{
  "markdown": "# AMD Skills\n\n<div align=\"center\">\n\n![AMD](https://img.shields.io/badge/AMD-Skills-ED1C24?logo=amd&logoColor=white)\n![ROCm](https://img.shields.io/badge/ROCm-Enabled-green)\n![Ryzen AI](https://img.shields.io/badge/Ryzen_AI-Ready-1F6FEB)\n![Agent Skills](https://img.shields.io/badge/Agent_Skills-Standard-7B2D8E)\n[![Cursor](https://img.shields.io/badge/Cursor-Compatible-000000?logo=cursor&logoColor=white)](https://cursor.com)\n[![Claude Code](https://img.shields.io/badge/Claude_Code-Compatible-F07535?logo=claude&logoColor=white)](https://www.anthropic.com/claude-code)\n[![OpenAI Codex](https://img.shields.io/badge/OpenAI_Codex-Compatible-412991?logo=data:image/svg+xml;base64,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)](https://openai.com/codex/)\n[![Gemini CLI](https://img.shields.io/badge/Gemini_CLI-Compatible-4285F4?logo=googlegemini&logoColor=white)](https://ai.google.dev/gemini-api/docs)\n[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)\n\n<img src=\"assets/banner.gif\" alt=\"AMD Skills\"/>\n\n[**Browse the Skill Catalog ->**](#the-catalog)\n\n</div>\n\nAMD Skills provide agents with knowledge, scripts, and conventions for working with AMD hardware and software.\n\nSkills in this repository follow the standardized [Agent Skills](https://github.com/anthropics/skills) format and are designed to interoperate with the major coding agents like Cursor, Claude Code, OpenAI Codex, and Gemini CLI.\n\n> [!IMPORTANT]\n> This catalog is being built in the open and will evolve frequently as skills, categories, and descriptions take shape. Some skills may be in Tech Preview; see the underlying product for status.\n\n## Installation\n\nInstall AMD Skills with the [`skills` CLI](https://github.com/vercel-labs/skills) via `npx`. No clone or manual copying required.\n\n```bash\nnpx skills add amd/skills\n```\n\nThis prompts you to pick a skill and an install destination. To install a specific skill into specific agents, pass `--skill` with one or more `--agent` flags (e.g. `cursor`, `claude-code`, `codex`):\n\n```bash\nnpx skills add amd/skills --skill local-ai-use --agent claude-code\n```\n\nBrowse everything available before installing:\n\n```bash\nnpx skills add amd/skills --list\n```\n\n`npx` requires [Node.js](https://nodejs.org). Prefer to do it by hand? See [Manual installation](#manual-installation).\n\n## Using a skill\n\nOnce a skill is installed, reference it in plain language while talking to your agent. For example:\n\n- \"Use AMD Skills to learn how to generate images locally instead of burning cloud tokens.\"\n- \"Use AMD Skills to deploy this LLM for inference on my AMD Instinct GPUs.\"\n\nIn most cases the agent picks the right skill on its own from the description; explicit invocation is a fallback, not a requirement.\n\nFor hands-on, step-by-step guides that show a skill in action, see the [walkthroughs](walkthroughs/README.md).\n\n## The catalog\n\nThe initial catalog is organized into three focus areas, spanning the full stack from client to cloud. This catalog is expected to grow significantly as more skills land.\n\n### Client-native\n\nRun and optimize on Ryzen AI.\n\n| Skill | What it does | Source |\n| --- | --- | --- |\n| [`local-ai-use`](https://github.com/amd/skills/blob/main/skills/local-ai-use/SKILL.md) | Route image generation, text-to-speech, and speech-to-text through a local AI server to reduce token cost. | in-repo |\n| [`local-ai-app-integration`](https://github.com/amd/skills/blob/main/skills/local-ai-app-integration/SKILL.md) | Integrate local AI into cloud LLM apps for offline support, better privacy, and lower API costs. | in-repo |\n\n### Cross-stack\n\nCross-stack skills, from client to cloud.