to flush (reason: %s).",
+                log_identifier,
+                reason,
+            )
+            return False
+
+        buffer_size = len(buffer_content.encode("utf-8"))
+        log.info(
+            "%s Flushing buffer content (size: %d bytes, reason: %s).",
+            log_identifier,
+            buffer_size,
+            reason,
+        )
+
+        try:
+            resolved_text, unprocessed_tail = await self._flush_and_resolve_buffer(
+                a2a_context, is_final=False
+            )
+
+            if resolved_text:
+                await self._publish_text_as_partial_a2a_status_update(
+                    resolved_text,
+                    a2a_context,
+                    is_stream_terminating_content=False,
+                )
+                log.debug(
+                    "%s Successfully flushed and published buffer content (resolved: %d bytes).",
+                    log_identifier,
+                    len(resolved_text.encode("utf-8")),
+                )
+                return True
+            else:
+                log.debug(
+                    "%s Buffer flush completed but no resolved text to publish.",
+                    log_identifier,
+                )
+                return False
+
+        except Exception as e:
+            log.exception(
+                "%s Error during buffer flush (reason: %s): %s",
+                log_identifier,
+                reason,
+                e,
+            )
+            return False
+
+    async def _publish_status_update_with_buffer_flush(
+        self,
+        status_update_event: TaskStatusUpdateEvent,
+        a2a_context: Dict,
+        skip_buffer_flush: bool = False,
+    ) -> None:
+        """
+        Central method for publishing status updates with automatic buffer flushing.
+
+        Args:
+            status_update_event: The status update event to publish
+            a2a_context: The A2A context dictionary for the current task
+            skip_buffer_flush: If True, skip buffer flushing (used for LLM streaming text)
+        """
+        logical_task_id = a2a_context.get("logical_task_id", "unknown_task")
+        jsonrpc_request_id = a2a_context.get("jsonrpc_request_id")
+        log_identifier = f"{self.log_identifier}[StatusUpdate:{logical_task_id}]"
+
+        status_type = "unknown"
+        if status_update_event.metadata:
+            if status_update_event.metadata.get("type") == "tool_invocation_start":
+                status_type = "tool_invocation_start"
+            elif "agent_name" in status_update_event.metadata:
+                status_type = "agent_status"
+
+        if (
+            status_update_event.status
+            and status_update_event.status.message
+            and status_update_event.status.message.parts
+        ):
+            for part in status_update_event.status.message.parts:
+                if hasattr(part, "data") and part.data:
+                    if part.data.get("a2a_signal_type") == "agent_status_message":
+                        status_type = "agent_status_signal"
+                        break
+                    elif "tool_error" in part.data:
+                        status_type = "tool_failure"
+                        break
+
+        log.debug(
+            "%s Publishing status update (type: %s, skip_buffer_flush: %s).",
+            log_identifier,
+            status_type,
+            skip_buffer_flush,
+        )
+
+        if not skip_buffer_flush:
+            buffer_was_flushed = await self._flush_buffer_if_needed(
+                a2a_context, reason=f"before_{status_type}_status"
+            )
+            if buffer_was_flushed:
+                log.info(
+                    "%s Buffer flushed before %s status update.",
+                    log_identifier,
+                    status_type,
+                )
+
+        try:
+            rpc_response = JSONRPCResponse(
+                id=jsonrpc_request_id, result=status_update_event
+            )
+            payload_to_publish = rpc_response.model_dump(exclude_none=True)
+
+            target_topic = a2a_context.get("statusTopic") or get_gateway_status_topic(
+                self.namespace, self.get_gateway_id(), logical_task_id
+            )
+
+            self._publish_a2a_event(payload_to_publish, target_topic, a2a_context)
+
+            log.info(
+                "%s Published %s status update to %s.",
+                log_identifier,
+                status_type,
+                target_topic,
+            )
+
+        except Exception as e:
+            log.exception(
+                "%s Error publishing %s status update: %s",
+                log_identifier,
+                status_type,
+                e,
+            )
+            raise
+
+    async def _translate_adk_part_to_a2a_filepart(
+        self,
+        adk_part: adk_types.Part,
+        filename: str,
+        a2a_context: Dict,
+        version: Optional[int] = None,
+    ) -> Optional[FilePart]:
+        """
+        Translates a loaded ADK Part (with inline_data) to an A2A FilePart
+        based on the configured artifact_handling_mode.
+        If version is not provided, it will be resolved to the latest.
+        """
+        if self.artifact_handling_mode == "ignore":
+            log.debug(
+                "%s Artifact handling mode is 'ignore'. Skipping translation for '%s'.",
+                self.log_identifier,
+                filename,
+            )
+            return None
+
+        if not adk_part or not adk_part.inline_data:
+            log.warning(
+                "%s Cannot translate artifact '%s': ADK Part is missing or has no inline_data.",
+                self.log_identifier,
+                filename,
+            )
+            return None
+
+        resolved_version = version
+        if resolved_version is None:
+            try:
+                resolved_version = await get_latest_artifact_version(
+                    artifact_service=self.artifact_service,
+                    app_name=self.get_config("agent_name"),
+                    user_id=a2a_context.get("user_id"),
+                    session_id=a2a_context.get("session_id"),
+                    filename=filename,
+                )
+                if resolved_version is None:
+                    log.error(
+                        "%s Could not resolve latest version for artifact '%s'.",
+                        self.log_identifier,
+                        filename,
+                    )
+                    return None
+            except Exception as e:
+                log.exception(
+                    "%s Failed to resolve latest version for artifact '%s': %s",
+                    self.log_identifier,
+                    filename,
+                    e,
+                )
+                return None
+
+        mime_type = adk_part.inline_data.mime_type
+        data_bytes = adk_part.inline_data.data
+        file_content: Optional[FileContent] = None
+
+        try:
+            if self.artifact_handling_mode == "embed":
+                encoded_bytes = base64.b64encode(data_bytes).decode("utf-8")
+                file_content = FileContent(
+                    name=filename, mimeType=mime_type, bytes=encoded_bytes
+                )
+                log.debug(
+                    "%s Embedding artifact '%s' (size: %d bytes) for A2A message.",
+                    self.log_identifier,
+                    filename,
+                    len(data_bytes),
+                )
+
+            elif self.artifact_handling_mode == "reference":
+                adk_app_name = self.get_config("agent_name")
+                user_id = a2a_context.get("user_id")
+                original_session_id = a2a_context.get("session_id")
+
+                if not all([adk_app_name, user_id, original_session_id]):
+                    log.error(
+                        "%s Cannot create artifact reference URI: missing context (app_name, user_id, or session_id).",
+                        self.log_identifier,
+                    )
+                    return None
+
+                artifact_uri = f"artifact://{adk_app_name}/{user_id}/{original_session_id}/{filename}?version={resolved_version}"
+
+                log.info(
+                    "%s Creating reference URI for artifact: %s",
+                    self.log_identifier,
+                    artifact_uri,
+                )
+                file_content = FileContent(
+                    name=filename, mimeType=mime_type, uri=artifact_uri
+                )
+
+            if file_content:
+                return FilePart(file=file_content)
+            else:
+                log.warning(
+                    "%s No FileContent created for artifact '%s' despite mode '%s'.",
+                    self.log_identifier,
+                    filename,
+                    self.artifact_handling_mode,
+                )
+                return None
+
+        except Exception as e:
+            log.exception(
+                "%s Error translating artifact '%s' to A2A FilePart (mode: %s): %s",
+                self.log_identifier,
+                filename,
+                self.artifact_handling_mode,
+                e,
+            )
+            return None
+
+    async def _filter_text_from_final_streaming_event(
+        self, adk_event: ADKEvent, a2a_context: Dict
+    ) -> ADKEvent:
+        """
+        Filters out text parts from the final ADKEvent of a turn for PERSISTENT streaming sessions.
+        This prevents sending redundant, aggregated text that was already streamed.
+        Non-text parts like function calls are preserved.
+        """
+        is_run_based_session = a2a_context.get("is_run_based_session", False)
+        is_streaming = a2a_context.get("is_streaming", False)
+        is_final_turn_event = not adk_event.partial
+        has_content_parts = adk_event.content and adk_event.content.parts
+
+        # Only filter for PERSISTENT (not run-based) streaming sessions.
