model-registry-governance

1# Model Registry Governance
2 
3Create a trustworthy system of record for model artifacts, prompts, adapters, and evaluation evidence.
4 
5## When to Use This Skill
6 
7- Setting up a centralized model registry for your organization
8- Defining metadata standards for model artifacts
9- Building approval workflows for model promotion to production
10- Implementing lifecycle policies for model retirement
11- Preparing for compliance audits of AI systems
12 
13## Prerequisites
14 
15- MLflow Tracking Server or Weights & Biases instance deployed
16- Object storage for model artifacts (S3, GCS, or MinIO)
17- CI/CD pipeline with access to the registry API
18- OPA or similar policy engine for governance checks
19- Git repository for policy definitions and promotion scripts
20 
21## Core Principles
22 
23- **Traceability**: every production model maps to source code, data snapshot, and evaluation results.
24- **Reproducibility**: builds are deterministic with pinned dependencies.
25- **Policy-driven promotion**: no manual bypass for critical safety checks.
26- **Lifecycle hygiene**: stale, vulnerable, or unowned models are retired automatically.
27 
28## MLflow Registry Setup
29 
30```bash
31# Install MLflow with required backends
32pip install mlflow[extras] psycopg2-binary boto3
33 
34# Start MLflow tracking server with PostgreSQL backend and S3 artifact store
35mlflow server \
36  --backend-store-uri postgresql://mlflow:password@db:5432/mlflow \
37  --default-artifact-root s3://mlflow-artifacts/models \
38  --host 0.0.0.0 \
39  --port 5000 \
40…