SageMaker AI Agent Plugin

One portable package for building, training, deploying, monitoring, and operating machine-learning workloads on Amazon SageMaker AI.

GitHub stars License: MIT Agent Plugin

Codex · Kiro · Claude Code

What it covers

Area	Capabilities
SDK and workflows	SageMaker Python SDK v3, v2 migration, Pipelines, processing, HPO
Training	Classical ML, SFT, LoRA, QLoRA, DPO, CPT, RLVR, RLAIF, Trainium
Inference	Real-time, batch, JumpStart, DJL LMI, vLLM, HyperPod
Operations	HyperPod with EKS or Slurm, Model Monitor, AutoGluon
Iteration	Managed warm pools for repeated training and experimentation

The package uses one canonical skill under skills/sagemaker-ai/. IDE-specific manifests only handle discovery and installation.

Install

Codex

Add this repository as a plugin marketplace:

codex plugin marketplace add dgallitelli/sagemaker-ai-agent-plugin
codex plugin add sagemaker-ai@dgallitelli-sagemaker-ai

Restart Codex to load the plugin. Alternatively, run /plugins, open the SageMaker AI Plugins marketplace, and install sagemaker-ai interactively.

The Codex package includes the SageMaker skill and the optional official AWS Labs SageMaker AI MCP server configuration.

Claude Code

Add the marketplace and install the plugin:

claude plugin marketplace add dgallitelli/sagemaker-ai-agent-plugin
claude plugin install sagemaker-ai@dgallitelli-sagemaker-ai

Restart Claude Code after installation. You can inspect or update it later with:

claude plugin list
claude plugin update sagemaker-ai@dgallitelli-sagemaker-ai

Kiro

Open the Powers panel in Kiro.
Select Add Custom Power.
Choose Import power from GitHub.

Enter:

https://github.com/dgallitelli/sagemaker-ai-agent-plugin

Install the Power, then enable and trust it when prompted.

For a skill-only installation, import this folder instead:

https://github.com/dgallitelli/sagemaker-ai-agent-plugin/tree/main/skills/sagemaker-ai

Manual skill-only installation

Clients that support Agent Skills can link the canonical skill directly:

git clone https://github.com/dgallitelli/sagemaker-ai-agent-plugin.git
cd sagemaker-ai-agent-plugin

Choose the destination for your client:

# Codex
mkdir -p ~/.codex/skills
ln -s "$PWD/skills/sagemaker-ai" ~/.codex/skills/sagemaker-ai

# Claude Code
mkdir -p ~/.claude/skills
ln -s "$PWD/skills/sagemaker-ai" ~/.claude/skills/sagemaker-ai

# Kiro
mkdir -p ~/.kiro/skills
ln -s "$PWD/skills/sagemaker-ai" ~/.kiro/skills/sagemaker-ai

Package layout

.
├── plugin.json                       # Agent Plugins / Kiro
├── mcp.json                          # Agent Plugins MCP definition
├── .mcp.json                         # Shared Codex / Claude MCP definition
├── .agents/plugins/marketplace.json  # Codex marketplace
├── .codex-plugin/                    # Codex manifest
├── .claude-plugin/                   # Claude Code manifest and marketplace
└── skills/sagemaker-ai/
    ├── SKILL.md                      # Routing and operating rules
    ├── references/                   # Detailed guidance
    ├── scripts/                      # Reusable utilities
    ├── templates/                    # Training and inference templates
    └── assets/                       # HyperPod examples

Optional AWS MCP server

The plugin configures the official AWS Labs package:

awslabs.sagemaker-ai-mcp-server@latest

It is intentionally configured without write or sensitive-data flags. The server currently focuses on SageMaker HyperPod operations; the skill uses the AWS CLI, boto3, and SageMaker Python SDK v3 for other workflows.

The skill remains usable when the MCP server is unavailable.

Requirements

Python 3.10–3.13 for SageMaker Python SDK v3 workflows
AWS CLI with configured credentials
uv/uvx when using the optional MCP server

Consolidated projects

This plugin brings together capabilities previously spread across:

claude-code-skill-for-sagemaker-ai
sagemaker-python-sdk-skill
aws-hyperpod-skill
kiro-power-for-sagemaker-ai
the original LLM-training skill in this repository

It does not include or depend on dgallitelli/sagemaker-ai-mcp-server.

The focused sagemaker-warm-pool-researcher remains independently installable, while its capabilities are also available inside this plugin.

License

MIT. HyperPod-derived material retains its Apache-2.0 terms; see THIRD_PARTY_NOTICES.md.