+        if (
+            not is_run_based_session
+            and is_streaming
+            and is_final_turn_event
+            and has_content_parts
+        ):
+            log_id = f"{self.log_identifier}[FilterFinalStreamEvent:{a2a_context.get('logical_task_id', 'unknown')}]"
+            log.debug(
+                "%s Filtering final streaming event to remove redundant text.", log_id
+            )
+
+            non_text_parts = [
+                part for part in adk_event.content.parts if part.text is None
+            ]
+
+            if len(non_text_parts) < len(adk_event.content.parts):
+                event_copy = adk_event.model_copy(deep=True)
+                event_copy.content = (
+                    adk_types.Content(parts=non_text_parts) if non_text_parts else None
+                )
+                log.info(
+                    "%s Removed text from final streaming event. Kept %d non-text part(s).",
+                    log_id,
+                    len(non_text_parts),
+                )
+                return event_copy
+
+        return adk_event
+
+    async def process_and_publish_adk_event(
+        self, adk_event: ADKEvent, a2a_context: Dict
+    ):
+        """
+        Main orchestrator for processing ADK events.
+        Handles text buffering, embed resolution, and event routing based on
+        whether the event is partial or the final event of a turn.
+        """
+        logical_task_id = a2a_context.get("logical_task_id", "unknown_task")
+        log_id_main = (
+            f"{self.log_identifier}[ProcessADKEvent:{logical_task_id}:{adk_event.id}]"
+        )
+        log.debug(
+            "%s Received ADKEvent (Partial: %s, Final Turn: %s).",
+            log_id_main,
+            adk_event.partial,
+            not adk_event.partial,
+        )
+
+        if adk_event.content and adk_event.content.parts:
+            if any(
+                p.function_response
+                and p.function_response.name == "_continue_generation"
+                for p in adk_event.content.parts
+            ):
+                log.debug(
+                    "%s Discarding _continue_generation tool response event.",
+                    log_id_main,
+                )
+                return
+
+        if adk_event.custom_metadata and adk_event.custom_metadata.get(
+            "was_interrupted"
+        ):
+            log.debug(
+                "%s Found 'was_interrupted' signal. Skipping event.",
+                log_id_main,
+            )
+            return
+
+        with self.active_tasks_lock:
+            task_context = self.active_tasks.get(logical_task_id)
+
+        if not task_context:
+            log.error(
+                "%s TaskExecutionContext not found for task %s. Cannot process ADK event.",
+                log_id_main,
+                logical_task_id,
+            )
+            return
+
+        is_run_based_session = a2a_context.get("is_run_based_session", False)
+        is_final_turn_event = not adk_event.partial
+
+        if not is_final_turn_event:
+            if adk_event.content and adk_event.content.parts:
+                for part in adk_event.content.parts:
+                    if part.text is not None:
+                        task_context.append_to_streaming_buffer(part.text)
+                        log.debug(
+                            "%s Appended text to buffer. New buffer size: %d bytes",
+                            log_id_main,
+                            len(
+                                task_context.get_streaming_buffer_content().encode(
+                                    "utf-8"
+                                )
+                            ),
+                        )
+
+            buffer_content = task_context.get_streaming_buffer_content()
+            batching_disabled = self.stream_batching_threshold_bytes <= 0
+            buffer_has_content = bool(buffer_content)
+            threshold_met = (
+                buffer_has_content
+                and not batching_disabled
+                and (
+                    len(buffer_content.encode("utf-8"))
+                    >= self.stream_batching_threshold_bytes
+                )
+            )
+
+            if buffer_has_content and (batching_disabled or threshold_met):
+                log.info(
+                    "%s Partial event triggered buffer flush due to size/batching config.",
+                    log_id_main,
+                )
+                resolved_text, _ = await self._flush_and_resolve_buffer(
+                    a2a_context, is_final=False
+                )
+
+                if resolved_text:
+                    if is_run_based_session:
+                        task_context.append_to_run_based_buffer(resolved_text)
+                        log.debug(
+                            "%s [RUN_BASED] Appended %d bytes to run_based_response_buffer.",
+                            log_id_main,
+                            len(resolved_text.encode("utf-8")),
+                        )
+                    else:
+                        await self._publish_text_as_partial_a2a_status_update(
+                            resolved_text, a2a_context
+                        )
+        else:
+            buffer_content = task_context.get_streaming_buffer_content()
+            if buffer_content:
+                log.info(
+                    "%s Final event triggered flush of remaining buffer content.",
+                    log_id_main,
+                )
+                resolved_text, _ = await self._flush_and_resolve_buffer(
+                    a2a_context, is_final=True
+                )
+                if resolved_text:
+                    if is_run_based_session:
+                        task_context.append_to_run_based_buffer(resolved_text)
+                        log.debug(
+                            "%s [RUN_BASED] Appended final %d bytes to run_based_response_buffer.",
+                            log_id_main,
+                            len(resolved_text.encode("utf-8")),
+                        )
+                    else:
+                        await self._publish_text_as_partial_a2a_status_update(
+                            resolved_text, a2a_context
+                        )
+
+            # Prepare and publish the final event for observability
+            event_to_publish = await self._filter_text_from_final_streaming_event(
+                adk_event, a2a_context
+            )
+
+            (
+                a2a_payload,
+                target_topic,
+                user_properties,
+                _,
+            ) = await format_and_route_adk_event(event_to_publish, a2a_context, self)
+
+            if a2a_payload and target_topic:
+                self._publish_a2a_event(a2a_payload, target_topic, a2a_context)
+                log.info(
+                    "%s Published final turn event (e.g., tool call) to %s.",
+                    log_id_main,
+                    target_topic,
+                )
+            else:
+                log.debug(
+                    "%s Final turn event did not result in a publishable A2A message.",
+                    log_id_main,
+                )
+
+            await self._handle_artifact_return_signals(adk_event, a2a_context)
+
+    async def _flush_and_resolve_buffer(
+        self, a2a_context: Dict, is_final: bool
+    ) -> Tuple[str, str]:
+        """Flushes buffer, resolves embeds, handles signals, returns (resolved_text, unprocessed_tail)."""
+        logical_task_id = a2a_context.get("logical_task_id", "unknown_task")
+        log_id = f"{self.log_identifier}[FlushBuffer:{logical_task_id}]"
+
+        with self.active_tasks_lock:
+            task_context = self.active_tasks.get(logical_task_id)
+
+        if not task_context:
+            log.error(
+                "%s TaskExecutionContext not found for task %s. Cannot flush/resolve buffer.",
+                log_id,
+                logical_task_id,
+            )
+            return "", ""
+
+        text_to_process = task_context.flush_streaming_buffer()
+
+        resolved_text, signals_found, unprocessed_tail = (
+            await self._resolve_early_embeds_and_handle_signals(
+                text_to_process, a2a_context
+            )
+        )
+
+        if not is_final:
+            if unprocessed_tail:
+                task_context.append_to_streaming_buffer(unprocessed_tail)
+                log.debug(
+                    "%s Placed unprocessed tail (length %d) back into buffer.",
+                    log_id,
+                    len(unprocessed_tail.encode("utf-8")),
+                )
+        else:
+            if unprocessed_tail is not None and unprocessed_tail != "":
+                resolved_text = resolved_text + unprocessed_tail
+
+        if signals_found:
+            log.info(
+                "%s Handling %d signals from buffer resolution.",
+                log_id,
+                len(signals_found),
+            )
+            for _signal_index, signal_data_tuple in signals_found:
+                if (
+                    isinstance(signal_data_tuple, tuple)
+                    and len(signal_data_tuple) == 3
+                    and signal_data_tuple[0] is None
+                    and signal_data_tuple[1] == "SIGNAL_STATUS_UPDATE"
+                ):
+                    status_text = signal_data_tuple[2]
+                    log.info(
+                        "%s Publishing SIGNAL_STATUS_UPDATE from buffer: '%s'",
+                        log_id,
+                        status_text,
+                    )
+                    await self._publish_agent_status_signal_update(
+                        status_text, a2a_context
+                    )
+
+        return resolved_text, unprocessed_tail
+
+    async def _handle_artifact_return_signals(
+        self, adk_event: ADKEvent, a2a_context: Dict
+    ):
+        """
+        Processes artifact return signals.
+        This method is triggered by a placeholder in state_delta, but reads the
+        actual list of signals from the TaskExecutionContext.
