Clasificar Mensaje Publicitario de Twitch

Workflow manual que clasifica un mensaje de chat de Twitch como publicitario o no usando un LLM con salida estructurada (JSON schema). Demuestra workflowInput, concatenación de cadenas en CEL, structured_output_schema y returnExpression con get(step(0), ...).

manual llm structured-output json-schema moderation twitch

Descargar ZIP

Qué está incluido

Campo	Valor
Nombre	Clasificar Mensaje Publicitario de Twitch
Tipo	Workflow
Trigger	manual
Pasos	1
Entradas	2

JSON Completo

Clic para expandir

{
  "description": "Manual workflow that classifies a Twitch chat message as advertising or not using an LLM with structured output (JSON schema). Demonstrates workflowInput, CEL string concatenation, structured_output_schema, and returnExpression with get(step(0), ...).",
  "tags": [
    "manual",
    "llm",
    "structured-output",
    "json-schema",
    "moderation",
    "twitch"
  ],
  "document": {
    "name": "Classify Twitch Advertising Message",
    "enabled": true,
    "log_level": "info",
    "trigger": {
      "type": "manual"
    },
    "workflowInput": [
      {
        "name": "twitch_username",
        "description": "Twitch username that sent the chat message",
        "schema": {
          "type": "string",
          "required": true,
          "enableEnums": false,
          "enum": []
        }
      },
      {
        "name": "twitch_message",
        "description": "Raw Twitch chat message text to classify",
        "schema": {
          "type": "string",
          "required": true,
          "enableEnums": false,
          "enum": []
        }
      }
    ],
    "steps": [
      {
        "condition": {
          "expr": "true"
        },
        "action": {
          "actionType": "actions/ai/llm/chat/generate",
          "actions/ai/llm/chat/generate": {
            "is_advanced": true,
            "platform": "openai",
            "model": "gpt-4.1-mini",
            "temperature": 0,
            "max_tokens": 400,
            "enable_structured_outputs": true,
            "structured_output_schema_name": "output",
            "structured_output_schema": {
              "expr": "{\n  \"type\": \"object\",\n  \"description\": \"Classify Twitch chat for moderation.\",\n  \"additionalProperties\": false,\n  \"required\": [\"is_advertising\", \"confidence\", \"category\", \"reason\", \"signals\"],\n  \"properties\": {\n    \"is_advertising\": { \"type\": \"boolean\" },\n    \"confidence\": { \"type\": \"number\", \"minimum\": 0, \"maximum\": 1 },\n    \"category\": {\n      \"type\": \"string\",\n      \"enum\": [\"viewer_advertising\", \"bot_spam_advertising\", \"not_advertising\"]\n    },\n    \"reason\": { \"type\": \"string\", \"minLength\": 1, \"maxLength\": 200 },\n    \"signals\": {\n      \"type\": \"array\",\n      \"items\": { \"type\": \"string\", \"minLength\": 1, \"maxLength\": 80 }\n    }\n  }\n}"
            },
            "messages": [
              {
                "role": "user",
                "content_type": "text",
                "content": {
                  "expr": "'You classify Twitch chat for moderation. Return ONLY valid JSON with exact keys: is_advertising (boolean), confidence (number from 0 to 1), category (viewer_advertising|bot_spam_advertising|not_advertising), reason (short string), signals (array of short strings). Mark as advertising for promotions, self-promo, follower/follow-back, links, codes, giveaways, paid offers, suspicious repetitive spam, or DM bait. If uncertain, return not_advertising with lower confidence. Do not include markdown code fences.'"
                }
              },
              {
                "role": "user",
                "content_type": "text",
                "content": {
                  "expr": "'Twitch username: ' + twitch_username + '\\nMessage: ' + twitch_message"
                }
              }
            ]
          }
        }
      }
    ],
    "returnExpression": {
      "expr": "get(step(0), 'choices.0.message.content')"
    }
  }
}

Como Importar

1. Descargue el archivo ZIP usando el botón de arriba
2. Navegue a Configuración de la Empresa → Transferencia de Datos en su panel de AutoTalk
3. Haga clic en Importar y seleccione el archivo ZIP descargado
4. Revise el resumen de importación y confirme
5. Todos los IDs de documentos se remapean automáticamente — no se necesita edición manual de IDs