\n\n| Skill | What it does | Source |\n| --- | --- | --- |\n| [`rocm-doctor`](https://github.com/amd/skills/blob/main/staging/rocm-doctor/SKILL.md) | Diagnose ROCm, HIP, PyTorch, or llama.cpp failures on AMD GPUs (Linux and Windows) against a closed list of known misconfigurations, then fix with consent or route upstream. Thin driver over the `rocm` CLI (`examine`, `diagnose`, or `fix`). | _planned_ |\n| [`lemonade-router-builder`](https://github.com/amd/skills/blob/main/skills/lemonade-router-builder/SKILL.md) | Set up a Lemonade model router that handles requests based on content, sensitivity, or required capabilities. | in-repo |\n| `hrr-replay-analysis` | Record, replay, and analyze GPU workload behavior on ROCm across AMD Instinct, Radeon, and Ryzen hardware using HIP Record and Replay archives. | _planned_ |\n\n### Server-native\n\nRun and optimize on AMD Instinct.\n\n| Skill | What it does | Source |\n| --- | --- | --- |\n| [`serving-llms-on-instinct`](https://github.com/amd/skills/blob/main/skills/serving-llms-on-instinct/SKILL.md) | Deploy LLM inference on AMD Instinct GPUs end-to-end: detect hardware (or onboard via AMD Developer Cloud), validate model fit, apply the right vLLM recipe, and launch a benchmarked endpoint. SGLang and engine or backend selection in later phases. | in-repo |\n| [`serving-llms-on-epyc`](https://github.com/amd/skills/blob/main/skills/serving-llms-on-epyc/SKILL.md) | Serve LLMs on AMD EPYC CPUs with vLLM and zentorch, in a container (Docker or Podman) or conda. Handles CPU detection, runtime and env validation, vLLM model-support and RAM-fit checks, hardware-sized threads and KV, launch, and health verification. Single instance; reports and stops on failure. | in-repo |\n| [`hyperloom-workload-optimizer`](https://github.com/amd/skills/blob/main/skills/hyperloom-workload-optimizer/SKILL.md) | Set up Hyperloom and autonomously optimize end-to-end LLM inference throughput on AMD Instinct GPUs, reporting a validated gain. | in-repo |\n| [`magpie-kernel-evaluator`](https://github.com/amd/skills/blob/main/skills/magpie-kernel-evaluator/SKILL.md) | Evaluate GPU kernel correctness and performance, compare kernel implementations, and benchmark vLLM or SGLang inference with profiling, TraceLens, and torch-trace gap analysis. | [Magpie](https://github.com/AMD-AGI/Magpie) |\n| [`tracelens-analysis-orchestrator`](https://github.com/amd/skills/blob/main/skills/tracelens-analysis-orchestrator/SKILL.md) | Orchestrate modular PyTorch profiler trace analysis with TraceLens: generate perf reports, run system-level and compute-kernel subagents in parallel, and write a prioritized stakeholder report. | [TraceLens](https://github.com/AMD-AGI/TraceLens) |\n\n## What is a skill?\n\nA skill is a self-contained folder that bundles everything an agent needs to perform a focused task: instructions, helper scripts, prompts, templates, and references. At its core is a `SKILL.md` file with YAML frontmatter, a `name`, and a short `description` that tells the agent *when* the skill should activate, followed by the guidance the agent reads while the skill is in use.\n\n```\nskills/\n  <skill-name>/\n    SKILL.md\n    skill-card.md\n    scripts/       # optional\n    references/    # optional\n```\n\nWhen an agent decides a skill is relevant (or you invoke it explicitly), it loads that `SKILL.md` and follows the instructions inside. Descriptions stay in context cheaply; the full body of a skill only loads when the task actually matches.\n\nEvery skill also ships a `skill-card.md`: a short, human-facing governance card (Description, Owner, License) that tells a reviewer what the skill is and who stands behind it without reading the source. See [docs/skill-requirements.md](https://github.com/amd/skills/blob/main/docs/skill-requirements.md#skill-cardmd).\n\n## Why a skill, not a doc?\n\nDocumentation describes an API surface: every flag, every option, neutral by design. A skill encodes the opinionated path: which flags, which container image, which `gfx` target, which environment variables, in what order. It captures the decisions a senior AMD engineer makes without thinking, in a form the agent can apply consistently across teams and repositories.\n\nSkills earn their keep on repeated, opinionated workflows, exactly where the AMD stack lives.