Steps

Workflows in AgentScope AI are structured sequences of actions that guide agents through complex tasks. Each workflow consists of multiple steps, where each step represents a discrete operation or decision. Steps can include tool invocations, data transformations, conditional logic, or external API calls.

Defining a Workflow Step

Each step within a workflow follows a structured schema that includes:

• Step ID: A unique identifier for the step.

• Action: The operation to be performed (e.g., tool execution, branching, or API call).

• Input Parameters: Data required for executing the step.

• Output Handling: How results are processed and passed to the next step.

• Conditions (Optional): Logic that determines whether a step should be executed.

Example: Defining a Workflow Step

Below is an example of a workflow step that fetches user information from an API and processes it:

const fetchUserData = {
  id: "fetch_user_data",
  action: "call_api",
  parameters: {
    url: "https://api.agentscope-ai.io/user",
    method: "GET",
    headers: {
      Authorization: "Bearer {token}"
    }
  },
  output: {
    storeAs: "userData"
  }
};

Workflow Execution Flow

A workflow consists of multiple steps executed sequentially or based on conditional logic. The execution flow typically follows this pattern:

1. Initialize Workflow: Set up the required context and input parameters.

2. Execute Steps: Perform actions as defined in each step.

3. Evaluate Conditions: Determine the next step based on execution results.

4. Handle Errors: Apply error-handling mechanisms for failed steps.

5. Finalize Workflow: Return processed results or trigger follow-up actions.

Example: Multi-Step Workflow

The following example demonstrates a simple workflow that verifies a user and retrieves their transaction history:

const userVerificationWorkflow = {
  id: "user_verification",
  steps: [
    {
      id: "fetch_user_data",
      action: "call_api",
      parameters: {
        url: "https://api.agentscope-ai.io/user",
        method: "GET"
      },
      output: {
        storeAs: "userData"
      }
    },
    {
      id: "check_user_status",
      action: "condition",
      condition: "userData.status === 'verified'",
      nextStep: "fetch_transaction_history"
    },
    {
      id: "fetch_transaction_history",
      action: "call_api",
      parameters: {
        url: "https://api.agentscope-ai.io/transactions",
        method: "GET",
        query: {
          userId: "{userData.id}"
        }
      },
      output: {
        storeAs: "transactionHistory"
      }
    }
  ]
};

Error Handling in Workflow Steps

To ensure reliability, workflows can include error-handling mechanisms:

• Retries: Automatically retry failed steps.

• Fallback Steps: Define alternative actions if a step fails.

• Logging & Alerts: Capture error details and notify relevant systems.

Example of a step with error handling:

const fetchUserDataWithRetry = {
  id: "fetch_user_data",
  action: "call_api",
  parameters: {
    url: "https://api.agentscope-ai.io/user",
    method: "GET"
  },
  output: {
    storeAs: "userData"
  },
  errorHandling: {
    retry: 3,
    fallback: "use_cached_user_data"
  }
};