- Adds a secret safety skill to the aws-core plugin in the Agent Toolkit for AWS, letting agents use Secrets Manager without exposing raw secret values.
- Implements a two-layer approach that prevents the model from handling plaintext secrets and resolves them only in a child process at execution time.
- Available now for all supported agent harnesses (Claude Code, Codex, Cursor) across AWS regions; install via the aws-core plugin and refer to the documentation.