The Agent Skills Directory

Command Execution: The skill orchestrates its workflow by using Python's subprocess.run to invoke its own local scripts (e.g., calculate_costs.py, deploy_model.py, benchmark_model.py). This allows for modular analysis and benchmarking. These commands are invoked using specific file paths and do not use shell interpolation for user-provided data.
Dynamic Execution: For live validation, the skill dynamically assembles a Lambda deployment package. This process includes vendoring local versions of boto3 and botocore into a zip archive to ensure compatibility with specific DynamoDB features like vector search. This is a targeted implementation detail for benchmarking and does not involve executing arbitrary remote code.
AWS Resource Management: The skill possesses the capability to create and delete real AWS resources in the user's account. To prevent unintended use in sensitive environments, the deployment scripts include built-in safeguards that check for production keywords in account aliases or ARNs. Furthermore, the skill requires explicit user consent before proceeding with any deployment and generates a standalone teardown.sh script for the user to review and execute for cleanup.
Authorization Integrity: The skill's instructions emphasize the importance of aligning partition keys with authorization boundaries (IDOR prevention) as a primary security check for any design it reviews, promoting secure architectural patterns.

amazon-dynamodb