+        """
+        logical_task_id = a2a_context.get("logical_task_id", "unknown_task")
+        log_id = f"{self.log_identifier}[ArtifactSignals:{logical_task_id}]"
+
+        # Check for the trigger in state_delta. The presence of any key is enough.
+        has_signal_trigger = (
+            adk_event.actions
+            and adk_event.actions.state_delta
+            and any(
+                k.startswith("temp:a2a_return_artifact:")
+                for k in adk_event.actions.state_delta
+            )
+        )
+
+        if not has_signal_trigger:
+            return
+
+        with self.active_tasks_lock:
+            task_context = self.active_tasks.get(logical_task_id)
+
+        if not task_context:
+            log.warning(
+                "%s No TaskExecutionContext found for task %s. Cannot process artifact signals.",
+                log_id,
+                logical_task_id,
+            )
+            return
+
+        all_signals = task_context.get_and_clear_artifact_signals()
+
+        if not all_signals:
+            log.info(
+                "%s Triggered for artifact signals, but none were found in the execution context.",
+                log_id,
+            )
+            return
+
+        log.info(
+            "%s Found %d artifact return signal(s) in the execution context.",
+            log_id,
+            len(all_signals),
+        )
+
+        original_session_id = a2a_context.get("session_id")
+        user_id = a2a_context.get("user_id")
+        adk_app_name = self.get_config("agent_name")
+
+        peer_status_topic = a2a_context.get("statusTopic")
+        namespace = self.get_config("namespace")
+        gateway_id = self.get_gateway_id()
+
+        artifact_topic = peer_status_topic or get_gateway_status_topic(
+            namespace, gateway_id, logical_task_id
+        )
+
+        if not self.artifact_service:
+            log.error("%s Artifact service not available.", log_id)
+            return
+        if not artifact_topic:
+            log.error("%s Could not determine artifact topic.", log_id)
+            return
+
+        for item in all_signals:
+            try:
+                filename = item["filename"]
+                version = item["version"]
+
+                log.info(
+                    "%s Processing artifact return signal for '%s' v%d from context.",
+                    log_id,
+                    filename,
+                    version,
+                )
+
+                loaded_adk_part = await self.artifact_service.load_artifact(
+                    app_name=adk_app_name,
+                    user_id=user_id,
+                    session_id=original_session_id,
+                    filename=filename,
+                    version=version,
+                )
+
+                if not loaded_adk_part:
+                    log.warning(
+                        "%s Failed to load artifact '%s' v%d.",
+                        log_id,
+                        filename,
+                        version,
+                    )
+                    continue
+
+                a2a_file_part = await self._translate_adk_part_to_a2a_filepart(
+                    loaded_adk_part, filename, a2a_context, version=version
+                )
+
+                if a2a_file_part:
+                    a2a_artifact = A2AArtifact(name=filename, parts=[a2a_file_part])
+                    artifact_update_event = TaskArtifactUpdateEvent(
+                        id=logical_task_id, artifact=a2a_artifact
+                    )
+                    artifact_payload = JSONRPCResponse(
+                        id=a2a_context.get("jsonrpc_request_id"),
+                        result=artifact_update_event,
+                    ).model_dump(exclude_none=True)
+
+                    self._publish_a2a_event(
+                        artifact_payload, artifact_topic, a2a_context
+                    )
+
+                    log.info(
+                        "%s Published TaskArtifactUpdateEvent for '%s' to %s",
+                        log_id,
+                        filename,
+                        artifact_topic,
+                    )
+                else:
+                    log.warning(
+                        "%s Failed to translate artifact '%s' v%d to A2A FilePart.",
+                        log_id,
+                        filename,
+                        version,
+                    )
+
+            except Exception as e:
+                log.exception(
+                    "%s Error processing artifact signal item %s from context: %s",
+                    log_id,
+                    item,
+                    e,
+                )
+
+    def _format_final_task_status(self, last_event: ADKEvent) -> TaskStatus:
+        """Helper to format the final TaskStatus based on the last ADK event."""
+        log.debug(
+            "%s Formatting final task status from last ADK event %s",
+            self.log_identifier,
+            last_event.id,
+        )
+        a2a_state = TaskState.COMPLETED
+        a2a_parts = []
+
+        if last_event.content and last_event.content.parts:
+            for part in last_event.content.parts:
+                if part.text:
+                    a2a_parts.append(TextPart(text=part.text))
+                elif part.function_response:
+                    try:
+                        response_data = part.function_response.response
+                        if isinstance(response_data, dict):
+                            a2a_parts.append(
+                                DataPart(
+                                    data=response_data,
+                                    metadata={"tool_name": part.function_response.name},
+                                )
+                            )
+                        else:
+                            a2a_parts.append(
+                                TextPart(
+                                    text=f"Tool {part.function_response.name} result: {str(response_data)}"
+                                )
+                            )
+                    except Exception:
+                        a2a_parts.append(
+                            TextPart(
+                                text=f"[Tool {part.function_response.name} result omitted]"
+                            )
+                        )
+
+        elif last_event.actions:
+            if last_event.actions.requested_auth_configs:
+                a2a_state = TaskState.INPUT_REQUIRED
+                a2a_parts.append(TextPart(text="[Agent requires input/authentication]"))
+
+        if not a2a_parts:
+            a2a_parts.append(TextPart(text=""))
+
+        a2a_message = A2AMessage(role="agent", parts=a2a_parts)
+        return TaskStatus(state=a2a_state, message=a2a_message)
+
+    async def finalize_task_success(self, a2a_context: Dict):
+        """
+        Finalizes a task successfully. Fetches final state, publishes final A2A response,
+        and ACKs the original message.
+        For RUN_BASED tasks, it uses the aggregated response buffer.
+        For STREAMING tasks, it uses the content of the last ADK event.
+        """
+        logical_task_id = a2a_context.get("logical_task_id")
+        original_message: Optional[SolaceMessage] = a2a_context.get(
+            "original_solace_message"
+        )
+        log.info(
+            "%s Finalizing task %s successfully.", self.log_identifier, logical_task_id
+        )
+        try:
+            session_id_to_retrieve = a2a_context.get(
+                "effective_session_id", a2a_context.get("session_id")
+            )
+            original_session_id = a2a_context.get("session_id")
+            user_id = a2a_context.get("user_id")
+            client_id = a2a_context.get("client_id")
+            jsonrpc_request_id = a2a_context.get("jsonrpc_request_id")
+            peer_reply_topic = a2a_context.get("replyToTopic")
+            namespace = self.get_config("namespace")
+            agent_name = self.get_config("agent_name")
+            is_run_based_session = a2a_context.get("is_run_based_session", False)
+
+            final_status: TaskStatus
+
+            with self.active_tasks_lock:
+                task_context = self.active_tasks.get(logical_task_id)
+
+            final_adk_session = await self.session_service.get_session(
+                app_name=agent_name,
+                user_id=user_id,
+                session_id=session_id_to_retrieve,
+            )
+            if not final_adk_session:
+                raise RuntimeError(
+                    f"Could not retrieve final session state for {session_id_to_retrieve}"
+                )
+
+            last_event = (
+                final_adk_session.events[-1] if final_adk_session.events else None
+            )
+
+            if is_run_based_session:
+                aggregated_text = ""
+                if task_context:
+                    aggregated_text = task_context.run_based_response_buffer
+                    log.info(
+                        "%s Using aggregated response buffer for RUN_BASED task %s (length: %d bytes).",
+                        self.log_identifier,
+                        logical_task_id,
+                        len(aggregated_text.encode("utf-8")),
+                    )
+
+                final_a2a_parts = []
+                if aggregated_text:
+                    final_a2a_parts.append(TextPart(text=aggregated_text))
+
+                if last_event and last_event.content and last_event.content.parts:
+                    for part in last_event.content.parts:
+                        if part.text is None:
+                            if part.function_response:
+                                try:
+                                    response_data = part.function_response.response
+                                    if isinstance(response_data, dict):
+                                        final_a2a_parts.append(
+                                            DataPart(
+                                                data=response_data,
+                                                metadata={
+                                                    "tool_name": part.function_response.name
+                                                },
+                                            )
+                                        )
+                                    else:
+                                        final_a2a_parts.append(
+                                            TextPart(
+                                                text=f"Tool {part.function_response.name} result: {str(response_data)}"
+                                            )
+                                        )
+                                except Exception:
+                                    final_a2a_parts.append(
+                                        TextPart(
+                                            text=f"[Tool {part.function_response.name} result omitted]"
+                                        )
+                                    )
+
+                if not final_a2a_parts:
+                    final_a2a_parts.append(TextPart(text=""))
+
+                final_status = TaskStatus(
+                    state=TaskState.COMPLETED,
+                    message=A2AMessage(role="agent", parts=final_a2a_parts),
+                )
+            else:
+                if last_event:
+                    final_status = self._format_final_task_status(last_event)
+                else:
+                    final_status = TaskStatus(
+                        state=TaskState.COMPLETED,
+                        message=A2AMessage(
+                            role="agent", parts=[TextPart(text="Task completed.")]
