Skip to content
← Tools / Skills
ToolAMD/10 trustMetrics-checked

AMD Skills

Official AMD agent skills for ROCm, Instinct, and EPYC AI work.

Visit official site →

About

AMD Skills is AMD's official catalog of AI agent skills, released under the MIT license. It equips AI agents with workflows for AMD's optimized software stack: serving LLMs on EPYC CPUs and Instinct accelerators, PTQ quantization of Torch LLMs with Quark, ROCm environment diagnosis with rocm-doctor, local AI app integration, and workload optimization with HyperLoom and Magpie kernel evaluation.

Skills install via the skills CLI and work across Claude Code, Cursor, Codex, and OpenCode. The catalog also accepts product-repo skills authored in AMD product repositories and registered through a sources manifest with pinned tags, so skills version alongside the products they describe.

Honest limitations: this is deeply AMD-hardware-centric. If your agents run on NVIDIA GPUs or cloud TPUs, almost nothing here applies. The catalog is relatively new and skill quality varies across the stack, and some skills assume ROCm familiarity that the skill itself cannot fully teach.

What it can do

  • Official AMD-maintained skill catalog
  • LLM serving guides for EPYC and Instinct
  • ROCm setup and troubleshooting skills
  • LLM quantization with Quark
  • Cross-agent install via skills CLI

Setup

Install with the skills CLI: npx skills add amd/skills, or clone and copy the skill folders you want into your agent's skills directory.

Prerequisites

A compatible agent (Claude Code, Cursor, Codex, OpenCode); AMD GPU/CPU hardware for most skills to be useful.

Use cases

agent-skillsgpu-computingllm-servingrocm

Works with

Trust score — 7/10

/10 trustMetrics-checked

Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 88; Trending score 74/100 from researcher assessment.

★ 417 GitHub stars

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.