Perhaps we are a little late on this, but we've published a portable Agent Skills package for Proxylity UDP Gateway. It's available now on GitHub at github.com/proxylity/skills. It's free, open source (MIT), and compatible with every major AI coding tool that supports the Agent Skills format.
If you use GitHub Copilot, Claude Code, Cursor, Gemini CLI, or any of the dozens of other agents that support Agent Skills, you can point it at this package and immediately get a well-informed collaborator for designing, implementing, deploying, and troubleshooting UDP Gateway solutions on AWS.
What Are Agent Skills?
Agent Skills are an open format for giving AI
agents specialized knowledge and workflows. At its core, a skill is simply a folder containing a
SKILL.md file with metadata and instructions. Skills can also bundle scripts,
reference materials, templates, and other resources that the agent loads on demand.
The format was originally developed by Anthropic and has since been adopted across a wide range of AI tools. Because it's a plain file format (no proprietary APIs, no vendor lock-in) a skill written once works across every compatible agent. The client showcase lists the currently supported tools.
Skills load using progressive disclosure: an agent reads only the name and description at startup (a tiny context footprint), then loads the full instructions only when a task matches. This means you can keep many skills available without worrying about context inflation.
The udp-gateway Skill
The repository contains one skill for now: udp-gateway/. It guides an agent through the
complete workflow for building with Proxylity UDP Gateway — from initial requirements gathering to
production deployment.
Specifically, the skill covers:
- Identifying use cases where Proxylity UDP Gateway is the right solution
- Choosing the right architecture pattern for your use case (Lambda, Step Functions, direct destination, or composite)
- Generating CloudFormation and SAM templates for Listeners, Destinations, IAM roles, and Client Restrictions
- Writing packet handler code for Lambda and Step Functions, including the JSON packet request/response format
- Configuring direct destinations: EventBridge, SQS, Kinesis, DynamoDB, S3, Firehose, CloudWatch Logs, API Gateway, and IoT Core
- WireGuard setup and encryption, including Open Endpoints and Decapsulated Delivery
- CoAP and syslog packet handling with Proxylity's built-in formatters
- Batching configuration to reduce API call volume and costs
- Packet Sources for server-initiated delivery to connected clients
- Multi-region deployment patterns and cross-region failover
- Deploying, testing, and troubleshooting end-to-end
The skill also includes reference materials in the references/ subfolder — covering
the CloudFormation resource schemas, the JSON packet format, BREX syntax, and other details the
agent can pull in when needed — as well as example CloudFormation snippets and Lambda handler
stubs in assets/.
Getting Started
Installation follows your agent's own skill setup instructions. The general pattern is to clone
(or reference) the udp-gateway/ folder and point your agent configuration at it.
The repository README links to the setup
guide for each supported tool.
For example, with GitHub Copilot in VS Code you'd add the skill via the agent customization
settings. With Claude Code you'd reference the folder in your project's
.claude/skills/ directory. Cursor, Gemini CLI, and most other tools follow similar
patterns — check the client showcase for links to
each tool's skill installation docs.
Once installed, just describe what you want to build. The agent will recognize when the task should make use of UDP Gateway and load the full skill instructions. You can also invoke it directly: ask your agent to "use the udp-gateway skill to help me set up a CoAP listener with a Lambda destination" and it will follow the skill's guided workflow.
Requirements
To build and deploy using the skill, you'll need:
- An AWS account subscribed to Proxylity UDP Gateway on AWS Marketplace
- AWS CLI configured with your account credentials
- SAM CLI (only required for compiled Lambda runtimes: .NET NativeAOT, Go, Java, Rust)
Why a Skill Instead of Documentation Links?
You might wonder how this is different from an agent that can already browse documentation. The difference is precision and reliability. Documentation pages are written for humans to read sequentially; they assume context that accumulates across pages. Agents often miss that context when pulling in random pages mid-task.
A skill is written specifically for agents: it encodes the decision tree, the common failure modes, the ordering constraints (like needing an IAM policy before a CloudFormation resource), and the patterns that tend to produce working solutions on the first try. It's the difference between handing someone a manual and working alongside someone who's done the job before.
We'll keep the skill updated as the platform evolves. When we add a new feature, the skill
will be updated alongside the documentation. Cloning or referencing a specific commit pins you to
a stable version; tracking main keeps you current.
Open Source and Contributions
The repository is MIT-licensed. If you find something missing — a pattern the skill doesn't handle well, a reference document that would help, an example that would save someone an hour of troubleshooting — pull requests are welcome. The goal is to make the skill as useful as possible for the community.
Related Resources
- github.com/proxylity/skills — the Agent Skills repository
- github.com/proxylity/examples — canonical working templates and handlers
- agentskills.io — the Agent Skills format specification and client showcase
- Proxylity Documentation — full reference documentation