Engineer training pipelines, features, model artifacts, batch or online serving, performance, testing, deployment interfaces and resilience. Use for ML Engineer implementation and productionization work. Route general batch, CDC or streaming ingestion to data-engineering, and registry, drift or model rollout operations to mlops.
Installs into .claude/skills of the current project.
Are you the author of Machine Learning Engineering?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/kina2711-machine-learning-engineering)
---
name: machine-learning-engineering
description: "Engineer training pipelines, features, model artifacts, batch or online serving, performance, testing, deployment interfaces and resilience. Use for ML Engineer implementation and productionization work. Route general batch, CDC or streaming ingestion to data-engineering, and registry, drift or model rollout operations to mlops."
---
# Machine Learning Engineering
## Operating contract
1. Identify the requested deliverable, then select exactly one atomic task below.
2. Read that task file completely before acting.
3. Read [references/lifecycle-standard.md](references/lifecycle-standard.md) and use the task's profile, risk tier and execution path.
4. Load only the company context and adapter references needed for the selected task.
5. On a cross-role handoff, compare logical ID and SHA-256 in `references/shared-reference-manifest.json`; reuse an already loaded shared reference when both match instead of loading its duplicate.
6. Inspect real artifacts and systems before making change-sensitive claims.
7. Apply Definition of Ready, stage gates, test strategy, Definition of Done, approval and handoff requirements.
8. Do not invent access, approvals, successful execution, test results or business confirmation.
9. For any Git-backed mutation, regardless of the task's verb or profile, require a pre-change success/scope contract, a post-change `core-audit-change-scope`, and fresh `core-verify-deliverable` evidence before completion.
10. For learning, coaching or skill-transition work, resolve [the learner-memory interoperability contract](references/learning-memory-interoperability.md). Reuse only relevant verified summaries; never infer mastery from exposure, and expand stale, uncertain, changed-version or safety-critical prerequisites.
## Ownership
Own only deliverables listed in this role catalog. When the requested deliverable belongs to another role, produce a handoff instead of silently taking ownership. Use the department orchestrator for multi-role work.
## Stack-native adapter routing
After selecting the task, detect the real stack/version and read only the matching adapter below plus [the adapter index](references/technology-adapters.md). Do not load every adapter.
- [spark](references/adapter-spark.md)
- [kafka-flink](references/adapter-kafka-flink.md)
- [databricks](references/adapter-databricks.md)
- [mlflow-kubeflow](references/adapter-mlflow-kubeflow.md)
## Atomic task routing
- **Plan, define, design, map, specify or create a proposed artifact** (4 tasks): read [references/catalog-plan-design.md](references/catalog-plan-design.md).
- **Build, implement, configure, teach, interview or deliver an artifact** (10 tasks): read [references/catalog-build-deliver.md](references/catalog-build-deliver.md).
- **Inspect, analyze, test, review, validate, assess, certify or audit** (3 tasks): read [references/catalog-test-assure.md](references/catalog-test-assure.md).
- **Deploy, release, monitor, recover, migrate, optimize, retire or improve** (3 tasks): read [references/catalog-operate-improve.md](references/catalog-operate-improve.md).
Read only the best-matching catalog. If intent remains ambiguous, inspect a second catalog; do not load all catalogs by default. Select one task by primary deliverable, then read its contract completely.
## Completion response
State the selected task ID, primary deliverable, evidence inspected, validation performed, approval status, residual risks and next owner. A draft or plan is not an executed production outcome.
