AWS adds skill-routing checks for AI agents

AWS has added evaluators for measuring whether AI agents select the appropriate reusable instructions and follow their required steps. Strands Evals can test recorded development runs, while Amazon Bedrock AgentCore Evaluations can inspect OpenTelemetry traces on demand, in batches or through continuous production sampling. The tools separate routing errors from incomplete execution and return results for individual skills. Strands also provides a deterministic check that a required skill loaded, but that check has no AgentCore command-line equivalent. Judge-based evaluators produce no score when an agent invokes no skill.