ml-feature-governance

Plugin ml-feature-governance from marketplace snowflake-ml-pipeline in vahid-ai/snowflake-ml-pipeline. Commands read from the repo's manifest — 1 month ago.

Skills (9)

• architecture

  Enforce the ML feature-platform architecture whenever designing, editing, reviewing, or implementing feature engineering, training datasets, model inputs, feature stores, schemas, backend adapters, data transformations, or ML data pipelines.

• change

  Start or continue a governed ML feature, transformation, feature-set, backend, training-data, or pipeline change using the cross-client standalone workflow.

• experiment

  Define or review ML feature ablations, feature combinations, alternate transforms, feature counts, or backend comparisons as reproducible versioned experiments.

• init

  Initialize a repository for Claude Code, Cursor, or OpenAI Codex with the standalone ML feature-platform scaffold, repository instructions, local governance metadata, and deterministic validation.

• review-architecture

  Independently review canonical ML feature architecture, DAG design, versioning, fitted-state boundaries, feature-set contracts, and source-of-truth discipline.

• review-leakage

  Independently review ML feature and training pipelines for temporal leakage, split contamination, fitted-state leakage, target leakage, late data, and train-serve skew.

• review-portability

  Independently review ML feature implementations for cross-engine typing, determinism, capability, and semantic-equivalence risks.

• review

  Review an ML feature-platform design or code change for portability, leakage, reproducibility, data typing, transformation-state, and cross-backend correctness.

• validate

  Run ML feature governance validation, inspect architecture violations, and report precise fixes without changing generated outputs by hand.

Subagents (3)

• feature-architect

  Independently reviews canonical ML feature architecture, DAG design, versioning, fitted-state boundaries, feature-set contracts, and source-of-truth discipline.

• leakage-reviewer

  Reviews training and feature pipelines for temporal leakage, split contamination, state fitting leakage, target leakage, late-arriving data issues, and train/serve skew.

• portability-reviewer

  Reviews feature implementations for Spark/Snowflake/DuckDB/Polars/Ray/Dask/RAPIDS portability, type coercion, determinism, and cross-engine equivalence risks.

Hooks (3)

• SessionStart

  python "${CLAUDE_PLUGIN_ROOT}/scripts/session_context.py"

• PreToolUse

  python "${CLAUDE_PLUGIN_ROOT}/scripts/guard_write.py"

• PostToolUse

  python "${CLAUDE_PLUGIN_ROOT}/scripts/validate_after_write.py"

Similar plugins