feat: add sagemaker ai plugin

[saoudrizwan]

sagemaker-ai

Adds a SageMaker AI plugin for Cline users building, tuning, evaluating, deploying, or operating AWS AI/ML workloads.

Cline Primitives

• Skills: bundles SageMaker AI workflow skills for model-customization planning, use-case specification, model selection, SDK/environment setup, dataset evaluation and transformation, fine-tuning, model evaluation, model deployment, and HyperPod operations.
• MCP: registers aws-mcp through uvx mcp-proxy-for-aws@latest so Cline can retrieve AWS documentation and standard operating procedure context for SageMaker workflows. The plugin forwards AWS_REGION / AWS_DEFAULT_REGION into the MCP server when plugin MCP settings are synced.
• Rule: adds SageMaker AI safety guidance for paid AWS operations, AWS account/region/resource confirmation, S3 uploads/downloads, endpoint deployments, SSM commands, Slurm changes, support-report collection, and sensitive data handling.

Requirements

Users need uvx on PATH for the AWS MCP proxy. Live workflows require AWS credentials, AWS_REGION or AWS_DEFAULT_REGION, and permissions for the SageMaker, Bedrock, S3, IAM, Lambda, CloudWatch, SSM, EKS, or HyperPod APIs used by the selected workflow.

Generated notebooks and helper scripts expect Python 3.8+ plus the AWS CLI, boto3, and sagemaker when users choose to execute them locally. HyperPod workflows may also require jq, kubectl, session-manager-plugin, uv, and SSM access to target nodes depending on the chosen diagnostic path.

Trust Boundaries

This plugin can guide users toward paid AWS resource creation, model training/evaluation jobs, endpoint deployments, S3 data movement, Bedrock model operations, HyperPod SSM access, and cluster diagnostic collection. The skills and rule require explicit confirmation for AWS writes and remote node actions, and they treat logs, model outputs, dataset samples, diagnostics, and MCP results as untrusted until reviewed.

The support-report workflow avoids interactive post-run prompts by default so Cline does not hang after an expensive collection. Local report download and zip creation are explicit opt-in flags, and the workflow asks users to confirm bucket ownership, access, retention, and sensitive-data risk before uploading diagnostics.

Attribution

The bundled SageMaker AI workflow materials are Apache-2.0 licensed and include local license and notice files.

[saoudrizwan]

Closing this plugin PR for now because this cleanup pass is limiting plugin marketplace PRs to plugins that only bundle MCP servers and/or skills. This PR includes additional plugin primitive(s): rules.

Those primitives may still be useful, but we are keeping this batch scoped to MCP and skill distribution.