AWS adds skill-focused evaluators to Strands Evals and Bedrock AgentCore

Amazon Web Services announced that the Strands Evals SDK and the Amazon Bedrock AgentCore Evaluations capability include three skill‑focused evaluators: Skill Selection Accuracy, which returns a binary result indicating whether an invoked skill matches the task; Skill Instruction Following, which rates how completely an agent follows a skill’s prescribed steps on a five‑level scale; and Skill Invoked, a deterministic check that confirms a named skill was loaded successfully. These evaluators operate on recorded agent trajectories in Strands Evals or on OpenTelemetry traces captured by AgentCore, allowing developers to isolate routing errors and missed procedural steps that generic output metrics might miss.

The tools are available through a Python SDK (requiring Python 3.10+ and Bedrock access) and via the AgentCore CLI. By integrating the evaluators into test suites or deployment gates, teams can diagnose skill selection problems, verify instruction compliance, and enforce critical skill loading without invoking additional models. This supports more maintainable, modular agent development for use cases such as HR assistants or document processing.