The Agent Skills Directory

[Data Ingestion and Processing]: The skill interacts with user-specified datasets in S3 or local storage to facilitate training and evaluation. While this introduces an ingestion surface for potential indirect prompt injection, the skill implements format validation through a dedicated format_detector.py script and adheres to standard SageMaker and Bedrock evaluation schemas.
[Controlled Code Execution]: The agent is instructed to generate and execute Python code for tasks like dataset transformation and environment verification. This behavior is restricted to the specific context of model customization and includes built-in safeguards, such as checking for existing jobs before re-execution to prevent redundant or unintended operations.
[Infrastructure Management]: The skill manages AWS resources like SageMaker endpoints, MLflow apps, and IAM roles using official AWS APIs and standard tools. It provides helpful guidance on IAM prerequisites and requires explicit user confirmation before destructive actions like deleting MLflow applications.
[Verified Dependency Usage]: The skill relies on well-known Python packages such as sagemaker, boto3, and mlflow. These are standard libraries for AWS environment interaction and are managed through version-pinned installation commands to ensure compatibility and stability.

aws-ai-ml