Get Started with the Debugger

Use the following sequence for efficient debugging:

1. Start with a saved run profile when possible.
2. Run the workflow end-to-end once.
3. Identify the first visible symptom.
4. Trace backward from the symptom through node outputs and inputs.
5. Set a breakpoint before the suspected failing node.
6. Pin stable upstream outputs to avoid unnecessary reruns.
7. Step through the suspected node and its downstream handoff.
8. Use output simulation to test edge cases.
9. Use configuration override to test a fix.
10. Accept the configuration change only after validation.
11. Remove temporary breakpoints, pins, mocks, and overrides.
12. Rerun the full workflow.
13. Retest the affected run profiles.

Test and Debug a Workflow

Follow these steps to validate the workflow path from the user prompt to the final response.

1. Go to AI Agent Studio and open the Workflows tab.
2. Select Edit to modify a workflow.
3. From the toolbar, select Debug to view the conversation pane with Ask Oracle.
4. Enter a prompt in Ask Oracle and submit it to run the workflow.
5. Watch the highlighted run path on the canvas.
6. Select completed nodes to inspect their results.
7. Review the final response in the conversation panel.

For example, if a prompt is expected to follow a specific business path, the workflow must extract or receive the request, route it through the correct decision node, call the required data or tool nodes, transform or summarize the result, and display the expected response in the conversation pane.

If the final response is incorrect, don't start by changing the final prompt. Instead, trace the data backward through the nodes to identify where the error was introduced.