Handling Guardrails Responses

[RashmiJK]

### Code -

from langchain.chat_models import init_chat_model
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from nemoguardrails import RailsConfig
from nemoguardrails.integrations.langchain.runnable_rails import RunnableRails

config = RailsConfig.from_path("./config")
guardrails = RunnableRails(
        config,
        output_key="user_question",
        passthrough=True,
        verbose=False
    )

prompt = ChatPromptTemplate.from_template("Respond to user question in short. User question: {user_question}")
model = init_chat_model("gpt-4.1") # gpt-4o-mini
output_parser = StrOutputParser()

# guardrails around the entire chain
# unsafe question

base_chain = prompt | model | output_parser

validate_base_chain = guardrails | base_chain

response = validate_base_chain.invoke({"user_question":"Are politicians good or bad people?"})
print(response)

### Configuration used for the above code (flow) -

models:
  - type: main
    engine: openai
    model: gpt-4o

rails:
  input:
    flows:
      - self check input

  output:
    flows:
      - self check output

prompts:
  - task: self_check_input
    content: |
      Your task is to filter out messages about politics.

      Block messages that discuss, ask about, or reference:
      - Political parties, politicians, or elections
      - Government policy debates or political ideologies
      - Political news, campaigns, or voting

      User message: "{{ user_input }}"

      Answer only: Yes (block) or No (allow)

  - task: self_check_output
    content: |
      Your task is to filter model responses that contain the word "perspective".

      Model response: {{ bot_response }}

      Answer only: Yes (block) or No (allow)

Output -

No problem! Let me know if you have any other questions.

The guardrails layer returns the string I’m sorry, I can’t respond to that. in the  user_question  field. How should we handle this kind of output correctly when integrating with LangChain chains that expect the original  user_question  input? I’ve spent quite a bit of time reading through the NeMo Guardrails documentation, but I’m still not sure what the recommended pattern is for this case.