Guides

CI/CD Integration

Automate data quality and target leakage checks in your deployment pipelines

Data quality issues and target leakage often sneak into production because datasets are updated out-of-band or model retraining runs automatically. Featuresmith is designed to run inside CI/CD pipelines to prevent model degradation by gating deployments on strict rule audits.

Exit-Code Gating

The featuresmith analyze command returns deterministic exit codes that shell runners can check to decide whether to block or proceed with the build:

0: Clean run — no findings detected at or above the requested severity threshold.
1: Rule violation(s) detected — one or more audits failed. This will automatically fail standard CI steps unless ignored.
2: Invalid input — misspelled arguments, target column not in schema, or incorrect options.
3: Ingestion/load failure — database or files could not be read or parsed.

GitHub Actions Workflow

Here is a complete, copy-pasteable GitHub Actions workflow file (.github/workflows/data-audit.yml) that runs Featuresmith on every pull request to check for schema drift and target leakage:

yaml

1name: Data Quality Audit
2
3on:
4  push:
5    branches: [ main ]
6  pull_request:
7    branches: [ main ]
8
9jobs:
10  audit-data:
11    runs-on: ubuntu-latest
12    steps:
13      - name: Checkout Code
14        uses: actions/checkout@v4
15
16      - name: Install Python & uv
17        uses: astral-sh/setup-uv@v3
18        with:
19          python-version: "3.11"
20
21      - name: Install Featuresmith
22        run: |
23          uv pip install featuresmith-core featuresmith-cli --system
24
25      - name: Run Data Quality Audit
26        run: |
27          # Gate pipeline on 'warning' and 'critical' severity levels
28          featuresmith analyze data/train.csv --target churn --severity warning

Customizing Gates

You can customize the strictness of your CI gate by overriding severity settings in your local .featuresmith.yml configuration file, or by passing the --severity flag to the CLI.

bash

1# Only fail builds on critical violations (fully empty columns or target leakage)
2featuresmith analyze data/train.csv --target churn --severity critical