Why does passing a ToolMessage trigger a 'Template error: unknown filter' in LangChain?

I'm following the tutorial provided here to create a simple tool-calling function in LangChain. I've used the exact code from the tutorial, only modifying the tool itself. The code used is as follows:

PYTHONCopied
from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint
from langchain_core.messages import HumanMessage, AIMessage

@tool
def duddify(first_number: int, second_number: int):
    """Concatenate two numbers twice."""
    return int(f"{str(first_number)}{str(second_number)}{str(first_number)}{str(second_number)}")

@tool
def pippify(number: int):
    """Repeats the number three times."""
    return f"{str(number)} {str(number)} {str(number)}"

llm = ChatHuggingFace(
    llm=HuggingFaceEndpoint(
        repo_id="mistralai/Mistral-7B-Instruct-v0.3",
        task='text-generation'
    ),
    verbose=True
)

llm_with_tools = llm.bind_tools([duddify, pippify])

query = "pippify the number 102"

messages = [HumanMessage(content=query)]

ai_msg = llm_with_tools.invoke(messages)
messages.append(ai_msg)

for tool_call in ai_msg.tool_calls:
    selected_tool = {"duddify": duddify, "pippify": pippify}[tool_call["name"].lower()]
    tool_msg = selected_tool.invoke(tool_call)
    messages.append(tool_msg)

print(f"Messages:\n{messages}\n")
for message in messages:
    print(f'{message}\ntype {type(message)}\n')

final_response = llm_with_tools.invoke(messages)
print(final_response)

The conversation history now includes three messages: a HumanMessage (with the query), an AIMessage (from the initial invocation), and a ToolMessage (containing the tool's output). This is exemplified in the print outputs:

PYTHONCopied
Messages:
[HumanMessage(content='pippify the number 102 and calculate 8 bryanify 3', additional_kwargs={}, response_metadata={}), AIMessage(content='', additional_kwargs={'tool_calls': [ChatCompletionOutputToolCall(function=ChatCompletionOutputFunctionDefinition(arguments={'number': 102}, name='pippify', description=None), id='0', type='function')]}, response_metadata={'token_usage': ChatCompletionOutputUsage(completion_tokens=19, prompt_tokens=277, total_tokens=296), 'model': '', 'finish_reason': 'stop'}, id='run-b00d61a6-0a4f-4b02-8cac-30663ae5e5e3-0', tool_calls=[{'name': 'pippify', 'args': {'number': 102}, 'id': '0', 'type': 'tool_call'}]), ToolMessage(content='102 102 102', name='pippify', tool_call_id='0')]

content='pippify the number 102' additional_kwargs={} response_metadata={}
type <class 'langchain_core.messages.human.HumanMessage'>

content='' additional_kwargs={'tool_calls': [ChatCompletionOutputToolCall(function=ChatCompletionOutputFunctionDefinition(arguments={'number': 102}, name='pippify', description=None), id='0', type='function')]} response_metadata={'token_usage': ChatCompletionOutputUsage(completion_tokens=19, prompt_tokens=277, total_tokens=296), 'model': '', 'finish_reason': 'stop'} id='run-f65ed238-9c6f-42f0-84f1-c7b521920f09-0' tool_calls=[{'name': 'pippify', 'args': {'number': 102}, 'id': '0', 'type': 'tool_call'}]
type <class 'langchain_core.messages.ai.AIMessage'>

content='102 102 102' name='pippify' tool_call_id='0'
type <class 'langchain_core.messages.tool.ToolMessage'>

Despite the tool calls working correctly, when invoking llm_with_tools a second time with the full messages list, I receive the error below:

Template error: unknown filter: filter string is unknown (in <string>:79)

Full error message:

Traceback (most recent call last):
  File "C:\Users\path\.venv\Lib\site-packages\huggingface_hub\utils\_http.py", line 406, in hf_raise_for_status
    response.raise_for_status()
  File "C:\Users\path\.venv\Lib\site-packages\requests\models.py", line 1024, in raise_for_status
    raise HTTPError(http_error_msg, response=self)
requests.exceptions.HTTPError: 422 Client Error: Unprocessable Entity for url: https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.3/v1/chat/completions

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "c:\Users\path\testing_langchain_agents.py", line 57, in <module>
    print(f"\n{llm_with_tools.invoke([messages[-1]])}\n")
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\path\.venv\Lib\site-packages\langchain_core\runnables\base.py", line 535, in invoke
    return self.bound.invoke(
  File "C:\Users\path\.venv\Lib\site-packages\langchain_core\language_models\chat_models.py", line 286, in invoke
    self.generate_prompt(
  File "C:\Users\path\.venv\Lib\site-packages\langchain_core\language_models\chat_models.py", line 786, in generate_prompt
    return self.generate(prompt_messages, stop=stop, callbacks=callbacks, **kwargs)
  File "C:\Users\path\.venv\Lib\site-packages\langchain_core\language_models\chat_models.py", line 643, in generate
    raise e
  File "C:\Users\path\.venv\Lib\site-packages\langchain_core\language_models\chat_models.py", line 633, in generate
    self._generate_with_cache(
  File "C:\Users\path\.venv\Lib\site-packages\langchain_core\language_models\chat_models.py", line 851, in _generate_with_cache
    result = self._generate(
             ^^^^^^^^^^^^^^^
  File "C:\Users\path\.venv\Lib\site-packages\langchain_huggingface\chat_models\huggingface.py", line 370, in _generate
    answer = self.llm.client.chat_completion(messages=message_dicts, **kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\path\.venv\Lib\site-packages\huggingface_hub\inference\_client.py", line 892, in chat_completion
    data = self.post(model=model_url, json=payload, stream=stream)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\path\.venv\Lib\site-packages\huggingface_hub\inference\_client.py", line 306, in post
    hf_raise_for_status(response)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\path\.venv\Lib\site-packages\huggingface_hub\utils\_http.py", line 477, in hf_raise_for_status
    raise _format(HfHubHTTPError, str(e), response) from e
huggingface_hub.errors.HfHubHTTPError: 422 Client Error: Unprocessable Entity for url: https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.3/v1/chat/completions (Request ID: z1qkv0HDiNWIwsBGxA-xp)
Template error: unknown filter: filter string is unknown (in <string>:79)

I determined that the error only occurs when a ToolMessage is included in the list passed to llm_with_tools.invoke(). For instance:

• llm_with_tools.invoke(HumanMessage(content=query)) works
• llm_with_tools.invoke([HumanMessage(content=query)]) works
• llm_with_tools.invoke([HumanMessage(content=query), ai_msg]) works
• llm_with_tools.invoke(messages[-1]) does not work, since it expects a PromptValue, str, or list of BaseMessages rather than a ToolMessage.
• llm_with_tools.invoke([messages[-1]]) fails with the templating error.

I have tried the following:

• Reinstalling langchain, langchain_huggingface, and related packages
• Testing on multiple Python versions (3.11 as the original, 3.12 with the same error, and 3.13 where torch isn’t installable)
• Invoking with different message combinations, which confirmed the error only appears when a ToolMessage is inside the list.

I understand that ToolMessage objects are meant for internal use and to be appended to the conversation log, but they aren’t supported in the prompt generation template. This leads to the error when they are passed directly without conversion to a valid message type (such as HumanMessage or AIMessage).

The recommended workarounds include:

1. Filtering out ToolMessages when re-invoking the model.
2. Converting ToolMessage objects into a supported message type (e.g., wrapping the tool output in an AIMessage with an explanation).

Examples of both approaches were provided in the answers above.

My question is: Is this behavior intentional, and what is the best practice for including tool results in subsequent LLM invocations?