Add ROCM kernel skill

docs/source/cli-skills.md

@@ -1,6 +1,11 @@
 ### kernels skills add
 
-Use `kernels skills add` to install the skills for AI coding assistants like Claude, Codex, and OpenCode. For now, only the `cuda-kernels` skill is supported. Skill files are downloaded from the `huggingface/kernels` directory in this [repository](https://github.com/huggingface/kernels/tree/main/skills).
+Use `kernels skills add` to install the skills for AI coding assistants like Claude, Codex, and OpenCode.
+Supported skills include:
+- `cuda-kernels` (default)
+- `rocm-kernels`
+
+Skill files are downloaded from the `huggingface/kernels` directory in this [repository](https://github.com/huggingface/kernels/tree/main/skills).
 
 Skills instruct agents how to deal with hardware-specific optimizations, integrate with libraries like diffusers and transformers, and benchmark kernel performance in consistent ways.
 
@@ -10,6 +15,9 @@ Examples:
 # install for Claude in the current project
 kernels skills add --claude
 
+# install ROCm kernels skill for Codex
+kernels skills add --skill rocm-kernels --codex
+
 # install globally for Codex
 kernels skills add --codex --global
 
@@ -18,6 +26,9 @@ kernels skills add --claude --codex --opencode
 
 # install to a custom destination and overwrite if already present
 kernels skills add --dest ~/my-skills --force
+```
+
 ### Create a new kernel project
+```bash
 kernels init my-username/my-kernel --skills ~/my-skills
 ```

kernels/src/kernels/cli/__init__.py

@@ -4,7 +4,7 @@
 import sys
 from pathlib import Path
 
-from kernels.cli.skills import add_skill
+from kernels.cli.skills import DEFAULT_SKILL_ID, SUPPORTED_SKILL_IDS, add_skill
 from kernels.cli.upload import upload_kernels_dir
 from kernels.cli.versions import print_kernel_versions
 from kernels.compat import tomllib
@@ -98,7 +98,17 @@ def main():
     skills_subparsers = skills_parser.add_subparsers(required=True)
     skills_add_parser = skills_subparsers.add_parser(
         "add",
-        help="Install the cuda-kernels skill for an AI assistant",
+        help="Install a kernels skill for an AI assistant",
+    )
+    skills_add_parser.add_argument(
+        "--skill",
+        type=str,
+        choices=SUPPORTED_SKILL_IDS,
+        default=DEFAULT_SKILL_ID,
+        help=(
+            f"Skill ID to install. Defaults to `{DEFAULT_SKILL_ID}`. "
+            "Use `rocm-kernels` for ROCm-focused kernels."
+        ),
     )
     skills_add_parser.add_argument(
         "--claude",

kernels/src/kernels/cli/skills.py

@@ -6,11 +6,11 @@
 from huggingface_hub.utils import get_session
 
 DEFAULT_SKILL_ID = "cuda-kernels"
-_GITHUB_RAW_BASE = (
-    "https://raw.githubusercontent.com/huggingface/kernels/main/" "skills/cuda-kernels"
+SUPPORTED_SKILL_IDS = ("cuda-kernels", "rocm-kernels")
+_GITHUB_RAW_BASE_TEMPLATE = (
+    "https://raw.githubusercontent.com/huggingface/kernels/main/skills/{skill_id}"
 )
-_MANIFEST_URL = f"{_GITHUB_RAW_BASE}/manifest.txt"
-_LOCAL_SKILLS_ROOT = Path(__file__).resolve().parents[4] / "skills" / "cuda-kernels"
+_LOCAL_SKILLS_DIR = Path(__file__).resolve().parents[4] / "skills"
 
 GLOBAL_TARGETS = {
     "codex": Path("~/.codex/skills"),
@@ -31,12 +31,24 @@ def _download(url: str) -> str:
     return response.text
 
 
-def _download_manifest() -> list[str]:
+def _github_raw_base(skill_id: str) -> str:
+    return _GITHUB_RAW_BASE_TEMPLATE.format(skill_id=skill_id)
+
+
+def _manifest_url(skill_id: str) -> str:
+    return f"{_github_raw_base(skill_id)}/manifest.txt"
+
+
+def _local_skill_root(skill_id: str) -> Path:
+    return _LOCAL_SKILLS_DIR / skill_id
+
+
+def _download_manifest(skill_id: str) -> list[str]:
     entries: list[str] = []
     try:
-        raw_manifest = _download(_MANIFEST_URL)
+        raw_manifest = _download(_manifest_url(skill_id))
     except Exception:
-        local_manifest = _LOCAL_SKILLS_ROOT / "manifest.txt"
+        local_manifest = _local_skill_root(skill_id) / "manifest.txt"
         if not local_manifest.exists():
             raise
         raw_manifest = local_manifest.read_text(encoding="utf-8")
@@ -49,11 +61,11 @@ def _download_manifest() -> list[str]:
     return entries
 
 
-def _download_file(rel_path: str) -> str:
+def _download_file(skill_id: str, rel_path: str) -> str:
     try:
-        return _download(f"{_GITHUB_RAW_BASE}/{rel_path}")
+        return _download(f"{_github_raw_base(skill_id)}/{rel_path}")
     except Exception:
-        local_file = _LOCAL_SKILLS_ROOT / rel_path
+        local_file = _local_skill_root(skill_id) / rel_path
         if local_file.exists():
             return local_file.read_text(encoding="utf-8")
         raise
@@ -67,10 +79,10 @@ def _remove_existing(path: Path) -> None:
     shutil.rmtree(path)
 
 
-def _install_to(target: Path, force: bool) -> Path:
+def _install_to(target: Path, force: bool, skill_id: str) -> Path:
     target = target.expanduser().resolve()
     target.mkdir(parents=True, exist_ok=True)
-    dest = target / DEFAULT_SKILL_ID
+    dest = target / skill_id
 
     if dest.exists():
         if not force:
@@ -79,8 +91,8 @@ def _install_to(target: Path, force: bool) -> Path:
