mbtorch-api-design

What problem does it solve?

Ensures MbTorch public APIs are consistent, discoverable, and easy to use by making design decisions from the user's perspective rather than implementation convenience. It reduces fragmentation, surprising behaviors, and costly breaking changes by enforcing naming conventions, sample-first design, clear error messages, and a checklist-driven review process.

Core Features & Use Cases

• Naming & Conventions: Prescribes PascalCase for types, snake_case for functions, and alignment with PyTorch/JAX conventions where appropriate.
• Sample-first API design: Forces writing caller-facing examples before committing signatures to ensure ergonomics and discoverability.
• Compatibility & Error Guidance: Provides a backward-compatibility workflow, deprecation policy, and structured error message format for io/nn/core/optim changes.
• Use case: Use this Skill when adding a new optimizer, changing Model I/O signatures (ONNX/safetensors/.mbt), standardizing layer names, or preparing breaking-change migration notes.

**Quick Start

Ask the assistant to review a proposed public API change for core/nn/optim, produce a sample-first usage example, validate naming consistency, and list backward-compatibility and documentation tasks.