Testing Center test case criteria
Key Takeaway
A Testing Center test case pairs an utterance with the subagent, actions, and response you expect. In a CSV file, Expected Subagent and Expected Actions take API names, not labels.
Testing Center stores test cases for an Agentforce agent and runs them again and again in batches. Each test case pairs an utterance, something a user might say, with criteria for what should happen. The criteria are which subagent should run, which actions should fire, and what the response should cover. AI agents are non-deterministic, which means the same request can produce different results. For that reason, you test them across many scenarios to confirm they perform consistently and accurately.
Why store and rerun tests
You first test and troubleshoot an agent by hand in Agentforce Builder. Then you use Testing Center to store those tests and run them over and over. A manual check is one-off. A stored test case can be rerun after you update instructions, subagents, actions, filters, or guardrails.