What problem does it solve? When a retrieval-augmented generation system returns the wrong files at the top of query results despite good semantic similarity, this Skill fixes the ranking by editing weight multipliers in the RAG config — no re-indexing or re-embedding required. ## Core Features & Use Cases - Query-time weight tuning: Edits the ranking.weights table in tools/rag/rag-config.yaml using fnmatch glob patterns, with first-match-wins ordering. - Guided diagnosis workflow: Runs failing queries, interprets score gaps, and maps symptoms (right file ranked low, wrong file winning, missing file) to specific weight adjustments. - Regression validation: Re-runs the query immediately and executes the full eval benchmark (eval.py) to confirm no previously passing queries regressed. - Use Case: An ADR about coupons consistently ranks below router README pages for coupon queries. Add a glob rule docs/adr/0016/** with weight 1.15 before the broader patterns, then verify the file moves to position one. ## Quick Start Ask the assistant to tune the RAG ranking weights so that a specific file or folder ranks higher or lower for a given query, providing the file path and whether to boost or suppress it.
