name: customer_support_agent
version: 1.0.0
description: Autonomous customer support agent

states:
  - id: start
    type: initial
    next: analyze_inquiry
  
  - id: analyze_inquiry
    type: llm
    model: analyzer
    prompt: |
      Analyze this customer inquiry:
      {inquiry}
      
      Determine:
      1. Type of inquiry (general, technical, billing, etc.)
      2. Priority level
      3. Required information
      4. Potential actions      
    input:
      inquiry: "${input.message}"
    next: gather_context
  
  - id: gather_context
    type: llm
    model: researcher
    prompt: |
      Gather relevant information for:
      {inquiry_type}
      
      Search in:
      - Product documentation
      - Knowledge base
      - Previous tickets      
    input:
      inquiry_type: "${previous.output.type}"
    next: formulate_response
  
  - id: formulate_response
    type: llm
    model: responder
    prompt: |
      Create response for:
      
      Inquiry: {inquiry}
      Type: {type}
      Context: {context}
      
      Guidelines:
      - Be clear and concise
      - Use appropriate tone
      - Include relevant links
      - Suggest next steps      
    input:
      inquiry: "${input.message}"
      type: "${states.analyze_inquiry.output.type}"
      context: "${previous.output}"
    next: evaluate_response
  
  - id: evaluate_response
    type: llm
    model: evaluator
    prompt: |
      Evaluate response quality:
      
      Response: {response}
      Inquiry: {inquiry}
      Type: {type}
      
      Check for:
      - Accuracy
      - Completeness
      - Tone
      - Actionability      
    input:
      response: "${previous.output}"
      inquiry: "${input.message}"
      type: "${states.analyze_inquiry.output.type}"
    transitions:
      - when: "${output.quality === 'high'}"
        next: send_response
      - when: "${output.quality === 'medium'}"
        next: improve_response
      - when: "${output.quality === 'low'}"
        next: escalate
  
  - id: improve_response
    type: llm
    model: improver
    prompt: |
      Improve this response:
      {response}
      
      Issues to address:
      {issues}      
    input:
      response: "${states.formulate_response.output}"
      issues: "${previous.output.issues}"
    next: evaluate_response
  
  - id: send_response
    type: tool
    tool: send_message
    parameters:
      message: "${states.formulate_response.output}"
      channel: "${input.channel}"
    next: check_followup
  
  - id: check_followup
    type: llm
    model: analyzer
    prompt: |
      Check if follow-up is needed:
      
      Inquiry: {inquiry}
      Response: {response}
      Customer History: {history}      
    input:
      inquiry: "${input.message}"
      response: "${states.formulate_response.output}"
      history: "${context.customer_history}"
    transitions:
      - when: "${output.needs_followup}"
        next: plan_followup
      - next: end
  
  - id: plan_followup
    type: llm
    model: planner
    prompt: |
      Plan follow-up action:
      
      Inquiry: {inquiry}
      Response: {response}
      Follow-up Type: {type}      
    input:
      inquiry: "${input.message}"
      response: "${states.formulate_response.output}"
      type: "${previous.output.followup_type}"
    next: execute_followup
  
  - id: execute_followup
    type: tool
    tool: "${previous.output.tool}"
    parameters: "${previous.output.parameters}"
    next: end
  
  - id: escalate
    type: tool
    tool: create_ticket
    parameters:
      inquiry: "${input.message}"
      type: "${states.analyze_inquiry.output.type}"
      priority: "${states.analyze_inquiry.output.priority}"
      context: "${context.gathered_context}"
    next: notify_agent
  
  - id: notify_agent
    type: tool
    tool: send_notification
    parameters:
      message: "New ticket created: ${previous.output.ticket_id}"
      agent: "${previous.output.assigned_agent}"
    next: end
  
  - id: end
    type: output
    value: "${context.final_response}"