Claude Batch Process and Retrieve Results

arazzo: 1.0.1
info:
  title: Claude Batch Process and Retrieve Results
  summary: Submit a message batch, poll until it ends, then retrieve the JSONL results.
  description: >-
    The canonical asynchronous batch flow. The workflow submits a batch of
    message-creation requests for processing, polls the batch's status on a loop
    until its processing_status becomes ended (batches can take up to 24 hours
    and are billed at a 50 percent discount), and then retrieves the results as
    a JSONL stream. Polling uses a retry loop with a delay, and a branch routes
    the flow to result retrieval only once the batch has ended. Every step
    spells out its request inline — including the required x-api-key and
    anthropic-version headers — so the flow can be read and executed without
    opening the underlying OpenAPI description.
  version: 1.0.0
sourceDescriptions:
- name: claudeApi
  url: ../openapi/claude-messages-api.yml
  type: openapi
workflows:
- workflowId: batch-process-and-retrieve-results
  summary: Create a message batch, wait for it to end, and pull its results.
  description: >-
    Submits a batch of requests, polls retrieveMessageBatch until processing
    ends, then streams the JSONL results.
  inputs:
    type: object
    required:
    - apiKey
    - requests
    properties:
      apiKey:
        type: string
        description: Anthropic API key sent in the x-api-key header.
      anthropicVersion:
        type: string
        description: Value for the required anthropic-version header.
        default: '2023-06-01'
      requests:
        type: array
        description: >-
          Array of batch request items, each with a custom_id and a params
          object matching a Create Message request.
        items:
          type: object
  steps:
  - stepId: submitBatch
    description: >-
      Submit the batch of message-creation requests for asynchronous
      processing, capturing the batch id and initial processing status.
    operationId: createMessageBatch
    parameters:
    - name: x-api-key
      in: header
      value: $inputs.apiKey
    - name: anthropic-version
      in: header
      value: $inputs.anthropicVersion
    requestBody:
      contentType: application/json
      payload:
        requests: $inputs.requests
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      batchId: $response.body#/id
      processingStatus: $response.body#/processing_status
  - stepId: pollBatchStatus
    description: >-
      Poll the batch status. While processing_status is not "ended", retry this
      step after a delay; once ended, proceed to retrieve the results.
    operationId: retrieveMessageBatch
    parameters:
    - name: x-api-key
      in: header
      value: $inputs.apiKey
    - name: anthropic-version
      in: header
      value: $inputs.anthropicVersion
    - name: message_batch_id
      in: path
      value: $steps.submitBatch.outputs.batchId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      processingStatus: $response.body#/processing_status
      resultsUrl: $response.body#/results_url
      requestCounts: $response.body#/request_counts
    onSuccess:
    - name: retryWhileProcessing
      type: retry
      retryAfter: 30
      retryLimit: 2880
      criteria:
      - context: $response.body
        condition: $.processing_status != "ended"
        type: jsonpath
    - name: batchEnded
      type: goto
      stepId: retrieveResults
      criteria:
      - context: $response.body
        condition: $.processing_status == "ended"
        type: jsonpath
  - stepId: retrieveResults
    description: >-
      Stream the completed batch results as a JSONL file once the batch has
      ended.
    operationId: retrieveMessageBatchResults
    parameters:
    - name: x-api-key
      in: header
      value: $inputs.apiKey
    - name: anthropic-version
      in: header
      value: $inputs.anthropicVersion
    - name: message_batch_id
      in: path
      value: $steps.submitBatch.outputs.batchId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      results: $response.body
  outputs:
    batchId: $steps.submitBatch.outputs.batchId
    requestCounts: $steps.pollBatchStatus.outputs.requestCounts
    resultsUrl: $steps.pollBatchStatus.outputs.resultsUrl
    results: $steps.retrieveResults.outputs.results