AMD Skills

AMD's own plugin of Agent Skills for working with AMD hardware and software. Each skill gives your agent the path an AMD engineer would take, with the right flags, containers, targets and settings in order, and loads only when a request matches. What its skills teach your agent - local-ai-use: generate images, transcribe audio and read text aloud on your own machine through a local Lemonade Server instead of a paid cloud API, for one request or from now on - local-ai-app-integration: add a local AI mode to an existing app with Embeddable Lemonade, for offline use, privacy or lower API costs, using the NPU, iGPU or CPU - serving-llms-on-instinct: serve a model with vLLM on AMD Instinct GPUs (MI300X, MI325X, MI350X, MI355X), from GPU detection and environment checks to launch and a health check - hyperloom-workload-optimizer: set up Hyperloom and optimize LLM inference throughput with vLLM or SGLang on Instinct GPUs, reporting a validated gain - tracelens-analysis-orchestrator: analyze a PyTorch profiler trace with TraceLens, or compare two, and write a prioritized performance report How to install it In Claude Code it installs from Anthropic's official plugin marketplace as amd-skills. AMD also publishes each skill on its own: npx skills add amd/skills lists the full catalog, including skills this plugin doesn't bundle such as rocm-doctor, and installs them in Claude Code, Cursor, Codex and more. Good to know Some skills need the hardware or software they cover: an AMD Instinct GPU host with ROCm and Docker (on the machine or over SSH) for serving and Hyperloom, and Lemonade Server, which the local AI skill installs if it's missing. AMD builds the catalog in the open, so skills change often and some are in Tech Preview.