Custom evaluation criteria in Agentforce allow developers to define precise validation rules for AI agent responses using string and numeric comparisons. The post demonstrates how to create test definitions using XML files, deploy them via Salesforce CLI, and validate agent outputs against expected values using JSONPath expressions and comparison operators. This testing approach helps ensure AI agents return correct, consistent results by checking for specific phrases, actions, and numeric values in responses.
Table of contents
What are custom evaluation criteria?Setting up your first testExample: Making sure the agent knows its acronymDeploying and running your testsConclusionResourcesAbout the authorSort: