BuiltinEvaluator

class aws_cdk.aws_bedrockagentcore.BuiltinEvaluator(value)

Bases: object

Built-in evaluators provided by Amazon Bedrock AgentCore.

These evaluators assess different aspects of agent performance at various levels (session, trace, or tool call).

ExampleMetadata:: fixture=default infused

Example:

# custom_evaluator: agentcore.Evaluator


evaluation = agentcore.OnlineEvaluationConfig(self, "MixedEvaluation",
    online_evaluation_config_name="mixed_evaluation",
    evaluators=[
        # Built-in evaluators
        agentcore.EvaluatorSelector.builtin(agentcore.BuiltinEvaluator.HELPFULNESS),
        agentcore.EvaluatorSelector.builtin(agentcore.BuiltinEvaluator.CORRECTNESS),
        # Custom evaluator
        agentcore.EvaluatorSelector.custom(custom_evaluator)
    ],
    data_source=agentcore.DataSourceConfig.from_cloud_watch_logs(
        log_group_names=["/aws/bedrock-agentcore/my-agent"],
        service_names=["my-agent.default"]
    )
)

Parameters:

value (str) –

The evaluator identifier string.

Attributes

COHERENCE = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

CONCISENESS = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

CORRECTNESS = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

FAITHFULNESS = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

GOAL_SUCCESS_RATE = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

HARMFULNESS = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

HELPFULNESS = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

INSTRUCTION_FOLLOWING = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

REFUSAL = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

RESPONSE_RELEVANCE = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

STEREOTYPING = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

TOOL_PARAMETER_ACCURACY = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

TOOL_SELECTION_ACCURACY = <aws_cdk.aws_bedrockagentcore.BuiltinEvaluator object>

value: The string value of the built-in evaluator.