amd/skills

What it solves

AMD Skills provides AI coding agents with the specific knowledge, scripts, and operational conventions required to work effectively with AMD hardware and software. It bridges the gap between general AI capabilities and the highly specialized, opinionated workflows needed to optimize AI workloads on Ryzen AI, ROCm, and AMD Instinct GPUs.

How it works

The project uses a standardized "skill" format—self-contained folders containing a SKILL.md file (with YAML metadata and instructions), helper scripts, and references. These skills are designed to be integrated into agents like Cursor, Claude Code, OpenAI Codex, and Gemini CLI.

To maintain accuracy across a vast hardware stack, the project uses a federated model: skills are owned and versioned by the product teams (e.g., ROCm or Ryzen AI) in their own repositories and then automatically vendored into this central catalog via a nightly synchronization process.

Who it’s for

Developers and engineers using AI coding agents who need to deploy, optimize, or diagnose AI models and workloads on AMD hardware, ranging from client-side Ryzen AI to server-side Instinct GPUs.

Highlights

• Agent Interoperability: Compatible with major coding agents including Cursor, Claude Code, Codex, and Gemini CLI.
• Federated Catalog: Skills are maintained by the original product teams to ensure they stay current with hardware releases.
• Full-Stack Coverage: Includes skills for client-native (Ryzen AI), cross-stack (ROCm), and server-native (AMD Instinct/EPYC) environments.
• Opinionated Workflows: Unlike general documentation, skills encode the specific paths and decisions a senior AMD engineer would take to achieve a result.