| name | mlx-optimizer |
| description | Route Python-first MLX optimization work on Apple Silicon to focused audit, training, inference, Metal, or bridge workflows. |
MLX Optimizer
Use this skill when the user asks Codex to optimize, audit, benchmark, profile,
or explain MLX code on Apple Silicon.
First Actions
- Identify the target repo or file.
- Check for a repo-local
.venv before any Python execution.
- Inspect project structure before proposing new modules.
- Treat static observations as candidate findings until verified.
- Require concrete evidence for optimization claims.
- For version-sensitive MLX API behavior, retrieve current authoritative MLX
docs before implementation or recommendations.
Routing
- Repo-wide or unknown performance issue: read
../mlx-performance-audit/SKILL.md.
- Training loop issue: read
../mlx-training-optimizer/SKILL.md.
- Inference or generation issue: read
../mlx-inference-optimizer/SKILL.md.
- Metal capture, custom kernel, or C++ extension question: read
../mlx-metal-kernels/SKILL.md.
- Swift/C/C++ or other language integration: read
../mlx-portability-bridges/SKILL.md.
Core References
Read only the files needed for the routed task:
../../references/mlx-core-concepts.md../../references/eval-and-synchronization.md../../references/compile-and-transforms.md../../references/memory-and-dtypes.md../../references/reporting-format.md
Non-Negotiables
- Never install Python packages globally.
- Preserve unrelated user changes.
- Prefer existing repo abstractions.
- Do not recommend automatic rewrites without active code-path evidence.
- Include commands, outcomes, and residual risks in the final handoff.