+                        ),
+                    )
+
+            final_a2a_artifacts: List[A2AArtifact] = []
+            log.debug(
+                "%s Final artifact bundling is removed. Artifacts sent via TaskArtifactUpdateEvent.",
+                self.log_identifier,
+            )
+
+            final_task_metadata = {"agent_name": agent_name}
+            if task_context and task_context.produced_artifacts:
+                final_task_metadata["produced_artifacts"] = (
+                    task_context.produced_artifacts
+                )
+                log.info(
+                    "%s Attaching manifest of %d produced artifacts to final task metadata.",
+                    self.log_identifier,
+                    len(task_context.produced_artifacts),
+                )
+
+            final_task = Task(
+                id=logical_task_id,
+                sessionId=original_session_id,
+                status=final_status,
+                artifacts=(final_a2a_artifacts if final_a2a_artifacts else None),
+                metadata=final_task_metadata,
+            )
+            final_response = JSONRPCResponse(id=jsonrpc_request_id, result=final_task)
+            a2a_payload = final_response.model_dump(exclude_none=True)
+            target_topic = peer_reply_topic or get_client_response_topic(
+                namespace, client_id
+            )
+
+            self._publish_a2a_event(a2a_payload, target_topic, a2a_context)
+            log.info(
+                "%s Published final successful response for task %s to %s (Artifacts NOT bundled).",
+                self.log_identifier,
+                logical_task_id,
+                target_topic,
+            )
+            if original_message:
+                try:
+                    original_message.call_acknowledgements()
+                    log.info(
+                        "%s Called ACK for original message of task %s.",
+                        self.log_identifier,
+                        logical_task_id,
+                    )
+                except Exception as ack_e:
+                    log.error(
+                        "%s Failed to call ACK for task %s: %s",
+                        self.log_identifier,
+                        logical_task_id,
+                        ack_e,
+                    )
+            else:
+                log.warning(
+                    "%s Original Solace message not found in context for task %s. Cannot ACK.",
+                    self.log_identifier,
+                    logical_task_id,
+                )
+
+        except Exception as e:
+            log.exception(
+                "%s Error during successful finalization of task %s: %s",
+                self.log_identifier,
+                logical_task_id,
+                e,
+            )
+            if original_message:
+                try:
+                    original_message.call_negative_acknowledgements()
+                    log.warning(
+                        "%s Called NACK for original message of task %s due to finalization error.",
+                        self.log_identifier,
+                        logical_task_id,
+                    )
+                except Exception as nack_e:
+                    log.error(
+                        "%s Failed to call NACK for task %s after finalization error: %s",
+                        self.log_identifier,
+                        logical_task_id,
+                        nack_e,
+                    )
+            else:
+                log.warning(
+                    "%s Original Solace message not found in context for task %s during finalization error. Cannot NACK.",
+                    self.log_identifier,
+                    logical_task_id,
+                )
+
+            try:
+                jsonrpc_request_id = a2a_context.get("jsonrpc_request_id")
+                client_id = a2a_context.get("client_id")
+                peer_reply_topic = a2a_context.get("replyToTopic")
+                namespace = self.get_config("namespace")
+                error_response = JSONRPCResponse(
+                    id=jsonrpc_request_id,
+                    error=InternalError(
+                        message=f"Failed to finalize successful task: {e}",
+                        data={"taskId": logical_task_id},
+                    ),
+                )
+                target_topic = peer_reply_topic or get_client_response_topic(
+                    namespace, client_id
+                )
+                self._publish_a2a_message(
+                    error_response.model_dump(exclude_none=True), target_topic
+                )
+            except Exception as report_err:
+                log.error(
+                    "%s Failed to report finalization error for task %s: %s",
+                    self.log_identifier,
+                    logical_task_id,
+                    report_err,
+                )
+
+    def finalize_task_canceled(self, a2a_context: Dict):
+        """
+        Finalizes a task as CANCELED. Publishes A2A Task response with CANCELED state
+        and ACKs the original message if available.
+        Called by the background ADK thread wrapper when a task is cancelled.
+        """
+        logical_task_id = a2a_context.get("logical_task_id")
+        original_message: Optional[SolaceMessage] = a2a_context.get(
+            "original_solace_message"
+        )
+        log.info(
+            "%s Finalizing task %s as CANCELED.", self.log_identifier, logical_task_id
+        )
+        try:
+            jsonrpc_request_id = a2a_context.get("jsonrpc_request_id")
+            client_id = a2a_context.get("client_id")
+            peer_reply_topic = a2a_context.get("replyToTopic")
+            namespace = self.get_config("namespace")
+
+            canceled_status = TaskStatus(
+                state=TaskState.CANCELED,
+                message=A2AMessage(
+                    role="agent",
+                    parts=[TextPart(text="Task cancelled by request.")],
+                ),
+            )
+            agent_name = self.get_config("agent_name")
+            final_task = Task(
+                id=logical_task_id,
+                sessionId=a2a_context.get("session_id"),
+                status=canceled_status,
+                metadata={"agent_name": agent_name},
+            )
+            final_response = JSONRPCResponse(id=jsonrpc_request_id, result=final_task)
+            a2a_payload = final_response.model_dump(exclude_none=True)
+            target_topic = peer_reply_topic or get_client_response_topic(
+                namespace, client_id
+            )
+
+            self._publish_a2a_event(a2a_payload, target_topic, a2a_context)
+            log.info(
+                "%s Published final CANCELED response for task %s to %s.",
+                self.log_identifier,
+                logical_task_id,
+                target_topic,
+            )
+
+            if original_message:
+                try:
+                    original_message.call_acknowledgements()
+                    log.info(
+                        "%s Called ACK for original message of cancelled task %s.",
+                        self.log_identifier,
+                        logical_task_id,
+                    )
+                except Exception as ack_e:
+                    log.error(
+                        "%s Failed to call ACK for cancelled task %s: %s",
+                        self.log_identifier,
+                        logical_task_id,
+                        ack_e,
+                    )
+            else:
+                log.warning(
+                    "%s Original Solace message not found in context for cancelled task %s. Cannot ACK.",
+                    self.log_identifier,
+                    logical_task_id,
+                )
+
+        except Exception as e:
+            log.exception(
+                "%s Error during CANCELED finalization of task %s: %s",
+                self.log_identifier,
+                logical_task_id,
+                e,
+            )
+            if original_message:
+                try:
+                    original_message.call_negative_acknowledgements()
+                except Exception:
+                    pass
+
+    async def _publish_tool_failure_status(
+        self, exception: Exception, a2a_context: Dict
+    ):
+        """
+        Publishes an intermediate status update indicating a tool execution has failed.
+        This method will flush the buffer before publishing to maintain proper message ordering.
+        """
+        logical_task_id = a2a_context.get("logical_task_id")
+        log_identifier_helper = (
+            f"{self.log_identifier}[ToolFailureStatus:{logical_task_id}]"
+        )
+        try:
+            # Create the status update event
+            tool_error_data_part = DataPart(
+                data={
+                    "a2a_signal_type": "tool_execution_error",
+                    "error_message": str(exception),
+                    "details": "An unhandled exception occurred during tool execution.",
+                }
+            )
+
+            status_message = A2AMessage(role="agent", parts=[tool_error_data_part])
+            intermediate_status = TaskStatus(
+                state=TaskState.WORKING,
+                message=status_message,
+                timestamp=datetime.now(timezone.utc),
+            )
+
+            status_update_event = TaskStatusUpdateEvent(
+                id=logical_task_id,
+                status=intermediate_status,
+                final=False,
+                metadata={"agent_name": self.get_config("agent_name")},
+            )
+
+            await self._publish_status_update_with_buffer_flush(
+                status_update_event,
+                a2a_context,
+                skip_buffer_flush=False,
+            )
+
+            log.debug(
+                "%s Published tool failure status update.",
+                log_identifier_helper,
+            )
+
+        except Exception as e:
+            log.error(
+                "%s Failed to publish intermediate tool failure status: %s",
+                log_identifier_helper,
+                e,
+            )
+
+    async def _repair_session_history_on_error(
+        self, exception: Exception, a2a_context: Dict
+    ):
+        """
+        Reactively repairs the session history if the last event was a tool call.
