[BUG] JSON schema is invalid error when calling a tool with Claude

[tetiana-myronivska]

Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

1.01

Python Version

3.12.11

Operating System

Ubuntu 22.04.5 LTS

Installation Method

pip

Steps to Reproduce

strands-agents==1.0.1
strands-agents-tools==0.2.1
litellm==1.72.9
litellm-enterprise==0.1.7
litellm-proxy-extras==0.2.5

1. Define agent tool with two arguments: agent instance and enum
2. Initialize agent
3. Run agent with the prompt

Expected Behavior

Successful agent run. This setup and agent run used to work up until today; I did not update any of the strands or litellm packages between yesterday and today, so I'm having trouble pinning down what changed in the environment

Actual Behavior

---------------------------------------------------------------------------
HTTPStatusError                           Traceback (most recent call last)
File ~/.airconda-environments/production--ml_infra--redspot--default--v0.4.21/lib/python3.12/site-packages/litellm/llms/bedrock/chat/invoke_handler.py:200, in make_call(client, api_base, headers, data, model, messages, logging_obj, fake_stream, json_mode, bedrock_invoke_provider)
    196     client = get_async_httpx_client(
    197         llm_provider=litellm.LlmProviders.BEDROCK
    198     )  # Create a new client if none provided
--> 200 response = await client.post(
    201     api_base,
    202     headers=headers,
    203     data=data,
    204     stream=not fake_stream,
    205     logging_obj=logging_obj,
    206 )
    208 if response.status_code != 200:

File ~/.airconda-environments/production--ml_infra--redspot--default--v0.4.21/lib/python3.12/site-packages/litellm/litellm_core_utils/logging_utils.py:135, in track_llm_api_timing.<locals>.decorator.<locals>.async_wrapper(*args, **kwargs)
    134 try:
--> 135     result = await func(*args, **kwargs)
    136     return result

File ~/.airconda-environments/production--ml_infra--redspot--default--v0.4.21/lib/python3.12/site-packages/litellm/llms/custom_httpx/http_handler.py:278, in AsyncHTTPHandler.post(self, url, data, json, params, headers, timeout, stream, logging_obj, files, content)
    276     setattr(e, "status_code", e.response.status_code)
--> 278     raise e
    279 except Exception as e:

File ~/.airconda-environments/production--ml_infra--redspot--default--v0.4.21/lib/python3.12/site-packages/litellm/llms/custom_httpx/http_handler.py:234, in AsyncHTTPHandler.post(self, url, data, json, params, headers, timeout, stream, logging_obj, files, content)
    233 response = await self.client.send(req, stream=stream)
--> 234 response.raise_for_status()
    235 return response

File ~/.airconda-environments/production--ml_infra--redspot--default--v0.4.21/lib/python3.12/site-packages/httpx/_models.py:829, in Response.raise_for_status(self)
    828 message = message.format(self, error_type=error_type)
--> 829 raise HTTPStatusError(message, request=request, response=self)

HTTPStatusError: Client error '400 Bad Request' for url 'https://llm-fusion-hub.a.musta.ch/api/v2/proxy/aws/bedrock/model/us.anthropic.claude-sonnet-4-20250514-v1%3A0/converse-stream'
For more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/400

During handling of the above exception, another exception occurred:

BedrockError                              Traceback (most recent call last)
File ~/.airconda-environments/production--ml_infra--redspot--default--v0.4.21/lib/python3.12/site-packages/litellm/main.py:525, in acompletion(model, messages, functions, function_call, timeout, temperature, top_p, n, stream, stream_options, stop, max_tokens, max_completion_tokens, modalities, prediction, audio, presence_penalty, frequency_penalty, logit_bias, user, response_format, seed, tools, tool_choice, parallel_tool_calls, logprobs, top_logprobs, deployment_id, reasoning_effort, base_url, api_version, api_key, model_list, extra_headers, thinking, web_search_options, **kwargs)
    524 elif asyncio.iscoroutine(init_response):
--> 525     response = await init_response
    526 else:

File ~/.airconda-environments/production--ml_infra--redspot--default--v0.4.21/lib/python3.12/site-packages/litellm/llms/bedrock/chat/converse_handler.py:144, in BedrockConverseLLM.async_streaming(self, model, messages, api_base, model_response, timeout, encoding, logging_obj, stream, optional_params, litellm_params, credentials, logger_fn, headers, client, fake_stream, json_mode)
    134 logging_obj.pre_call(
    135     input=messages,
    136     api_key="",
   (...)
    141     },
    142 )
--> 144 completion_stream = await make_call(
    145     client=client,
    146     api_base=api_base,
    147     headers=dict(prepped.headers),
    148     data=data,
    149     model=model,
    150     messages=messages,
    151     logging_obj=logging_obj,
    152     fake_stream=fake_stream,
    153     json_mode=json_mode,
    154 )
    155 streaming_response = CustomStreamWrapper(
    156     completion_stream=completion_stream,
    157     model=model,
    158     custom_llm_provider="bedrock",
    159     logging_obj=logging_obj,
    160 )