             )
         _remove_existing(dest)
 
-    for rel_path in _download_manifest():
-        content = _download_file(rel_path)
+    for rel_path in _download_manifest(skill_id):
+        content = _download_file(skill_id, rel_path)
         output_file = dest / rel_path
         output_file.parent.mkdir(parents=True, exist_ok=True)
         output_file.write_text(content, encoding="utf-8")
@@ -89,6 +101,11 @@ def _install_to(target: Path, force: bool) -> Path:
 
 
 def add_skill(args: Namespace) -> None:
+    skill_id = getattr(args, "skill", DEFAULT_SKILL_ID)
+    if skill_id not in SUPPORTED_SKILL_IDS:
+        supported = ", ".join(SUPPORTED_SKILL_IDS)
+        raise SystemExit(f"Unsupported skill '{skill_id}'. Supported skills: {supported}")
+
     if not (args.claude or args.codex or args.opencode or args.dest):
         print(
             "Pick a destination via --claude, --codex, --opencode, or --dest.",
@@ -109,5 +126,5 @@ def add_skill(args: Namespace) -> None:
         targets.append(args.dest)
 
     for target in targets:
-        installed_path = _install_to(target, force=args.force)
-        print(f"Installed '{DEFAULT_SKILL_ID}' to {installed_path}")
+        installed_path = _install_to(target, force=args.force, skill_id=skill_id)
+        print(f"Installed '{skill_id}' to {installed_path}")

skills/rocm-kernels/CHANGELOG.md

@@ -0,0 +1,61 @@
+# Changelog
+
+## v0.3 (2026-04-10)
+
+### Added
+- **Consolidated benchmark package**: `examples/ltx-video-benchmark/benchmark_results.json` now provides a single JSON artifact with formal R9700 benchmark runs and summary statistics.
+- **Live harness benchmark trace**:
+  - `examples/ltx-video-benchmark/trace/opencode_live/opencode_trace_result.json`
+  - Supporting run output in `examples/ltx-video-benchmark/trace/opencode_live/results.json`
+- **Unified examples documentation**: `examples/ltx-video-benchmark/README.md` is now the single reviewer-facing README for the examples package.
+- **Dependency lock-in entrypoint**: `scripts/requirements.txt` is used as the canonical install file for benchmark/integration examples.
+
+### Changed
+- **E2E benchmark CLI/output semantics** (`scripts/benchmark_e2e.py`):
+  - Standardized output flag to `--output-dir`
+  - Updated docs and usage examples to match current examples layout
+  - Structured outputs under `examples/ltx-video-benchmark/` and `trace/*_live/`
+- **Examples content structure**:
+  - Merged previous top-level examples README content into `examples/ltx-video-benchmark/README.md`
+  - Updated trace/result paths in docs to current live locations
+- **Script requirement notes**:
+  - Updated benchmark/example script docstrings to install dependencies from `scripts/requirements.txt`
+
+### Fixed
+- Removed ambiguity between "harness replay" and "live benchmark execution" by recording explicit `live_benchmark: true` trace outputs with executed command/config.
+- Aligned reviewer-facing materials to use one benchmark documentation entrypoint and one consolidated formal benchmark JSON.
+
+### Removed
+- Legacy split/duplicate examples documentation and outdated trace path references in the examples package.
+- Codex-specific benchmark trace artifacts and non-essential video artifacts from the reviewer package.
+
+## v0.2 (2026-03-12)
+
+### Added
+- **Transformers integration**: `references/transformers-integration.md` — LLaMA/Mistral/Qwen RMSNorm patching, Flash Attention 2, epsilon handling differences
+- **Transformers injection script**: `scripts/transformers_injection_example.py` — minimal runnable example (~150 lines)
+- **HuggingFace Kernels Hub integration**: `references/huggingface-kernels-integration.md` — `get_kernel`, `has_kernel`, publishing, ROCm compatibility notes
+- **HuggingFace Kernels example script**: `scripts/huggingface_kernels_example.py` — Hub loading, benchmarking, model integration with fallback
+- **GEMM template with XCD swizzle**: Template 5 in `kernel-templates.md` — full GEMM kernel with XCD swizzle for MI355X, L2 cache grouping, autotune configs, Python API, and benchmark
+- **CHANGELOG.md**: Version tracking for skill iterations
+
+### Fixed
+- Broken cross-references: "Template 2" for GEMM → corrected to "Template 5" in `troubleshooting.md`, `kernelbench-classification.md`, and `skill-evaluation-methodology.md`
+- R9700 Memory Bandwidth: filled in ~608 GB/s (was TBD) in SKILL.md
+
+### Updated
+- `SKILL.md` See Also section: added new integration guides, scripts, and Hub links
+- `SKILL.md` argument-hint: added gemm, transformers, huggingface-kernels, get_kernel
+- `manifest.txt`: added all new files
+
+## v0.1 (2026-03-10)
+
+### Added
+- Initial skill with SKILL.md, 4 kernel templates (RMSNorm, RoPE 3D, GEGLU, AdaLN)
+- MI355X and R9700 GPU optimization guides
+- Diffusers integration guide (LTX-Video)
+- Troubleshooting guide (14 ROCm-specific issues)
+- Benchmark scripts: micro-benchmark (`benchmark_kernels.py`) and E2E (`benchmark_e2e.py`)
+- LTX-Video injection example (`ltx_kernel_injection_example.py`)
+- KernelBench classification and evaluation methodology docs
+- Kernel-agent knowledge base