MLX Engineering Helper

What it does

This skill provides an actionable engineering workflow for working with MLX and MLX-LM: verifying upstream behavior, running runtime probes, porting PyTorch/JAX code, debugging Metal kernels and memory/caching behavior, and validating local model loads on Apple silicon. It combines quick-start commands, probe scripts, and high-signal guidance for common MLX pitfalls.

When to use it

Use when you need to fact-check current MLX behavior, audit or patch MLX-based repos, port code from other frameworks, profile GPU execution, or validate local MLX-LM inference and prompt-cache behaviors. Ideal for engineers working on Apple-silicon deployments or MLX-LM model integration.

What's included

• Scripts: Bundled helpers and probes in scripts/ (e.g., mlx_probe.sh, mlx_release_info.sh) to verify runtime and releases.
• References: references/current-facts.md and references/porting-checklist.md with common caveats and checks.
• Instructions: Detailed workflows for classification (current facts vs porting), runtime validation, kernel escalation, and high-signal differences between MLX and MLX-LM.

Compatible agents

{}