skills: Give AI Agents AMD Workflow Guidance

Overview

AMD Skills is a catalog of agent-ready workflows for AMD software and hardware. Each skill packages focused instructions and, where needed, scripts or references so an AI coding agent can follow an opinionated workflow rather than rely on general documentation alone.

It is intended for developers using coding agents across AMD's client-to-cloud stack. Use it when you want an agent to assist with a supported AMD task, such as running local AI or deploying inference, and check the individual skill for its scope and product status. The catalog is evolving, and some entries are planned or in tech preview.

Key Features

• Follows the Agent Skills format, with a SKILL.md describing when and how an agent should apply a skill.
• Provides workflows spanning Ryzen AI, ROCm, AMD Instinct GPUs, and EPYC CPUs.
• Includes skills for local AI integration, LLM serving, quantization, troubleshooting, and performance analysis.
• Supports installation through the skills CLI, with options to choose skills and target agents.
• Targets compatible coding agents including Cursor, Claude Code, OpenAI Codex, and Gemini CLI.
• Federates selected skills from product repositories, with upstream changes synchronized through a repository workflow.

Use Cases

• An application developer wants an agent to route image, speech, or text-generation tasks to a local AI server instead of a cloud API.
• An infrastructure engineer needs guidance deploying an LLM on AMD Instinct GPUs or EPYC CPUs.
• A PyTorch user wants an agent-assisted workflow for quantizing a model with AMD Quark.
• A developer diagnosing a ROCm or HIP issue wants a skill for checking known configuration problems, where the relevant skill is available.
• A performance engineer wants help analyzing profiler traces or comparing GPU kernel behavior using the catalog's supported tools.

Getting Started

Install the catalog with Node.js and the skills CLI:

npx skills add amd/skills

See the repository README for selecting specific skills, manual installation, and catalog details.

Alternatives

• awesome-agent-skills: An index of skills from many providers and for many workflows, rather than AMD-focused guidance for hardware and software.
• skills: Provides portable skills for .NET and C# workflows instead of guidance for AMD hardware, ROCm, and LLM serving.
• skills: Focuses on general engineering practices such as TDD and debugging, rather than AMD-specific tools and workflows.

Considerations

• This is a growing catalog, not a general-purpose agent or standalone AMD runtime. The utility depends on whether a skill covers your specific task and product version.
• The README notes that some skills may be in tech preview; check the underlying product's status and each skill's documentation before relying on it.
• Installation through npx requires Node.js. Skills that operate on AMD software or hardware may have additional requirements described in their own instructions.
• The repository reports a Python language, but the supplied installation path uses the Node.js-based skills CLI.