AWS bedrock-agentcore documentation change

Service: bedrock-agentcore · 2026-06-28 · Documentation low

File: bedrock-agentcore/latest/devguide/ab-testing-3p-agents.md

Summary

Added CLI and SDK examples for configuring gateway targets and online evaluations in A/B testing workflows

Security assessment

Adds operational examples without security context. Includes signing parameters but these are standard AWS configurations, not security fixes.

Diff

diff --git a/bedrock-agentcore/latest/devguide/ab-testing-3p-agents.md b/bedrock-agentcore/latest/devguide/ab-testing-3p-agents.md
index 9732b915e..07f9192cf 100644
--- a//bedrock-agentcore/latest/devguide/ab-testing-3p-agents.md
+++ b//bedrock-agentcore/latest/devguide/ab-testing-3p-agents.md
@@ -115,0 +116,20 @@ To register a Lambda function URL as a target, set `protocolType` to `CUSTOM` an
+###### Example
+
+AgentCore CLI
+    
+    
+    
+    agentcore add gateway-target \
+      --name customer-support-control \
+      --gateway cs-3p-abtest-gw \
+      --type passthrough \
+      --passthrough-endpoint https://<control-id>.lambda-url.us-west-2.on.aws/ \
+      --passthrough-protocol CUSTOM \
+      --stickiness-identifier '$context.header.x-session-id' \
+      --stickiness-timeout 28800 \
+      --signing-service lambda \
+      --signing-region us-west-2
+
+AWS CLI
+    
+    
@@ -166,0 +187,43 @@ When you send traffic through the gateway, direct it to the control variant’s
+###### Example
+
+AgentCore CLI
+    
+
+Add one online evaluation config per variant — using the `service.name` and event log group from Step 1: Instrument your agent for AgentCore observability as the data source — then start the test in `target-based` mode:
+    
+    
+    agentcore add online-eval \
+      --name cs-control-eval \
+      --evaluator Builtin.Correctness \
+      --service-name customer-support-control \
+      --log-group-name /aws/bedrock-agentcore/agents/cs-agent-control/runtime-logs \
+      --enable-on-create
+    
+    agentcore add online-eval \
+      --name cs-treatment-eval \
+      --evaluator Builtin.Correctness \
+      --service-name customer-support-treatment \
+      --log-group-name /aws/bedrock-agentcore/agents/cs-agent-treatment/runtime-logs \
+      --enable-on-create
+    
+    agentcore deploy
+    
+    agentcore run ab-test \
+      --name cs-3p-abtest \
+      --gateway cs-3p-abtest-gw \
+      --mode target-based \
+      --control-target customer-support-control \
+      --treatment-target customer-support-treatment \
+      --control-online-eval cs-control-eval \
+      --treatment-online-eval cs-treatment-eval \
+      --control-weight 50 \
+      --treatment-weight 50 \
+      --wait
+
+Use `agentcore status` to view per-variant results while the test runs and `agentcore stop` to end it.
+
+AWS SDK (boto3)
+    
+
+Create one online evaluation config per variant with `create_online_evaluation_config`. Set `serviceNames` to the endpoint’s `service.name` and `logGroupNames` to its event log group. Then create the test with `create_ab_test`, using `perVariantOnlineEvaluationConfig`. For the full boto3 example — including sending traffic, polling results, and stopping the test — see [Run an A/B test with target-based routing](./ab-testing-target-based.html), starting from the **Create the A/B test** step.
+