Skip to main content

adding-model-support

Use this skill whenever contributing model support to the `winml` CLI: an unsupported Hugging Face model or architecture, an existing family that fails in one or more commands, a missing recipe, exporter, resolver, task, dataset adapter, evaluator, or promotion handoff. It diagnoses the gap, implements the narrowest reusable fix, validates the result, opens a draft model-support PR, and captures reusable findings. Trigger even when the user asks to repair support rather than explicitly saying "add model support". Do not activate for explanation-only questions about operator coverage reports or execution-provider internals, even when they mention an unsupported operator or a crash; those require a request to implement or repair model support. Do not use it merely to run or tune an already-supported model, optimize an existing recipe, or add a new execution provider backend.

Source facts

Repository
microsoft/winml-cli
Last source activity
October 28, 2026 at 08:02
Detected SKILL.md language
English
Stars
51
Forks
14

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

File Explorer
48 files

Showing SKILL.md

SKILL.md
Source instructions · Read-only preview
nameadding-model-support
descriptionUse this skill whenever contributing model support to the `winml` CLI: an unsupported Hugging Face model or architecture, an existing family that fails in one or more commands, a missing recipe, exporter, resolver, task, dataset adapter, evaluator, or promotion handoff. It diagnoses the gap, implements the narrowest reusable fix, validates the result, opens a draft model-support PR, and captures reusable findings. Trigger even when the user asks to repair support rather than explicitly saying "add model support". Do not activate for explanation-only questions about operator coverage reports or execution-provider internals, even when they mention an unsupported operator or a crash; those require a request to implement or repair model support. Do not use it merely to run or tune an already-supported model, optimize an existing recipe, or add a new execution provider backend.