fix: revert deferred tool call, add confirmToolCall support with ToolRunnableCallable [JAR-9208]

Author: JoshParkSJ

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "uipath-langchain"
-version = "0.9.24"
+version = "0.9.25"
 description = "Python SDK that enables developers to build and deploy LangGraph agents to the UiPath Cloud Platform"
 readme = { file = "README.md", content-type = "text/markdown" }
 requires-python = ">=3.11"

src/uipath_langchain/agent/tools/__init__.py

@@ -11,6 +11,7 @@
     create_tools_from_resources,
 )
 from .tool_node import (
+    ConversationalToolRunnableCallable,
     ToolWrapperMixin,
     UiPathToolNode,
     create_tool_node,
@@ -28,6 +29,7 @@
     "create_ixp_extraction_tool",
     "create_ixp_escalation_tool",
     "UiPathToolNode",
+    "ConversationalToolRunnableCallable",
     "ToolWrapperMixin",
     "wrap_tools_with_error_handling",
 ]

src/uipath_langchain/agent/tools/tool_node.py

@@ -22,7 +22,10 @@
     extract_current_tool_call_index,
     find_latest_ai_message,
 )
-from uipath_langchain.chat.hitl import request_conversational_tool_confirmation
+from uipath_langchain.chat.hitl import (
+    REQUIRE_CONVERSATIONAL_CONFIRMATION,
+    request_conversational_tool_confirmation,
+)
 
 # the type safety can be improved with generics
 ToolWrapperReturnType = dict[str, Any] | Command[Any] | None
@@ -274,9 +277,27 @@ async def _afunc(state: AgentGraphState) -> OutputType:
                 raise
             return result
 
+    tool = getattr(tool_node, "tool", None)
+
+    # Preserve tool ref so the runtime can discover which tools need confirmation
+    # (see runtime.py _get_tool_confirmation_info)
+    metadata = getattr(tool, "metadata", None) or {}
+    if isinstance(tool, BaseTool) and metadata.get(REQUIRE_CONVERSATIONAL_CONFIRMATION):
+        return ConversationalToolRunnableCallable(
+            func=_func, afunc=_afunc, name=tool_name, tool=tool
+        )
+
     return RunnableCallable(func=_func, afunc=_afunc, name=tool_name)
 
 
+class ConversationalToolRunnableCallable(RunnableCallable):
+    """Preserves a reference to the underlying BaseTool for conversational HITL confirmation."""
+
+    def __init__(self, *, func: Any, afunc: Any, name: str, tool: BaseTool):
+        super().__init__(func=func, afunc=afunc, name=name)
+        self.tool = tool
+
+
 class ToolWrapperMixin:
     wrapper: ToolWrapperType | None = None
     awrapper: AsyncToolWrapperType | None = None

src/uipath_langchain/chat/hitl.py

@@ -7,18 +7,43 @@
 from langchain_core.messages.tool import ToolCall, ToolMessage
 from langchain_core.tools import BaseTool, InjectedToolCallId
 from langchain_core.tools import tool as langchain_tool
-from uipath.core.chat import (
-    UiPathConversationToolCallConfirmationValue,
-)
+from uipath.core.chat import UiPathConversationToolCallConfirmationEvent
 
 from uipath_langchain._utils.durable_interrupt import durable_interrupt
 
 CANCELLED_MESSAGE = "Cancelled by user"
+ARGS_MODIFIED_MESSAGE = "User has modified the tool arguments"
 
 CONVERSATIONAL_APPROVED_TOOL_ARGS = "conversational_approved_tool_args"
 REQUIRE_CONVERSATIONAL_CONFIRMATION = "require_conversational_confirmation"
 
 
+def _wrap_with_args_modified_meta(result: Any, approved_args: dict[str, Any]) -> str:
+    """Wrap a tool result with metadata indicating the user modified the args."""
+    try:
+        result_value = json.loads(result) if isinstance(result, str) else result
+    except (json.JSONDecodeError, TypeError):
+        result_value = result
+    return json.dumps(
+        {
+            "meta": {
+                "message": ARGS_MODIFIED_MESSAGE,
+                "executed_args": approved_args,
+            },
+            "result": result_value,
+        }
+    )
+
+
+def get_confirmation_schema(tool: Any) -> dict[str, Any] | None:
+    """Return the JSON input schema if this tool requires confirmation, else None."""
+    metadata = getattr(tool, "metadata", None) or {}
+    if not metadata.get(REQUIRE_CONVERSATIONAL_CONFIRMATION):
+        return None
+    tool_call_schema = getattr(tool, "tool_call_schema", None)
+    return tool_call_schema.model_json_schema() if tool_call_schema is not None else {}
+
+
 class ConfirmationResult(NamedTuple):
     """Result of a tool confirmation check."""
 
