The Agent Skills Directory

[COMMAND_EXECUTION]: The skill scaffolds shell scripts (package_model.sh, test_local.sh) and Python code that use subprocess and CLI tools (like aws, tar, and docker) to perform model packaging, S3 uploads, and local container testing.
[PRIVILEGE_ESCALATION]: The generated Dockerfile templates run as the root user. This is a documented requirement of the SageMaker Marketplace contract to ensure that model artifacts can be properly mounted and accessed at /opt/ml/model/ during deployment.
[INDIRECT_PROMPT_INJECTION]: The skill performs an inventory phase where it reads existing local files (dependency manifests, Dockerfiles, and entry points). These contents are used to guide the walkthrough, representing a surface for instructions in processed data to influence the agent, although the skill mitigates this by requiring explicit user confirmation before moving between phases.

sagemaker-marketplace-onboarding