+        This is "the belt" in the belt-and-suspenders strategy.
+        """
+        log_identifier = f"{self.log_identifier}[HistoryRepair]"
+        try:
+            from ...agent.adk.callbacks import create_dangling_tool_call_repair_content
+
+            session_id = a2a_context.get("effective_session_id")
+            user_id = a2a_context.get("user_id")
+            agent_name = self.get_config("agent_name")
+
+            if not all([session_id, user_id, agent_name, self.session_service]):
+                log.warning(
+                    "%s Skipping history repair due to missing context.", log_identifier
+                )
+                return
+
+            session = await self.session_service.get_session(
+                app_name=agent_name, user_id=user_id, session_id=session_id
+            )
+
+            if not session or not session.events:
+                log.debug(
+                    "%s No session or events found for history repair.", log_identifier
+                )
+                return
+
+            last_event = session.events[-1]
+            function_calls = last_event.get_function_calls()
+
+            if not function_calls:
+                log.debug(
+                    "%s Last event was not a function call. No repair needed.",
+                    log_identifier,
+                )
+                return
+
+            log.info(
+                "%s Last event contained function_call(s). Repairing session history.",
+                log_identifier,
+            )
+
+            repair_content = create_dangling_tool_call_repair_content(
+                dangling_calls=function_calls,
+                error_message=f"Tool execution failed with an unhandled exception: {str(exception)}",
+            )
+
+            repair_event = ADKEvent(
+                invocation_id=last_event.invocation_id,
+                author=agent_name,
+                content=repair_content,
+            )
+
+            await self.session_service.append_event(session=session, event=repair_event)
+            log.info(
+                "%s Session history repaired successfully with an error function_response.",
+                log_identifier,
+            )
+
+        except Exception as e:
+            log.exception(
+                "%s Critical error during session history repair: %s", log_identifier, e
+            )
+
+    def finalize_task_limit_reached(
+        self, a2a_context: Dict, exception: LlmCallsLimitExceededError
+    ):
+        """
+        Finalizes a task when the LLM call limit is reached, prompting the user to continue.
+        Sends a COMPLETED status with an informative message.
+        """
+        logical_task_id = a2a_context.get("logical_task_id")
+        original_message: Optional[SolaceMessage] = a2a_context.get(
+            "original_solace_message"
+        )
+        log.info(
+            "%s Finalizing task %s as COMPLETED (LLM call limit reached).",
+            self.log_identifier,
+            logical_task_id,
+        )
+        try:
+            jsonrpc_request_id = a2a_context.get("jsonrpc_request_id")
+            client_id = a2a_context.get("client_id")
+            peer_reply_topic = a2a_context.get("replyToTopic")
+            namespace = self.get_config("namespace")
+            agent_name = self.get_config("agent_name")
+            original_session_id = a2a_context.get("session_id")
+
+            limit_message_text = (
+                f"This interaction has reached its processing limit. "
+                "If you'd like to continue this conversation, please type 'continue'. "
+                "Otherwise, you can start a new topic."
+            )
+
+            error_payload = InternalError(
+                message=limit_message_text,
+                data={"taskId": logical_task_id, "reason": "llm_call_limit_reached"},
+            )
+
+            final_response = JSONRPCResponse(id=jsonrpc_request_id, error=error_payload)
+            a2a_payload = final_response.model_dump(exclude_none=True)
+
+            target_topic = peer_reply_topic or get_client_response_topic(
+                namespace, client_id
+            )
+
+            self._publish_a2a_event(a2a_payload, target_topic, a2a_context)
+            log.info(
+                "%s Published ERROR response for task %s to %s (LLM limit reached, user guided to continue).",
+                self.log_identifier,
+                logical_task_id,
+                target_topic,
+            )
+
+            if original_message:
+                try:
+                    original_message.call_acknowledgements()
+                    log.info(
+                        "%s Called ACK for original message of task %s (LLM limit reached).",
+                        self.log_identifier,
+                        logical_task_id,
+                    )
+                except Exception as ack_e:
+                    log.error(
+                        "%s Failed to call ACK for task %s (LLM limit reached): %s",
+                        self.log_identifier,
+                        logical_task_id,
+                        ack_e,
+                    )
+            else:
+                log.warning(
+                    "%s Original Solace message not found in context for task %s (LLM limit reached). Cannot ACK.",
+                    self.log_identifier,
+                    logical_task_id,
+                )
+
+        except Exception as e:
+            log.exception(
+                "%s Error during COMPLETED (LLM limit) finalization of task %s: %s",
+                self.log_identifier,
+                logical_task_id,
+                e,
+            )
+            self.finalize_task_error(e, a2a_context)
+
+    async def finalize_task_error(self, exception: Exception, a2a_context: Dict):
+        """
+        Finalizes a task with an error. Publishes a final A2A Task with a FAILED
+        status and NACKs the original message.
+        Called by the background ADK thread wrapper.
+        """
+        logical_task_id = a2a_context.get("logical_task_id")
+        original_message: Optional[SolaceMessage] = a2a_context.get(
+            "original_solace_message"
+        )
+        log.error(
+            "%s Finalizing task %s with error: %s",
+            self.log_identifier,
+            logical_task_id,
+            exception,
+        )
+        try:
+            await self._repair_session_history_on_error(exception, a2a_context)
+
+            await self._publish_tool_failure_status(exception, a2a_context)
+
+            client_id = a2a_context.get("client_id")
+            jsonrpc_request_id = a2a_context.get("jsonrpc_request_id")
+            peer_reply_topic = a2a_context.get("replyToTopic")
+            namespace = self.get_config("namespace")
+
+            failed_status = TaskStatus(
+                state=TaskState.FAILED,
+                message=A2AMessage(
+                    role="agent",
+                    parts=[
+                        TextPart(
+                            text="An unexpected error occurred during tool execution. Please try your request again. If the problem persists, contact an administrator."
+                        )
+                    ],
+                ),
+            )
+
+            final_task = Task(
+                id=logical_task_id,
+                sessionId=a2a_context.get("session_id"),
+                status=failed_status,
+                metadata={"agent_name": self.get_config("agent_name")},
+            )
+
+            final_response = JSONRPCResponse(id=jsonrpc_request_id, result=final_task)
+            a2a_payload = final_response.model_dump(exclude_none=True)
+            target_topic = peer_reply_topic or get_client_response_topic(
+                namespace, client_id
+            )
+
+            self._publish_a2a_event(a2a_payload, target_topic, a2a_context)
+            log.info(
+                "%s Published final FAILED Task response for task %s to %s",
+                self.log_identifier,
+                logical_task_id,
+                target_topic,
+            )
+
+            if original_message:
+                try:
+                    original_message.call_negative_acknowledgements()
+                    log.info(
+                        "%s Called NACK for original message of failed task %s.",
+                        self.log_identifier,
+                        logical_task_id,
+                    )
+                except Exception as nack_e:
+                    log.error(
+                        "%s Failed to call NACK for failed task %s: %s",
+                        self.log_identifier,
+                        logical_task_id,
+                        nack_e,
+                    )
+            else:
+                log.warning(
+                    "%s Original Solace message not found in context for failed task %s. Cannot NACK.",
+                    self.log_identifier,
+                    logical_task_id,
+                )
+
+        except Exception as e:
+            log.exception(
+                "%s Error during error finalization of task %s: %s",
+                self.log_identifier,
+                logical_task_id,
+                e,
+            )
+            if original_message:
+                try:
+                    original_message.call_negative_acknowledgements()
+                    log.warning(
+                        "%s Called NACK for task %s during error finalization fallback.",
+                        self.log_identifier,
+                        logical_task_id,
+                    )
+                except Exception as nack_e:
+                    log.error(
+                        "%s Failed to call NACK for task %s during error finalization fallback: %s",
+                        self.log_identifier,
+                        logical_task_id,
+                        nack_e,
+                    )
+            else:
+                log.warning(
+                    "%s Original Solace message not found for task %s during error finalization fallback. Cannot NACK.",
+                    self.log_identifier,
+                    logical_task_id,
+                )
+
+    async def finalize_task_with_cleanup(
+        self, a2a_context: Dict, is_paused: bool, exception: Optional[Exception] = None
+    ):
+        """
+        Centralized async method to finalize a task and perform all necessary cleanup.
