arazzo: 1.0.1
info:
  title: OpenAI Moderate then Chat
  summary: Screen user input with the moderation endpoint, then chat only if it is safe.
  description: >-
    Runs the supplied user text through the moderation classifier and branches:
    when the content is flagged the flow ends without calling the model, and
    when it is not flagged it creates a chat completion. Every step spells out
    its request inline so the flow can be read and executed without opening the
    underlying OpenAPI description.
  version: 1.0.0
sourceDescriptions:
- name: openaiApi
  url: ../openapi/openai-openapi-master.yml
  type: openapi
workflows:
- workflowId: moderate-then-chat
  summary: Moderate input text and create a chat completion only when it is allowed.
  description: >-
    Classifies the user input for policy violations, and only when the content
    is not flagged does it forward the message to the chat completions endpoint.
  inputs:
    type: object
    required:
    - apiKey
    - chatModel
    - userMessage
    properties:
      apiKey:
        type: string
        description: OpenAI API key used as a Bearer token.
      chatModel:
        type: string
        description: The model id for the chat completion (e.g. gpt-4o-mini).
      userMessage:
        type: string
        description: The user message to moderate and then send to the model.
  steps:
  - stepId: moderate
    description: Classify the user input for policy violations.
    operationId: createModeration
    parameters:
    - name: Authorization
      in: header
      value: "Bearer $inputs.apiKey"
    requestBody:
      contentType: application/json
      payload:
        input: $inputs.userMessage
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      flagged: $response.body#/results/0/flagged
    onSuccess:
    - name: contentAllowed
      type: goto
      stepId: createChat
      criteria:
      - context: $response.body
        condition: $.results[0].flagged == false
        type: jsonpath
    - name: contentFlagged
      type: end
      criteria:
      - context: $response.body
        condition: $.results[0].flagged == true
        type: jsonpath
  - stepId: createChat
    description: Create a chat completion for the moderated, allowed user message.
    operationId: createChatCompletion
    parameters:
    - name: Authorization
      in: header
      value: "Bearer $inputs.apiKey"
    requestBody:
      contentType: application/json
      payload:
        model: $inputs.chatModel
        messages:
        - role: user
          content: $inputs.userMessage
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      completionId: $response.body#/id
      reply: $response.body#/choices/0/message/content
  outputs:
    flagged: $steps.moderate.outputs.flagged
    reply: $steps.createChat.outputs.reply