From mlx-optimizer
Guide MLX Metal profiling, mx.fast escalation, custom Metal kernels, and C++ extensions when profiling proves kernel-level bottlenecks.
How this skill is triggered — by the user, by Claude, or both
Slash command
/mlx-optimizer:mlx-metal-kernelsThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill only after normal MLX-level issues have been checked or when the
Use this skill only after normal MLX-level issues have been checked or when the user explicitly asks about Metal capture, custom kernels, or native extensions.
Before any Python execution, use the target repo's .venv. Never install
Python packages globally.
../../references/profiling-and-metal.md../../references/memory-and-dtypes.md../../references/compile-and-transforms.mdmx.fast, custom Metal kernels, or C++ extensions only when the
operation is a measured bottleneck and built-in MLX ops are insufficient.State whether custom native work is justified, which evidence supports that decision, and what maintenance burden it creates.
npx claudepluginhub sealad886/mlx-optimizer-plugin --plugin mlx-optimizerCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.