+        This is scheduled on the component's event loop to ensure it runs after
+        any pending status updates.
+
+        Args:
+            a2a_context: The context dictionary for the task.
+            is_paused: Boolean indicating if the task is paused for a long-running tool.
+            exception: The exception that occurred, if any.
+        """
+        logical_task_id = a2a_context.get("logical_task_id", "unknown_task")
+        log_id = f"{self.log_identifier}[FinalizeTask:{logical_task_id}]"
+        log.info(
+            "%s Starting finalization and cleanup. Paused: %s, Exception: %s",
+            log_id,
+            is_paused,
+            type(exception).__name__ if exception else "None",
+        )
+
+        try:
+            if is_paused:
+                log.info(
+                    "%s Task is paused for a long-running tool. Skipping finalization logic.",
+                    log_id,
+                )
+            else:
+                try:
+                    if exception:
+                        if isinstance(exception, TaskCancelledError):
+                            self.finalize_task_canceled(a2a_context)
+                        elif isinstance(exception, LlmCallsLimitExceededError):
+                            self.finalize_task_limit_reached(a2a_context, exception)
+                        else:
+                            await self.finalize_task_error(exception, a2a_context)
+                    else:
+                        await self.finalize_task_success(a2a_context)
+                except Exception as e:
+                    log.exception(
+                        "%s An unexpected error occurred during the finalization logic itself: %s",
+                        log_id,
+                        e,
+                    )
+                    original_message: Optional[SolaceMessage] = a2a_context.get(
+                        "original_solace_message"
+                    )
+                    if original_message:
+                        try:
+                            original_message.call_negative_acknowledgements()
+                        except Exception as nack_err:
+                            log.error(
+                                "%s Fallback NACK failed during finalization error: %s",
+                                log_id,
+                                nack_err,
+                            )
+        finally:
+            if not is_paused:
+                # Cleanup for RUN_BASED sessions remains, as it's a service-level concern
+                if a2a_context.get("is_run_based_session"):
+                    temp_session_id_to_delete = a2a_context.get(
+                        "temporary_run_session_id_for_cleanup"
+                    )
+                    agent_name_for_session = a2a_context.get("agent_name_for_session")
+                    user_id_for_session = a2a_context.get("user_id_for_session")
+
+                    if (
+                        temp_session_id_to_delete
+                        and agent_name_for_session
+                        and user_id_for_session
+                    ):
+                        log.info(
+                            "%s Cleaning up RUN_BASED session (app: %s, user: %s, id: %s) from shared service for task_id='%s'",
+                            log_id,
+                            agent_name_for_session,
+                            user_id_for_session,
+                            temp_session_id_to_delete,
+                            logical_task_id,
+                        )
+                        try:
+                            if self.session_service:
+                                await self.session_service.delete_session(
+                                    app_name=agent_name_for_session,
+                                    user_id=user_id_for_session,
+                                    session_id=temp_session_id_to_delete,
+                                )
+                            else:
+                                log.error(
+                                    "%s self.session_service is None, cannot delete RUN_BASED session %s.",
+                                    log_id,
+                                    temp_session_id_to_delete,
+                                )
+                        except AttributeError:
+                            log.error(
+                                "%s self.session_service does not support 'delete_session'. Cleanup for RUN_BASED session (app: %s, user: %s, id: %s) skipped.",
+                                log_id,
+                                agent_name_for_session,
+                                user_id_for_session,
+                                temp_session_id_to_delete,
+                            )
+                        except Exception as e_cleanup:
+                            log.error(
+                                "%s Error cleaning up RUN_BASED session (app: %s, user: %s, id: %s) from shared service: %s",
+                                log_id,
+                                agent_name_for_session,
+                                user_id_for_session,
+                                temp_session_id_to_delete,
+                                e_cleanup,
+                                exc_info=True,
+                            )
+                    else:
+                        log.warning(
+                            "%s Could not clean up RUN_BASED session for task %s due to missing context (id_to_delete: %s, agent_name: %s, user_id: %s).",
+                            log_id,
+                            logical_task_id,
+                            temp_session_id_to_delete,
+                            agent_name_for_session,
+                            user_id_for_session,
+                        )
+
+                with self.active_tasks_lock:
+                    removed_task_context = self.active_tasks.pop(logical_task_id, None)
+                    if removed_task_context:
+                        log.debug(
+                            "%s Removed TaskExecutionContext for task %s.",
+                            log_id,
+                            logical_task_id,
+                        )
+                    else:
+                        log.warning(
+                            "%s TaskExecutionContext for task %s was already removed.",
+                            log_id,
+                            logical_task_id,
+                        )
+            else:
+                log.info(
+                    "%s Task %s is paused for a long-running tool. Skipping all cleanup.",
+                    log_id,
+                    logical_task_id,
+                )
+
+            log.info(
+                "%s Finalization and cleanup complete for task %s.",
+                log_id,
+                logical_task_id,
+            )
+
+    def _resolve_instruction_provider(
+        self, config_value: Any
+    ) -> Union[str, InstructionProvider]:
+        """Resolves instruction config using helper."""
+        return resolve_instruction_provider(self, config_value)
+
+    def _get_a2a_base_topic(self) -> str:
+        """Returns the base topic prefix using helper."""
+        return get_a2a_base_topic(self.namespace)
+
+    def _get_discovery_topic(self) -> str:
+        """Returns the discovery topic using helper."""
+        return get_discovery_topic(self.namespace)
+
+    def _get_agent_request_topic(self, agent_id: str) -> str:
+        """Returns the agent request topic using helper."""
+        return get_agent_request_topic(self.namespace, agent_id)
+
+    def _get_agent_response_topic(
+        self, delegating_agent_name: str, sub_task_id: str
+    ) -> str:
+        """Returns the agent response topic using helper."""
+        return get_agent_response_topic(
+            self.namespace, delegating_agent_name, sub_task_id
+        )
+
+    def _get_peer_agent_status_topic(
+        self, delegating_agent_name: str, sub_task_id: str
+    ) -> str:
+        """Returns the peer agent status topic using helper."""
+        return get_peer_agent_status_topic(
+            self.namespace, delegating_agent_name, sub_task_id
+        )
+
+    def _get_client_response_topic(self, client_id: str) -> str:
+        """Returns the client response topic using helper."""
+        return get_client_response_topic(self.namespace, client_id)
+
+    def _publish_a2a_message(
+        self, payload: Dict, topic: str, user_properties: Optional[Dict] = None
+    ):
+        """Helper to publish A2A messages via the SAC App."""
+        try:
+            app = self.get_app()
+            if app:
+                if self.invocation_monitor:
+                    self.invocation_monitor.log_message_event(
+                        direction="PUBLISHED",
+                        topic=topic,
+                        payload=payload,
+                        component_identifier=self.log_identifier,
+                    )
+                app.send_message(
+                    payload=payload, topic=topic, user_properties=user_properties
+                )
+            else:
+                log.error(
+                    "%s Cannot publish message: Not running within a SAC App context.",
+                    self.log_identifier,
+                )
+        except Exception as e:
+            log.exception(
+                "%s Failed to publish A2A message to topic %s: %s",
+                self.log_identifier,
+                topic,
+                e,
+            )
+            raise
+
+    def _publish_a2a_event(self, payload: Dict, topic: str, a2a_context: Dict):
+        """
+        Centralized helper to publish an A2A event, ensuring user properties
+        are consistently attached from the a2a_context.
