Ml Ops Engineer

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass.

.codex/skills/a11y-audit

.codex/skills/ab-test-setup

.codex/skills/account-executive

.codex/skills/ad-creative

.codex/skills/aeo

Audit 0.4.1 · source sha 6e3fcbdbb5cf · 19 Oct 2026Report an issue or request a re-scan

• Publisherborghei GitHub userOwner of the repository this skill is read from
• Source folderborghei/Claude-Skills/data-analytics/ml-ops-engineerThe exact folder we audited
• LicenceNOASSERTIONAs the repository states it
• Commit pinned for this verdict6e3fcbdbb5cfA new audit runs when the source changes
• Repository statusNot archivedLast push 12 days ago
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: MLOps across model deployment, ML pipelines, monitoring, and feature stores. Use when deploying models to production, building training pipelines, setting up drift detection, configuring feature stores, or automating ML…

---
name: ml-ops-engineer
description: >
  MLOps across model deployment, ML pipelines, monitoring, and feature stores.
  Use when deploying models to production, building training pipelines, setting
  up drift detection, configuring feature stores, or automating ML CI/CD
  workflows.
license: MIT + Commons Clause
metadata:
  version: 1.0.0
  author: borghei
  category: data-analytics
  updated: 2026-03-31
  tags: [mlops, deployment, pipelines, monitoring, feature-store]
---
# MLOps Engineer

The agent operates as a senior MLOps engineer, deploying models to production, orchestrating training pipelines, monitoring model health, managing feature stores, and automating ML CI/CD.

## Clarify First

Before deploying, confirm these inputs. If any is unknown or vague, ASK — do not assume:

- [ ] **Serving mode + latency SLA** — real-time (FastAPI/K8s) or batch, and the P99 target (drives the entire deployment architecture)
- [ ] **Current MLOps maturity** — manual, pipeline, CI/CD, or full (identifies the highest-impact gap to close first)
- [ ] **Model artifact + registry/infra** — framework, where it is stored, and target platform (MLflow, K8s) (drives the serving and registry config)
- [ ] **Monitoring thresholds** — drift and accuracy-drop limits plus check cadence (drives alert rules and drift detection)

Stop rule: ask only the 2-3 that most change the output. If the user says "just draft it," proceed and list your assumptions at the top of the artifact.

## Workflow

1. **Assess ML maturity** -- Determine the current level (manual notebooks vs. automated pipelines vs. full CI/CD). Identify the highest-impact gap to close first.
2. **Build or extend training pipeline** -- Define fetch-data, validate, preprocess, train, evaluate stages. Use Kubeflow, Airflow, or equivalent. Gate deployment on an accuracy threshold (e.g., > 0.85).
3. **Deploy model for serving** -- Choose real-time (FastAPI + K8s) or batch (Spark/Parquet) based on latency requirements. Configure health checks, autoscaling, and resource limits.
4. **Register in model registry** -- Log parameters, metrics, and artifacts in MLflow. Transition the winning version to Production stage; archive the previous version.
5. **Instrument monitoring** -- Set up latency (P50/P95/P99), error rate, prediction-distribution, and feature-drift dashboards. Configure alerting thresholds.
6. **Validate end-to-end** -- Run smoke tests against the serving endpoint. Confirm monitoring dashboards populate. Verify rollback procedure works.

## MLOps Maturity Model

| Level | Capabilities | Key signals |
|-------|-------------|------------|
| 0 - Manual | Jupyter notebooks, manual deploy | No version control on models |
| 1 - Pipeline | Automated training, versioned models | MLflow tracking in use |
| 2 - CI/CD | Continuous training, automated tests | Feature store operational |
| 3 - Full MLOps | Auto-retraining on drift, A/B testing | SLA-backed monitoring |

## Real-Time Serving Example

```python
# model_server.py -- FastAPI model serving
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
import mlflow.pyfunc, time

app = FastAPI()
model = mlflow.pyfunc.load_model("models:/fraud_detector/Production")

class PredictionRequest(BaseModel):
    features: list[float]

class PredictionResponse(BaseModel):
    prediction: float
    model_version: str
    latency_ms: float

@app.post("/predict", response_model=PredictionResponse)
async def predict(req: PredictionRequest):
    start = time.time()
    try:
        pred = model.predict([req.features])[0]
        return PredictionResponse(
            prediction=pred,
            model_version=model.metadata.run_id,
            latency_ms=(time.time() - start) * 1000,
        )
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.get("/health")
async def health():
    return {"status": "healthy", "model_loaded": model is not None}
```

## Kubernetes Deployment

```yaml
# k8s/model-deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: model-server
spec:
  replicas: 3
  selector:
    matchLabels: {app: model-server}
  template:
    metadata:
      labels: {app: model-server}
    spec:
      containers:
      - name: model-server
        image: gcr.io/project/model-server:v1.2.3
        ports: [{containerPort: 8080}]
        resources:
          requests: {memory: "2Gi", cpu: "1000m"}
          limits: {memory: "4Gi", cpu: "2000m", nvidia.com/gpu: 1}
        env:
        - {name: MODEL_URI, value: "s3://models/production/v1.2.3"}
        readinessProbe:
          httpGet: {path: /health, port: 8080}
          initialDelaySeconds: 30
          periodSeconds: 10
---
apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: model-server-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: model-server
  minReplicas: 2
  maxReplicas: 10
  metrics:
  - type: Resource
    resource:
      name: cpu
      target: {type: Utilization, averageUtilization: 70}
```

## Drift Detection

```python
# monitoring/drift_detector.py
import numpy as np
from scipy import stats
from dataclasses import dataclass

@dataclass
class DriftResult:
    feature: str
    drift_score: float
    is_drifted: bool
    p_value: float

def detect_drift(reference: np.ndarray, current: np.ndarray, threshold: float = 0.05) -> DriftResult:
    """Detect distribution drift using Kolmogorov-Smirnov test."""
    statistic, p_value = stats.ks_2samp(reference, current)
    return DriftResult(feature="", drift_score=statistic, is_drifted=p_value < threshold, p_value=p_value)

def monitor_all_features(reference: dict, current: dict, threshold: float = 0.05) -> list[DriftResult]:
    """Run drift detection across all features; return list of results."""
    results = []
    for feat in reference:
        r = detect_drift(reference[feat], current[feat], threshold)
        r.feature = feat
        results.append(r)
    return results
```

Excerpt, first 6,000 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

git clone https://github.com/borghei/Claude-Skills
git -C Claude-Skills checkout 6e3fcbdbb5cf

Works in

1. 2026-10-19
   !CAUTION B 89 / 100

   first audit · source 6e3fcbdbb5cf

Questions about Ml Ops Engineer

• github.com/borghei/Claude-Skills/tree/main/data-analytics/ml-ops — the audited source at commit 6e3fcbdbb5cf and every finding aboveObserved · 19 Oct 2026

Spotted something wrong? Report a correction · Methodology