Definição de argumentos

Arguments pass runtime information into and out of an agent, just as inputs and outputs do for activities or processes. Define input and output arguments in the Data Manager panel, then reference them in the user prompt so the agent can use them at runtime.

Define input and output arguments

1. In the Data Manager panel, open the Input or Output tab.
2. For each argument, configure the Name, Type, Description, and whether it is Required.

To generate arguments from an existing JSON payload, select Generate from payload. To switch between the property builder and JSON editor views, select the angle brackets icon.

Observação:

Provide a description for every argument. An accurate description helps the agent understand how to use the argument effectively.

Result: The argument appears in the Data Manager and is available to select from the @ picker in the user prompt.

Reference arguments in the user prompt

The agent sees only arguments that are explicitly referenced in the user prompt.

1. In the user prompt, enter @ and select the argument by name from the picker.
2. The argument is inserted as {{argumentName}} in the stored prompt format and is substituted with its runtime value when the agent runs.

Observação:

Argument names are case-sensitive. Selecting the argument from the @ picker ensures the name matches exactly.

Result: The agent receives the argument's value in its context window when triggered.

Tipos de argumentos compatíveis

• String
• Número
• Número inteiro
• Booleano
• Object
• Matriz