awslabs/agent-plugins

AWS Transform custom is a tool for performing code upgrades, migrations, and transformations. It supports any-to-any transformations, including:

• Language version upgrades (Java, Python, Node.js, etc.)
• Framework migrations and library upgrades
• AWS SDK migrations (e.g., Java SDK v1 to v2, boto2 to boto3, JS SDK v2 to v3)
• Code refactoring and architecture changes
• Custom organization-specific transformations defined via natural language, docs, and code samples

Execution is handled in two modes:

• Local mode: Runs the AWS Transform CLI directly on your machine. Best for 1-9 repositories (max 3 concurrent).
• Remote mode: Runs transformations at scale via AWS Batch/Fargate containers. Best for 10+ repositories or when cloud execution is preferred. Infrastructure is auto-deployed with user consent.

The Evaluation Suite provides automated evaluation harnesses for the plugin's skills, generated using the skill-creator. These evals are used to validate database skills (such as Aurora DSQL) by testing triggering, functional correctness, and safety enforcement.

Important Note: Evaluation files are located in tools/evals/ and are not included in the plugin package when it is installed for end-users.

The VMware Migration capability uses generative AI to orchestrate the migration of VMware environments to AWS. It handles the full lifecycle including discovery, migration planning, landing zone setup, network migration, and server rehosting to Amazon EC2 (via AWS MGN).

Key Capabilities:

• VM to EC2: Rehosting/lift-and-shift via MGN (supports Windows and Linux).
• Network Conversion: AI-driven translation of VMware network configs to AWS VPC architecture (VPCs, subnets, security groups, Transit Gateway).
• Migration Planning: AI-driven application grouping, wave planning, and right-sizing.
• Discovery Options: Supports AWS Application Discovery Service (ADS) collectors, Export for vCenter tool, or independent data imports.
• Deployment Models: Supports single-account and multi-account migrations with multi-wave configuration.

The AWS Transform Agent Plugin is an AI-powered service designed to migrate and modernize codebases and workloads to AWS. It follows a workflow of assess, plan, transform, and validate.

It supports several modernization paths:

• Frameworks: .NET Framework to .NET 8/10.
• Mainframe: COBOL to Java.
• Virtualization: VMware VMs to EC2.
• Databases: SQL Server, Oracle, or MySQL to Aurora.
• Languages/SDKs: Upgrades for Java, Python, and Node.js, as well as AWS SDK upgrades.
• Custom: User-defined transformations.

Agent Plugins for AWS provide specialized skills to AI coding agents (currently supporting Claude Code and Cursor) to assist with architectural design, deployment, and operations on AWS.

Instead of pasting long AWS documentation into prompts, plugins package expertise into reusable, versioned artifacts. An agent plugin can contain:

• Agent Skills: Structured workflows and playbooks for complex tasks (e.g., deployment, code review).
• MCP Servers: Connections to external services, data sources, or APIs (e.g., live documentation or pricing data).
• Hooks: Automations and guardrails that run in response to developer actions (e.g., verifying changes).
• References: Documentation, default configurations, and knowledge used by agent skills to reduce prompt bloat.

The AWS Transform Server Migration Agent orchestrates the rehosting of VMware servers to Amazon EC2 using AWS Application Migration Service (MGN). It manages the full wave-based migration lifecycle, including wave setup, inventory validation, replication agent deployment, monitoring, testing, and production cutover.

Key Capabilities:

• Rehost VMware servers to EC2 via MGN.
• Continuous block-level data replication via AWS Replication Agent.
• Automated agent installation via MGN connector (SSH/WinRM).
• Support for multi-wave, single-account, and multi-account migrations.
• Selective or full-wave testing and cutover.

The hyperpod-performance-debugger is a read-only skill designed to diagnose two specific performance scenarios on Amazon SageMaker HyperPod clusters:

1. Uneven NCCL performance across nodes: Workloads running faster on some node sets than others, pairwise bandwidth variance, or suspected straggler nodes.
2. Poor filesystem performance: Training stalled due to data loading, checkpointing dominating step time, or FSx throughput saturation.

Operating Policy: The skill is read-only. It prints commands for the customer to run but does not execute state-modifying commands. It surfaces host-side signals (Xid, ECC, NVLink, EFA reachability, FSx saturation) and routes findings to specialized sibling skills for remediation.

The .NET Modernization Agent (dotnet-chatty-agent) is used to assess and transform .NET applications to newer versions. It supports modernizing applications from .NET Framework (v2.0–v4.8), .NET Core (1.x–3.x), or .NET 5–7 to target .NET 8, .NET 9, or .NET 10.

Supported Transformations:

• Frameworks: .NET Framework/Core/5-7 $\rightarrow$ .NET 8 or .NET 10
• Languages: VB.NET $\rightarrow$ VB.NET on .NET 8/10
• UI/Web: WPF $\rightarrow$ WPF on .NET 8/10, Xamarin $\rightarrow$ .NET MAUI, ASP.NET MVC 5 $\rightarrow$ ASP.NET Core, Web Forms $\rightarrow$ Blazor or Razor Pages
• Services/Data: WCF $\rightarrow$ gRPC or REST, Entity Framework 6 $\rightarrow$ EF Core
• Config/Deployment: Web.config $\rightarrow$ appsettings.json, IIS $\rightarrow$ ECS Fargate / App Runner
• Project Files: packages.config $\rightarrow$ PackageReference (SDK-style .csproj)

The sdk-getting-started skill performs preflight checks to verify that a user's environment is prepared for SageMaker AI operations. It validates the SageMaker Python SDK version, the AWS region, and the execution role.

Instead of generating code, the agent runs these checks directly and stores the results in the conversation context. This context is then used by downstream skills to ensure they have the necessary environment configuration for training, evaluation, or deployment tasks.

When to trigger this skill: Use this skill when a user expresses intent to:

• "set up"
• "get started"
• "check my environment"
• "configure SDK"
• Or as the initial step in any plan involving SageMaker or Bedrock operations.