| name | local-ai-runtime-review |
| description | Review LewLM local AI runtime, routing, serving, MLX, llama.cpp, OpenAI-compatible adapter, benchmark, and fallback changes. Use for local-first runtime correctness and capability reporting. |
| argument-hint | Describe the runtime or routing change |
Local AI Runtime Review
Use this skill for changes that affect model discovery, routing, serving profiles, runtime adapters, benchmarks, or capability/fallback reporting.
Procedure
- Identify which runtime path is affected: MLX text, MLX vision, MLX audio, llama.cpp/GGUF, external OpenAI-compatible adapter, documents, or core routing.
- Check whether the change affects public behavior through the CLI, API, events, Python facade, or app helpers.
- Verify capability claims are honest:
- return explicit fallback metadata when a runtime cannot support a constraint
- avoid universal performance claims across all backends
- keep optional dependencies profile-gated
- Map to tests under
tests\unit, tests\integration, or tests\e2e.
- Avoid requiring local model weights unless the user explicitly requested real-model validation.
Output
Return:
- affected runtime path
- capability/fallback impact
- public surface impact
- targeted tests