@@ -47,20 +72,8 @@ def annotate_result(self, output: dict[str, Any] | Any) -> None:
             msg.response_metadata[CONVERSATIONAL_APPROVED_TOOL_ARGS] = (
                 self.approved_args
             )
-        if self.args_modified:
-            try:
-                result_value = json.loads(msg.content)
-            except (json.JSONDecodeError, TypeError):
-                result_value = msg.content
-            msg.content = json.dumps(
-                {
-                    "meta": {
-                        "args_modified_by_user": True,
-                        "executed_args": self.approved_args,
-                    },
-                    "result": result_value,
-                }
-            )
+        if self.args_modified and self.approved_args is not None:
+            msg.content = _wrap_with_args_modified_meta(msg.content, self.approved_args)
 
 
 def _patch_span_input(approved_args: dict[str, Any]) -> None:
@@ -113,39 +126,24 @@ def request_approval(
     """
     tool_call_id: str = tool_args.pop("tool_call_id")
 
-    input_schema: dict[str, Any] = {}
-    tool_call_schema = getattr(
-        tool, "tool_call_schema", None
-    )  # doesn't include InjectedToolCallId (tool id from claude/oai/etc.)
-    if tool_call_schema is not None:
-        input_schema = tool_call_schema.model_json_schema()
-
     @durable_interrupt
     def ask_confirmation():
-        return UiPathConversationToolCallConfirmationValue(
-            tool_call_id=tool_call_id,
-            tool_name=tool.name,
-            input_schema=input_schema,
-            input_value=tool_args,
-        )
+        return {
+            "tool_call_id": tool_call_id,
+            "tool_name": tool.name,
+            "input": tool_args,
+        }
 
     response = ask_confirmation()
 
-    # The resume payload from CAS has shape:
-    #   {"type": "uipath_cas_tool_call_confirmation",
-    #    "value": {"approved": bool, "input": <edited args | None>}}
     if not isinstance(response, dict):
         return tool_args
 
-    confirmation = response.get("value", response)
-    if not confirmation.get("approved", True):
+    confirmation = UiPathConversationToolCallConfirmationEvent.model_validate(response)
+    if not confirmation.approved:
         return None
 
-    return (
-        confirmation.get("input")
-        if confirmation.get("input") is not None
-        else tool_args
-    )
+    return confirmation.input if confirmation.input is not None else tool_args
 
 
 # for conversational low code agents
@@ -200,8 +198,15 @@ def wrapper(**tool_args: Any) -> Any:
             if approved_args is None:
                 return json.dumps({"meta": CANCELLED_MESSAGE})
 
+            args_modified = approved_args != tool_args
+
             _patch_span_input(approved_args)
-            return fn(**approved_args)
+            result = fn(**approved_args)
+
+            if args_modified:
+                return _wrap_with_args_modified_meta(result, approved_args)
+
+            return result
 
         # rewrite the signature: e.g. (query: str) -> (query: str, *, tool_call_id: str)
         original_sig = inspect.signature(fn)
@@ -234,6 +239,10 @@ def wrapper(**tool_args: Any) -> Any:
                 return_direct=return_direct,
             )
 
+        if result.metadata is None:
+            result.metadata = {}
+        result.metadata[REQUIRE_CONVERSATIONAL_CONFIRMATION] = True
+
         _created_tool.append(result)
         return result
 

src/uipath_langchain/runtime/messages.py

@@ -59,7 +59,7 @@ def __init__(self, runtime_id: str, storage: UiPathRuntimeStorageProtocol | None
         self.runtime_id = runtime_id
         self.storage = storage
         self.current_message: AIMessageChunk | AIMessage
-        self.tool_names_requiring_confirmation: set[str] = set()
+        self.tool_confirmation_schemas: dict[str, Any] = {}
         self.seen_message_ids: set[str] = set()
         self._storage_lock = asyncio.Lock()
         self._citation_stream_processor = CitationStreamProcessor()
@@ -320,7 +320,6 @@ async def map_ai_message_chunk_to_events(
 
         events: list[UiPathConversationMessageEvent] = []
 
-        # For every new message_id, start a new message
         if message.id not in self.seen_message_ids:
             self.current_message = message
             self.seen_message_ids.add(message.id)
@@ -339,8 +338,13 @@ async def map_ai_message_chunk_to_events(
                                 self._chunk_to_message_event(message.id, chunk)
                             )
                     case "tool_call_chunk":
-                        # Accumulate the message chunk. Note that we assume no interweaving of AIMessage and AIMessageChunks for a given message.
-                        if isinstance(self.current_message, AIMessageChunk):
+                        # Skip the first chunk — it's already assigned as current_message above,
+                        # so accumulating it with itself would duplicate fields via string concat
+                        # (e.g. tool name "search_web" becomes "search_websearch_web").
+                        if (
+                            isinstance(self.current_message, AIMessageChunk)
+                            and self.current_message is not message
+                        ):
                             self.current_message = self.current_message + message
 
         elif isinstance(message.content, str) and message.content:
@@ -425,16 +429,19 @@ async def map_current_message_to_start_tool_call_events(self):
                             self.current_message.id
                         )
 
