MLflow Prompt Registry Patterns

When to Use

Use this skill when implementing agents that need:

• Versioned prompts: Track prompt changes over time
• A/B testing: Compare champion vs challenger prompts
• Runtime updates: Change prompts without code deployment
• Governance: Audit prompt changes and rollback capability

Core Principles

1. Single Source of Truth

• All prompts stored in Unity Catalog table (agent_config)
• Versioned in MLflow as artifacts
• Never hardcoded in agent code
• Runtime loading by alias (production, staging, champion, challenger)