'KeyError: 400' occurs when using langchain_community.agent_toolkits.openapi.spec.reduce_openapi_spec.

[eunhye1kim]

Checked other resources

• I added a very descriptive title to this issue.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.
• I am sure that this is a bug in LangChain rather than my code.
• The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).

Example Code

import os
import requests
import yaml

os.environ["OPENAI_API_KEY"] = "sk-REDACTED"

from langchain_community.agent_toolkits.openapi import planner
from langchain_openai.chat_models import ChatOpenAI

from langchain_community.agent_toolkits.openapi.spec import reduce_openapi_spec
from langchain.requests import RequestsWrapper
from requests.packages.urllib3.exceptions import InsecureRequestWarning

# Ignore SSL warnings
requests.packages.urllib3.disable_warnings(InsecureRequestWarning)

with open("/home/ehkim/git/testprj/code_snippet/swagger.yaml") as f:
    data = yaml.load(f, Loader=yaml.FullLoader)
swagger_api_spec = reduce_openapi_spec(data)

def construct_superset_aut_headers(url=None):
    import requests
    url = "https://superset.mydomain.com/api/v1/security/login"
    payload = {
        "username": "myusername",
        "password": "mypassword",
        "provider": "db",
        "refresh": True
    }
    headers = {
        "Content-Type": "application/json"
    }

    response = requests.post(url, json=payload, headers=headers, verify=False)
    data = response.json()
    return {"Authorization": f"Bearer {data['access_token']}"}

from langchain.globals import set_debug
set_debug(True)

llm = ChatOpenAI(model='gpt-4o')
swagger_requests_wrapper = RequestsWrapper(headers=construct_superset_aut_headers(), verify=False)
superset_agent = planner.create_openapi_agent(
    swagger_api_spec,
    swagger_requests_wrapper,
    llm, allow_dangerous_requests=True,
    handle_parsing_errors=True)

superset_agent.run(
    """
    1. Get the dataset using the following information. (tool: requests_post, API: /api/v1/dataset/get_or_create/, database_id: 1, table_name: raw_esg_volume, response : {{'result' : {{'table_id': (dataset_id)}}}})
    2. Retrieve the dataset information obtained in step 1. (tool: requests_get, API: /api/v1/dataset/dataset/{{dataset_id}}/, params: None)
    3. Create a chart referencing the dataset obtained in step 2. The chart should plot the trend of total, online_news, and (total - online_news) values as a line chart. (tool: requests_post, API: /api/v1/chart/, database_id: 1)
    4. Return the URL of the created chart. https://superset.mydomain.com/explore/?slice_id={{chart_id}}
    When specifying the action, only write the tool name without any additional explanation.
    """
)

In this file, I used swagger.yaml file from https://superset.demo.datahubproject.io/api/v1/_openapi.
It's json format, so I converted it with below code.

import json
import yaml

# read file
with open('swagger.json', 'r') as json_file:
    json_data = json.load(json_file)

# write file
with open('swagger.yaml', 'w') as yaml_file:
    yaml.dump(json_data, yaml_file, default_flow_style=False)

Error Message and Stack Trace (if applicable)

(myenv) ehkim@ehkim-400TEA-400SEA:~/git/testprj/code_snippet$ python openapi-agent.py
/home/ehkim/anaconda3/envs/myenv/lib/python3.12/site-packages/langchain/_api/module_import.py:92: LangChainDeprecationWarning: Importing RequestsWrapper from langchain is deprecated. Please replace deprecated imports:

from langchain import RequestsWrapper

with new imports of:

from langchain_community.utilities import RequestsWrapper
You can use the langchain cli to automatically upgrade many imports. Please see documentation here https://python.langchain.com/v0.2/docs/versions/v0_2/
warn_deprecated(
Traceback (most recent call last):
File "/home/ehkim/git/testprj/code_snippet/openapi-agent.py", line 23, in
swagger_api_spec = reduce_openapi_spec(data)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ehkim/anaconda3/envs/myenv/lib/python3.12/site-packages/langchain_community/agent_toolkits/openapi/spec.py", line 53, in reduce_openapi_spec
(name, description, dereference_refs(docs, full_schema=spec))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ehkim/anaconda3/envs/myenv/lib/python3.12/site-packages/langchain_core/utils/json_schema.py", line 108, in dereference_refs
else _infer_skip_keys(schema_obj, full_schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ehkim/anaconda3/envs/myenv/lib/python3.12/site-packages/langchain_core/utils/json_schema.py", line 80, in _infer_skip_keys
keys += _infer_skip_keys(v, full_schema, processed_refs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ehkim/anaconda3/envs/myenv/lib/python3.12/site-packages/langchain_core/utils/json_schema.py", line 80, in _infer_skip_keys
keys += _infer_skip_keys(v, full_schema, processed_refs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ehkim/anaconda3/envs/myenv/lib/python3.12/site-packages/langchain_core/utils/json_schema.py", line 76, in _infer_skip_keys
ref = _retrieve_ref(v, full_schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ehkim/anaconda3/envs/myenv/lib/python3.12/site-packages/langchain_core/utils/json_schema.py", line 17, in _retrieve_ref
out = out[int(component)]
~~~^^^^^^^^^^^^^^^^
KeyError: 400

Description

I'm trying to use the langchain library to execute the OpenAPI Agent and fully interpret an OpenAPI specification using the reduce_openapi_spec function in my script.
I expect the agent to execute normally without any errors.
Instead, it raises a KeyError: 400.

System Info

(myenv) ehkim@ehkim-400TEA-400SEA:~/git/testprj/code_snippet$ pip freeze | grep langchain
langchain==0.2.8
langchain-cli==0.0.21
langchain-community==0.2.7
langchain-core==0.2.20
langchain-experimental==0.0.37
langchain-google-vertexai==0.0.3
langchain-openai==0.1.16
langchain-robocorp==0.0.3
langchain-text-splitters==0.2.2
langchainhub==0.1.15