+        """
+        user_properties = {}
+        if a2a_context.get("a2a_user_config"):
+            user_properties["a2aUserConfig"] = a2a_context["a2a_user_config"]
+
+        self._publish_a2a_message(payload, topic, user_properties)
+
+    def submit_a2a_task(
+        self,
+        target_agent_name: str,
+        a2a_message: A2AMessage,
+        original_session_id: str,
+        main_logical_task_id: str,
+        user_id: str,
+        user_config: Dict[str, Any],
+        sub_task_id: str,
+        function_call_id: Optional[str] = None,
+    ) -> str:
+        """
+        Submits a task to a peer agent in a non-blocking way.
+        Returns the sub_task_id for correlation.
+        """
+        log_identifier_helper = (
+            f"{self.log_identifier}[SubmitA2ATask:{target_agent_name}]"
+        )
+        log.debug(
+            "%s Submitting non-blocking task for main task %s",
+            log_identifier_helper,
+            main_logical_task_id,
+        )
+
+        peer_request_topic = self._get_agent_request_topic(target_agent_name)
+
+        a2a_request_params = {
+            "id": sub_task_id,
+            "sessionId": original_session_id,
+            "message": a2a_message.model_dump(exclude_none=True),
+            "metadata": {
+                "sessionBehavior": "RUN_BASED",
+                "parentTaskId": main_logical_task_id,
+                "function_call_id": function_call_id,
+            },
+        }
+        a2a_request = SendTaskRequest(params=a2a_request_params)
+
+        delegating_agent_name = self.get_config("agent_name")
+        reply_to_topic = self._get_agent_response_topic(
+            delegating_agent_name=delegating_agent_name,
+            sub_task_id=sub_task_id,
+        )
+        status_topic = self._get_peer_agent_status_topic(
+            delegating_agent_name=delegating_agent_name,
+            sub_task_id=sub_task_id,
+        )
+
+        user_properties = {
+            "replyTo": reply_to_topic,
+            "a2aStatusTopic": status_topic,
+            "userId": user_id,
+        }
+        if isinstance(user_config, dict):
+            user_properties["a2aUserConfig"] = user_config
+
+        self._publish_a2a_message(
+            payload=a2a_request.model_dump(exclude_none=True),
+            topic=peer_request_topic,
+            user_properties=user_properties,
+        )
+        log.info(
+            "%s Published delegation request to %s (Sub-Task ID: %s, ReplyTo: %s, StatusTo: %s)",
+            log_identifier_helper,
+            peer_request_topic,
+            sub_task_id,
+            reply_to_topic,
+            status_topic,
+        )
+
+        return sub_task_id
+
+    def _handle_scheduled_task_completion(
+        self, future: concurrent.futures.Future, event_type_for_log: EventType
+    ):
+        """Callback to handle completion of futures from run_coroutine_threadsafe."""
+        try:
+            if future.cancelled():
+                log.warning(
+                    "%s Coroutine for event type %s (scheduled via run_coroutine_threadsafe) was cancelled.",
+                    self.log_identifier,
+                    event_type_for_log,
+                )
+            elif future.done() and future.exception() is not None:
+                exception = future.exception()
+                log.error(
+                    "%s Coroutine for event type %s (scheduled via run_coroutine_threadsafe) failed with exception: %s",
+                    self.log_identifier,
+                    event_type_for_log,
+                    exception,
+                    exc_info=exception,
+                )
+            else:
+                pass
+        except Exception as e:
+            log.error(
+                "%s Error during _handle_scheduled_task_completion (for run_coroutine_threadsafe future) for event type %s: %s",
+                self.log_identifier,
+                event_type_for_log,
+                e,
+                exc_info=e,
+            )
+
+    def _start_async_loop(self):
+        """Target method for the dedicated async thread."""
+        log.info("%s Dedicated async thread started.", self.log_identifier)
+        try:
+            asyncio.set_event_loop(self._async_loop)
+            self._async_loop.run_forever()
+        except Exception as e:
+            log.exception(
+                "%s Exception in dedicated async thread loop: %s",
+                self.log_identifier,
+                e,
+            )
+            if self._async_init_future and not self._async_init_future.done():
+                self._async_init_future.set_exception(e)
+        finally:
+            log.info("%s Dedicated async thread loop finishing.", self.log_identifier)
+            if self._async_loop.is_running():
+                self._async_loop.call_soon_threadsafe(self._async_loop.stop)
+
+    async def _perform_async_init(self):
+        """Coroutine executed on the dedicated loop to perform async initialization."""
+        try:
+            log.info(
+                "%s Loading tools asynchronously in dedicated thread...",
+                self.log_identifier,
+            )
+            loaded_tools, enabled_builtin_tools = await load_adk_tools(self)
+            log.info(
+                "%s Initializing ADK Agent/Runner asynchronously in dedicated thread...",
+                self.log_identifier,
+            )
+            self.adk_agent = initialize_adk_agent(
+                self, loaded_tools, enabled_builtin_tools
+            )
+            self.runner = initialize_adk_runner(self)
+
+            log.info("%s Populating agent card tool manifest...", self.log_identifier)
+            tool_manifest = []
+            for tool in loaded_tools:
+                if isinstance(tool, MCPToolset):
+                    try:
+                        log.debug(
+                            "%s Retrieving tools from MCPToolset for Agent %s...",
+                            self.log_identifier,
+                            self.agent_name,
+                        )
+                        mcp_tools = await tool.get_tools()
+                    except Exception as e:
+                        log.error(
+                            "%s Error retrieving tools from MCPToolset for Agent Card %s: %s",
+                            self.log_identifier,
+                            self.agent_name,
+                            e,
+                        )
+                        continue
+                    for mcp_tool in mcp_tools:
+                        tool_manifest.append(
+                            {
+                                "id": mcp_tool.name,
+                                "name": mcp_tool.name,
+                                "description": mcp_tool.description
+                                or "No description available.",
+                            }
+                        )
+                else:
+                    tool_name = getattr(tool, "name", getattr(tool, "__name__", None))
+                    if tool_name is not None:
+                        tool_manifest.append(
+                            {
+                                "id": tool_name,
+                                "name": tool_name,
+                                "description": getattr(
+                                    tool, "description", getattr(tool, "__doc__", None)
+                                )
+                                or "No description available.",
+                            }
+                        )
+
+            self.agent_card_tool_manifest = tool_manifest
+            log.info(
+                "%s Agent card tool manifest populated with %d tools.",
+                self.log_identifier,
+                len(self.agent_card_tool_manifest),
+            )
+
+            log.info(
+                "%s Async initialization steps complete in dedicated thread.",
+                self.log_identifier,
+            )
+            if self._async_init_future and not self._async_init_future.done():
+                log.info(
+                    "%s _perform_async_init: Signaling success to main thread.",
+                    self.log_identifier,
+                )
+                self._async_loop.call_soon_threadsafe(
+                    self._async_init_future.set_result, True
+                )
+            else:
+                log.warning(
+                    "%s _perform_async_init: _async_init_future is None or already done before signaling success.",
+                    self.log_identifier,
+                )
+        except Exception as e:
+            log.exception(
+                "%s _perform_async_init: Error during async initialization in dedicated thread: %s",
+                self.log_identifier,
+                e,
+            )
+            if self._async_init_future and not self._async_init_future.done():
+                log.error(
+                    "%s _perform_async_init: Signaling failure to main thread.",
+                    self.log_identifier,
+                )
+                self._async_loop.call_soon_threadsafe(
+                    self._async_init_future.set_exception, e
+                )
+            else:
+                log.warning(
+                    "%s _perform_async_init: _async_init_future is None or already done before signaling failure.",
+                    self.log_identifier,
+                )
+
+    def cleanup(self):
+        """Clean up resources on component shutdown."""
