Test Deep Agent prompt

Deep Agent prompt can be tested using various combinations of LLMs and model parameters, and comparing the responses. Refer to Test Agent Prompt for details about prompt testing.

Execution Control Using Interrupt Tool

The following steps outline the Deep Agent behavior when an Interrupt Tool enabled tool is triggered during testing.

• Navigate to the Test & Compare tab in deep agent prompt.

• Enter the required input in the User Input section.

• Click Test to execute the Deep Agent prompt

• Once testing starts, the Deep Agent begins processing the request.

  • All execution activity is displayed in the Events panel on the right side of the screen.

  • The Events panel provides a structured and real-time view of the execution flow. It displays:

    • The original user input

    • The Deep Agent’s processing steps

    • Any tool that is being invoked

    • The current execution status (e.g., running, interrupted, completed)

    This panel allows users to track how the request is being handled at each stage.

  • When the Deep Agent invokes a tool configured with Interrupt Tool, execution pauses automatically.

    • At this point:

      • The workflow is temporarily halted.

      • The execution state is preserved.

      • The Events panel indicates that the tool invocation has been interrupted.

  

• Click Format response to view the structured tool invocation details.

  This displays:

  • The tool name

  • The generated tool arguments

  • The available user decision options

  
• When interruption occurs, the following options are displayed:

  • Approve

    • The tool executes using the generated inputs.

    • The Deep Agent resumes execution from the paused state.

    • A Submit button becomes visible.

    • Click Submit to continue execution.

  

  • Edit

    • The system displays the tool inputs in editable form.

    • The user may modify:

      • Tool function arguments

      • File inputs (if applicable)

    • After making changes, click Submit.

    • The tool executes using the updated inputs.

    • The Deep Agent resumes execution from the preserved state.

  
• Reject

  • The tool invocation is canceled.

  • The current execution flow is terminated.

  • No further processing occurs for that test run.

  
• For lengthy or multi-step queries, the Deep Agent may trigger additional tool invocations that are also configured with interruption.

• In such cases:

  • Execution will pause again.

  • The user must review the tool invocation.

  • Approve, Edit, or Reject.

  • Click Submit to continue.

  This process may repeat until all required tool invocations are completed and the final response is generated.

After all required tool invocations are reviewed and approved (or edited and submitted), the Deep Agent completes execution and generates the final response. The completed output is displayed in the Events panel under the corresponding model, indicating that the workflow has successfully finished.