A practical skill pack for writing custom Metal compute kernels using MLX (mx.fast.metal_kernel) on Apple Silicon.
python -m venv .venv && source .venv/bin/activate
pip install -U mlx
python skill/scripts/rmsnorm_kernel.py
python skill/scripts/softmax_kernel.py
python -m skill.tests.smoke_testIf you just need task routing and patterns, start with skill/SKILL.md.
- macOS on Apple Silicon (Metal-capable GPU)
- Python 3.12+
- MLX 0.30.6+
skill/SKILL.md: skill entrypoint and routing guidedocs/USING_WITH_CLAUDE_CODE.md: using this project from Claude Codedocs/USING_WITH_CODEX_CHATGPT.md: using this project from Codex (ChatGPT)CONTRIBUTING.md: contribution workflow and review expectationsSECURITY.md: vulnerability reporting policyRELEASE_CHECKLIST.md: pre-publish and release checklist
skill/SKILL.md— primary skill guide / index of patternsskill/scripts/— runnable examplesskill/kernels/— reusable helpers (autotune + cache, scalar buffers, device capability checks)skill/tests/— smoke tests for local/CI useskill/references/— deep-dive reference docs
mlx-metal-kernels-skill/
├── README.md
├── CONTRIBUTING.md
├── SECURITY.md
├── RELEASE_CHECKLIST.md
├── docs/
│ ├── USING_WITH_CLAUDE_CODE.md
│ └── USING_WITH_CODEX_CHATGPT.md
├── .github/workflows/ci.yml
└── skill/
├── SKILL.md
├── manifest.txt
├── kernels/
├── scripts/
├── tests/
└── references/