verify-rag-pipeline

Verifying RAG pipeline integrity involves systematically checking response composition, citation consistency, and evidence completeness. This Skill validates system prompts, formatting rules, and API safety to prevent inaccurate answers and broken citations after code modifications.

RAG pipeline verification checks LLM prompt neutrality, guardrail application, verdict thresholds, and question decomposition patterns. It ensures prompts adhere to formatting standards without introducing unintended biases, hedging, or unsafe API behaviors during response generation.

Testing RAG retrieval fallback mechanisms involves validating indexing guards, sub-question completeness, and "I Don't Know" paths. This Skill systematically checks these edge cases alongside vector search filter conversion and multi-product filter searches to ensure robust query handling.

Yes, you can check citation consistency and per-question evidence mapping in a multi-product RAG system. This Skill verifies multi-product filter searches, product name extraction, and Korean query normalization to ensure accurate evidence summarization across diverse query patterns.

Incomplete evidence after updating prompt templates often stems from broken citation consistency, N+1 query issues, or flawed question decomposition. Run RAG pipeline verification to confirm new prompts adhere to formatting rules and evidence mapping standards without introducing regressions.