# Ml Feature Engineering

> ML feature engineering workflow for feature definition, lineage, and online-offline parity. Use when model performance depends on explicit feature design and parity controls; do not use for generic API-layer or infrastructure-only changes. Use when this capability is needed.

- Skill: `tomevault-io/ml-feature-engineering` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add tomevault-io/ml-feature-engineering`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/ml-feature-engineering/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: tomevault-io (https://skillmd.com/u/tomevault-io)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/tomevault-io/ml-feature-engineering

---


# Ml Feature Engineering

## Overview
Use this skill to design features that are useful, explainable, and consistent across training and serving.

## Scope Boundaries
- Use this skill when the task matches the trigger condition described in `description`.
- Do not use this skill when the primary task falls outside this skill's domain.

## Shared References
- Online/offline parity rules:
  - `references/online-offline-parity-rules.md`

## Templates And Assets
- Feature specification template:
  - `assets/feature-spec-template.csv`

## Inputs To Gather
- Candidate feature hypotheses and business rationale.
- Data sources and freshness constraints.
- Serving path capabilities and latency budget.
- Leakage/fairness/compliance constraints.

## Deliverables
- Feature catalog with lineage and ownership.
- Parity validation plan for train vs serve paths.
- Feature risk and maintenance notes.

## Workflow
1. Define feature specs in `assets/feature-spec-template.csv`.
2. Validate parity assumptions with `references/online-offline-parity-rules.md`.
3. Prioritize features by incremental value vs complexity.
4. Verify leakage and freshness assumptions.
5. Publish feature rollout and deprecation plan.

## Quality Standard
- Feature definitions are versioned and reproducible.
- Online/offline behavior is consistent for decision-critical features.
- Feature ownership and monitoring are explicit.

## Failure Conditions
- Stop when feature logic diverges between training and serving.
- Stop when feature value cannot justify operational complexity.
- Escalate when parity gaps remain unresolved.

---
> Converted and distributed by [TomeVault](https://tomevault.io/claim/kentoshimizu) — claim your Tome and manage your conversions.
<!-- tomevault:4.0:skill_md:2026-04-15 -->