\n\n\n## Catalog federation\n\nThe AMD stack is large and moves fast. ROCm, HIP, Ryzen AI, and framework integrations each have their own team, release cadence, and validation matrix. So skills here are **federated**: each skill is owned and versioned by the team that owns the product it describes, and this repository is the catalog that brings them together.\n\n```\n                ┌─────────────────────────────────────────────────────┐\n                │                amd/skills (this repo)               │\n                │                                                     │\n                │   skills/         .github/         .*-plugin/       │\n                │   vendored copies federation.json  agent manifests  │\n                └──────────────────────┬──────────────────────────────┘\n                                       │  one install\n                                       ▼\n                              your AI coding agent\n                                       ▲\n                                       │  resolves pointers to\n       ┌───────────────┬───────────────┼───────────────┬────────────────┐\n       │               │               │               │                │\n   ROCm/ROCm       ROCm/HIP        Ryzen AI repo   lemonade-sdk    ...more\n  rocm-doctor/    cuda-to-hip/    ryzen-ai-tools/   local-ai-app-   product\n   gfx-target-...  triton-amd-...  ...               integration/    repos\n```\n\n[`.github/federation.json`](.github/federation.json) is the whole registry: each\nentry names a source repo and the exact path of every skill folder to vendor\nfrom it. Sources are tracked at `main` only, so nothing reaches users that the\nowning team has not already merged.\n\nThe `federate-skills` workflow runs nightly and on demand. It clones each\ndeclared repo, compares a content hash of the upstream skill folder against the\nhash recorded in the vendored copy's `.federated.json`, and re-vendors only the\nskills that actually changed. When something did change it regenerates the agent\nmanifests and opens a pull request titled `Bump <skill> to <short commit>`,\nwhere the usual `validate` checks apply as they would to any other pull\nrequest. A quiet night produces no diff and therefore no pull request, so every\nbump that lands is a reviewed commit.\n\nA vendored skill mirrors its upstream folder with one exception: for now\nfederation does not carry the skill's `evals/` folder in either direction, so\nthe datasets this repo grades skills against live and are maintained here.\n\nSee [CONTRIBUTING.md](CONTRIBUTING.md) to register a repo.\n\n## Manual installation\n\nUntil marketplace integration lands, install skills manually: clone this repo, then copy (or symlink) the skill folders you want from `skills/` into your agent's skills directory. Each agent discovers `SKILL.md` automatically.\n\n```bash\ngit clone https://github.com/amd/skills.git amd-skills\ncp -r amd-skills/skills/local-ai-use <agent-skills-dir>/\n```\n\n| Agent | Skills directory (personal / project) |\n| --- | --- |\n| Cursor | `~/.cursor/skills/` or `.cursor/skills/` |\n| Claude Code | `~/.claude/skills/` or `.claude/skills/` |\n| Codex | `$HOME/.agents/skills` or `$REPO_ROOT/.agents/skills` |\n\n## Contributing\n\nContributions are welcome from AMD engineers and selected partners.\n\nSee [CONTRIBUTING.md](CONTRIBUTING.md) for the step-by-step instructions, then\n[docs/skill-requirements.md](docs/skill-requirements.md) for the rules CI enforces,\n[docs/best-practices.md](docs/best-practices.md) for writing guidance, and\n[docs/evals.md](docs/evals.md) for how skills are graded.\n\n## License\n\nReleased under the MIT License. See [LICENSE](https://github.com/amd/skills/blob/main/LICENSE) for details.\n\nCopyright (C) 2026 Advanced Micro Devices, Inc. All rights reserved.\n",
  "bytes": 13430,
  "sha": "dbf77823011e28717849aae3d10aab2e71fd5f4e448e2985e1ba291944da9150",
  "repo_slug": "amd/skills",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_amd_skills_amd_skills_f4152ed9/readme"
}