+        log.info("%s Cleaning up A2A ADK Host Component.", self.log_identifier)
+        self.cancel_timer(self._card_publish_timer_id)
+
+        cleanup_func_details = self.get_config("agent_cleanup_function")
+        if cleanup_func_details and isinstance(cleanup_func_details, dict):
+            module_name = cleanup_func_details.get("module")
+            func_name = cleanup_func_details.get("name")
+            base_path = cleanup_func_details.get("base_path")
+
+            if module_name and func_name:
+                log.info(
+                    "%s Attempting to load and execute cleanup_function: %s.%s",
+                    self.log_identifier,
+                    module_name,
+                    func_name,
+                )
+                try:
+                    module = import_module(module_name, base_path=base_path)
+                    cleanup_function = getattr(module, func_name)
+
+                    if not callable(cleanup_function):
+                        log.error(
+                            "%s Cleanup function '%s' in module '%s' is not callable. Skipping.",
+                            self.log_identifier,
+                            func_name,
+                            module_name,
+                        )
+                    else:
+                        cleanup_function(self)
+                        log.info(
+                            "%s Successfully executed cleanup_function: %s.%s",
+                            self.log_identifier,
+                            module_name,
+                            func_name,
+                        )
+                except Exception as e:
+                    log.exception(
+                        "%s Error during agent cleanup via cleanup_function '%s.%s': %s",
+                        self.log_identifier,
+                        module_name,
+                        func_name,
+                        e,
+                    )
+        if self.invocation_monitor:
+            try:
+                self.invocation_monitor.cleanup()
+            except Exception as im_clean_e:
+                log.error(
+                    "%s Error during InvocationMonitor cleanup: %s",
+                    self.log_identifier,
+                    im_clean_e,
+                )
+
+        if self._async_loop and self._async_loop.is_running():
+            log.info(
+                "%s Performing async cleanup via dedicated thread...",
+                self.log_identifier,
+            )
+
+            async def _perform_async_cleanup():
+                log.debug("%s Entering async cleanup coroutine...", self.log_identifier)
+                pass
+
+            try:
+                cleanup_future = asyncio.run_coroutine_threadsafe(
+                    _perform_async_cleanup(), self._async_loop
+                )
+                cleanup_future.result(timeout=30)
+                log.info("%s Async cleanup completed.", self.log_identifier)
+            except Exception as e:
+                log.exception(
+                    "%s Error during async cleanup: %s", self.log_identifier, e
+                )
+            finally:
+                if self._async_loop and self._async_loop.is_running():
+                    log.info(
+                        "%s Cleanup: Stopping dedicated async loop...",
+                        self.log_identifier,
+                    )
+                    self._async_loop.call_soon_threadsafe(self._async_loop.stop)
+                else:
+                    log.info(
+                        "%s Cleanup: Dedicated async loop is None or not running, no need to stop.",
+                        self.log_identifier,
+                    )
+                if self._async_thread and self._async_thread.is_alive():
+                    log.info(
+                        "%s Cleanup: Joining dedicated async thread...",
+                        self.log_identifier,
+                    )
+                    self._async_thread.join(timeout=5)
+                    if self._async_thread.is_alive():
+                        log.warning(
+                            "%s Dedicated async thread did not exit cleanly.",
+                            self.log_identifier,
+                        )
+                log.info(
+                    "%s Dedicated async thread stopped and joined.", self.log_identifier
+                )
+        else:
+            log.info(
+                "%s Dedicated async loop not running, skipping async cleanup.",
+                self.log_identifier,
+            )
+
+        with self.active_tasks_lock:
+            if self._async_loop and self._async_loop.is_running():
+                for task_context in self.active_tasks.values():
+                    task_context.cancel()
+            self.active_tasks.clear()
+            log.debug("%s Cleared all active tasks.", self.log_identifier)
+
+        super().cleanup()
+        log.info("%s Component cleanup finished.", self.log_identifier)
+
+    def set_agent_specific_state(self, key: str, value: Any):
+        """
+        Sets a key-value pair in the agent-specific state.
+        Intended to be used by the custom init_function.
+        """
+        if not hasattr(self, "agent_specific_state"):
+            self.agent_specific_state = {}
+        self.agent_specific_state[key] = value
+        log.debug("%s Set agent_specific_state['%s']", self.log_identifier, key)
+
+    def get_agent_specific_state(self, key: str, default: Optional[Any] = None) -> Any:
+        """
+        Gets a value from the agent-specific state.
+        Intended to be used by tools and the custom cleanup_function.
+        """
+        if not hasattr(self, "agent_specific_state"):
+            return default
+        return self.agent_specific_state.get(key, default)
+
+    def get_async_loop(self) -> Optional[asyncio.AbstractEventLoop]:
+        """Returns the dedicated asyncio event loop for this component's async tasks."""
+        return self._async_loop
+
+    def set_agent_system_instruction_string(self, instruction_string: str) -> None:
+        """
+        Sets a static string to be injected into the LLM system prompt.
+        Called by the agent's init_function.
+        """
+        if not isinstance(instruction_string, str):
+            log.error(
+                "%s Invalid type for instruction_string: %s. Must be a string.",
+                self.log_identifier,
+                type(instruction_string),
+            )
+            return
+        self._agent_system_instruction_string = instruction_string
+        self._agent_system_instruction_callback = None
+        log.info("%s Static agent system instruction string set.", self.log_identifier)
+
+    def set_agent_system_instruction_callback(
+        self,
+        callback_function: Callable[[CallbackContext, LlmRequest], Optional[str]],
+    ) -> None:
+        """
+        Sets a callback function to dynamically generate system prompt injections.
+        Called by the agent's init_function.
+        """
+        if not callable(callback_function):
+            log.error(
+                "%s Invalid type for callback_function: %s. Must be callable.",
+                self.log_identifier,
+                type(callback_function),
+            )
+            return
+        self._agent_system_instruction_callback = callback_function
+        self._agent_system_instruction_string = None
+        log.info("%s Agent system instruction callback set.", self.log_identifier)
+
+    def get_gateway_id(self) -> str:
+        """
+        Returns a unique identifier for this specific gateway/host instance.
+        For now, using the agent name, but could be made more robust (e.g., hostname + agent name).
+        """
+        return self.agent_name
+
+    async def _resolve_early_embeds_and_handle_signals(
+        self, raw_text: str, a2a_context: Dict
+    ) -> Tuple[str, List[Tuple[int, Any]], str]:
+        """
+        Resolves early-stage embeds in raw text and extracts signals.
+        Returns the resolved text, a list of signals, and any unprocessed tail.
+        This is called by process_and_publish_adk_event.
+        """
+        logical_task_id = a2a_context.get("logical_task_id", "unknown_task")
+        method_context_log_identifier = (
+            f"{self.log_identifier}[ResolveEmbeds:{logical_task_id}]"
+        )
+        log.debug(
+            "%s Resolving early embeds for text (length: %d).",
+            method_context_log_identifier,
+            len(raw_text),
+        )
+
+        original_session_id = a2a_context.get("session_id")
+        user_id = a2a_context.get("user_id")
+        adk_app_name = self.get_config("agent_name")
+
+        if not all([self.artifact_service, original_session_id, user_id, adk_app_name]):
+            log.error(
+                "%s Missing necessary context for embed resolution (artifact_service, session_id, user_id, or adk_app_name). Skipping.",
+                method_context_log_identifier,
+            )
+            return (
+                raw_text,
+                [],
+                "",
+            )
+        context_for_embeds = {
+            "artifact_service": self.artifact_service,
+            "session_context": {
+                "app_name": adk_app_name,
+                "user_id": user_id,
+                "session_id": original_session_id,
+            },
+            "config": {
+                "gateway_max_artifact_resolve_size_bytes": self.get_config(
+                    "tool_output_llm_return_max_bytes", 4096
+                ),
+                "gateway_recursive_embed_depth": self.get_config(
+                    "gateway_recursive_embed_depth", 12
+                ),
+            },
+        }
+
+        resolver_config = context_for_embeds["config"]
+
+        try:
+            from ...common.utils.embeds.resolver import (
+                resolve_embeds_in_string,
+                evaluate_embed,
+            )
+            from ...common.utils.embeds.constants import EARLY_EMBED_TYPES
+
+            resolved_text, processed_until_index, signals_found = (
+                await resolve_embeds_in_string(
+                    text=raw_text,
+                    context=context_for_embeds,
+                    resolver_func=evaluate_embed,
+                    types_to_resolve=EARLY_EMBED_TYPES,
+                    log_identifier=method_context_log_identifier,
+                    config=resolver_config,
+                )
+            )
+            unprocessed_tail = raw_text[processed_until_index:]
+            log.debug(
+                "%s Embed resolution complete. Resolved text: '%s...', Signals found: %d, Unprocessed tail: '%s...'",
+                method_context_log_identifier,
+                resolved_text[:100],
+                len(signals_found),
+                unprocessed_tail[:100],
+            )
+            return resolved_text, signals_found, unprocessed_tail
+        except Exception as e:
+            log.exception(
+                "%s Error during embed resolution: %s", method_context_log_identifier, e
+            )
+            return raw_text, [], ""