-                        # if tool requires confirmation, we skip start tool call
-                        if (
-                            tool_call["name"]
-                            not in self.tool_names_requiring_confirmation
-                        ):
-                            events.append(
-                                self.map_tool_call_to_tool_call_start_event(
-                                    self.current_message.id, tool_call
-                                )
+                        tool_name = tool_call["name"]
+                        require_confirmation = (
+                            tool_name in self.tool_confirmation_schemas
+                        )
+                        input_schema = self.tool_confirmation_schemas.get(tool_name)
+                        events.append(
+                            self.map_tool_call_to_tool_call_start_event(
+                                self.current_message.id,
+                                tool_call,
+                                require_confirmation=require_confirmation or None,
+                                input_schema=input_schema,
                             )
+                        )
 
                 if self.storage is not None:
                     await self.storage.set_value(
@@ -531,7 +538,12 @@ async def get_message_id_for_tool_call(
         return message_id, is_last
 
     def map_tool_call_to_tool_call_start_event(
-        self, message_id: str, tool_call: ToolCall
+        self,
+        message_id: str,
+        tool_call: ToolCall,
+        *,
+        require_confirmation: bool | None = None,
+        input_schema: Any | None = None,
     ) -> UiPathConversationMessageEvent:
         return UiPathConversationMessageEvent(
             message_id=message_id,
@@ -541,6 +553,8 @@ def map_tool_call_to_tool_call_start_event(
                     tool_name=tool_call["name"],
                     timestamp=self.get_timestamp(),
                     input=tool_call["args"],
+                    require_confirmation=require_confirmation,
+                    input_schema=input_schema,
                 ),
             ),
         )
@@ -658,7 +672,7 @@ def _map_langchain_human_message_to_uipath_message_data(
             )
 
         return UiPathConversationMessageData(
-            role="user", content_parts=content_parts, tool_calls=[], interrupts=[]
+            role="user", content_parts=content_parts, tool_calls=[]
         )
 
     @staticmethod
@@ -708,7 +722,6 @@ def _map_langchain_ai_message_to_uipath_message_data(
             role="assistant",
             content_parts=content_parts,
             tool_calls=uipath_tool_calls,
-            interrupts=[],
         )
 
 

src/uipath_langchain/runtime/runtime.py

@@ -29,7 +29,7 @@
 )
 from uipath.runtime.schema import UiPathRuntimeSchema
 
-from uipath_langchain.chat.hitl import REQUIRE_CONVERSATIONAL_CONFIRMATION
+from uipath_langchain.chat.hitl import get_confirmation_schema
 from uipath_langchain.runtime.errors import LangGraphErrorCode, LangGraphRuntimeError
 from uipath_langchain.runtime.messages import UiPathChatMessagesMapper
 from uipath_langchain.runtime.schema import get_entrypoints_schema, get_graph_schema
@@ -65,9 +65,7 @@ def __init__(
         self.entrypoint: str | None = entrypoint
         self.callbacks: list[BaseCallbackHandler] = callbacks or []
         self.chat = UiPathChatMessagesMapper(self.runtime_id, storage)
-        self.chat.tool_names_requiring_confirmation = (
-            self._get_tool_names_requiring_confirmation()
-        )
+        self.chat.tool_confirmation_schemas = self._get_tool_confirmation_info()
         self._middleware_node_names: set[str] = self._detect_middleware_nodes()
 
     async def execute(
@@ -490,17 +488,36 @@ def _detect_middleware_nodes(self) -> set[str]:
 
         return middleware_nodes
 
-    def _get_tool_names_requiring_confirmation(self) -> set[str]:
-        names: set[str] = set()
+    def _get_tool_confirmation_info(self) -> dict[str, Any]:
+        """Build {tool_name: input_schema} for tools requiring confirmation.
+
+        Walks compiled graph nodes once at runtime init. This is needed because coded agents
+        (create_agent) export a compiled graph as the only artifact — there's no side channel
+        to pass confirmation metadata from the build step to the runtime.
+        """
+        schemas: dict[str, Any] = {}
         for node_name, node_spec in self.graph.nodes.items():
-            # langgraph's processing node.bound -> runnable.tool -> baseTool (if tool node)
-            tool = getattr(getattr(node_spec, "bound", None), "tool", None)
-            if tool is None:
+            bound = getattr(node_spec, "bound", None)
+            if bound is None:
                 continue
-            metadata = getattr(tool, "metadata", None) or {}
-            if metadata.get(REQUIRE_CONVERSATIONAL_CONFIRMATION):
-                names.add(getattr(tool, "name", node_name))
-        return names
+
+            # Coded agents: one tool per node
+            tool = getattr(bound, "tool", None)
+            if tool is not None:
+                schema = get_confirmation_schema(tool)
+                if schema is not None:
+                    schemas[getattr(tool, "name", node_name)] = schema
+                continue
+
+            # Low-code agents: multiple tools in one node
+            tools_by_name = getattr(bound, "tools_by_name", None)
+            if isinstance(tools_by_name, dict):
+                for name, tool in tools_by_name.items():
+                    schema = get_confirmation_schema(tool)
+                    if schema is not None:
+                        schemas[str(getattr(tool, "name", name))] = schema
+
+        return schemas
 
     def _is_middleware_node(self, node_name: str) -> bool:
         """Check if a node name represents a middleware node."""