Automated prompt deployment requires quality gate evaluation

To prevent quality regressions during prompt changes, automated evaluation gates should verify new prompts meet quality standards before deployment. Evaluation criteria include relevance, completeness, faithfulness, and conciseness scores (0.0-1.0 scale). Minimum overall score threshold typically set at 0.7.

Implement prompt_deployment_gate in CI/CD pipeline. Run automated LLM-as-a-Judge evaluation against test dataset before deploying prompt changes. Use gpt-4o-mini as evaluator with temperature=0.0. Record scores to LangFuse using lf_client.score(). Block deployment if average score falls below min_overall_score threshold (default 0.7).