File ~/.airconda-environments/production--ml_infra--redspot--default--v0.4.21/lib/python3.12/site-packages/litellm/llms/bedrock/chat/invoke_handler.py:263, in make_call(client, api_base, headers, data, model, messages, logging_obj, fake_stream, json_mode, bedrock_invoke_provider)
    262     error_code = err.response.status_code
--> 263     raise BedrockError(status_code=error_code, message=err.response.text)
    264 except httpx.TimeoutException:

BedrockError: {"message":"The model returned the following errors: tools.0.custom.input_schema: JSON schema is invalid. It must match JSON Schema draft 2020-12 (https://json-schema.org/draft/2020-12). Learn more about tool use at https://docs.anthropic.com/en/docs/tool-use."}

During handling of the above exception, another exception occurred:

BadRequestError                           Traceback (most recent call last)
Cell In[8], line 9
      4 session_prompt = (
      5     f"Provide initial analysis of the user profile and activities, "
      6     f"focusing on identifying any signs of ATO DP. "
      7 )
      8 start_time = time.time()
----> 9 res = run_ato_dp_agent(ato_dp_agent, BAD_ACTOR_HOST_USER_ID, session_prompt)
     10 end_time = time.time()
     11 logging.info(f"Agent run time: {end_time - start_time:.2f} seconds")

Cell In[7], line 24
     21 logging.info(f"Complete session prompt for agent:\n{complete_prompt}")
     23 # 3. Run agent
---> 24 res = agent(complete_prompt)
     25 return res
...
    976     )
    977 elif original_exception.status_code == 404:
    978     exception_mapping_worked = True

BadRequestError: litellm.BadRequestError: BedrockException - {"message":"The model returned the following errors: tools.0.custom.input_schema: JSON schema is invalid. It must match JSON Schema draft 2020-12 (https://json-schema.org/draft/2020-12). Learn more about tool use at https://docs.anthropic.com/en/docs/tool-use."}
Output is truncated. View as a scrollable element or open in a text editor. Adjust cell output settings...

Additional Context

No response

Possible Solution

One potential issue is that AttributeName seems to be defined twice and is wrong here:

            allOf:
              - $ref: '#/$defs/UserAttributeName'
            description: The name of the attribute to retrieve history for.
            type: string

tools:
  - function:
      description: |-
        Retrieve the history of changes of a specific attribute for a user.

        Args:
            attribute_name (UserAttributeName): The name of the attribute to retrieve history for.
            agent (Agent): The agent instance containing the user ID and analysis end timestamp in its state.

        Returns:
            list[dict]: A list of dictionaries representing the history of changes of the specified attribute.
      name: get_user_attribute_history
      parameters:
        $defs:
          UserAttributeName:
            description: |-
              Attribute
            enum:
              - EMAIL
              - PHONE
    
            title: UserAttributeName
            type: string
        properties:
          attribute_name:
            allOf:
              - $ref: '#/$defs/UserAttributeName'
            description: The name of the attribute to retrieve history for.
            type: string
        required:
          - attribute_name
        type: object

Related Issues

No response

[poshinchen]

Hi, could you share your code snippet so we could reproduce the issue?

[tetiana-myronivska]

hi @poshinchen, thanks for taking a look! this is a very bizarre thing, but the error went away last week (without any obvious changes / upgrades in our codebase).

I also see @westonbrown put out a fix yesterday - thank you!

in case it can be helpful, this code was failing previously

class UserAttributeName(Enum):
    """Attribute (names) associated with user accounts.
    """

    email = "EMAIL"
    phone = "PHONE"
    phone_blocked = "PHONE_BLOCKED"
    payout = "PAYOUT"
    user_name = "NAME"
    password = "PASSWORD"
    work_email = "WORK_EMAIL"

@tool
def get_user_attribute_history(
    agent: Agent, attribute_name: UserAttributeName
) -> list[dict] | PayoutInstrumentUpdateActivities | UserAttributeUpdateActivities:
    """
    Retrieve the history of changes of a specific attribute for a user.

    Args:
        attribute_name UserAttributeName: The name of the attribute to retrieve history for.

    Returns:
        list[dict]: A list of dictionaries representing the history of changes of the specified attribute.
    """
    user_id = agent.state.get("user_id")
    analysis_timestamp = agent.state.get("analysis_timestamp")
    return get_attribute_history_by_user_id_and_attribute_name(
        user_id, attribute_name, analysis_timestamp
    )