Skip to content

[FEA] Add a repository-grounded cuML agent skill #8638

Description

@cafzal

Is your feature request related to a problem? Please describe.

Agents implementing machine-learning workflows can introduce leakage, fit preprocessing on held-out rows, lose stable output keys, or claim GPU acceleration without dispatch evidence. The repository does not currently provide an Agent Skills-compatible guide and executable routing/correctness evaluations grounded in the checked-in APIs.

Describe the solution you'd like

Add a generalized cuml-machine-learning skill for native cuML and cuml.accel. Cover task definition, deployment-aligned splitting, training-only preprocessing, baselines, supervised and unsupervised evaluation, stable result handoff, acceleration/fallback evidence, and library boundaries.

Describe alternatives you've considered

Relying only on estimator API reference pages does not provide an end-to-end statistical and execution-evidence contract. A workflow-specific guide would not generalize across cuML use cases.

Additional context

A draft implementation is available in #8639.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions