What problem does it solve? Answering questions from a local document corpus requires finding the most relevant passages, and a ranked list without similarity scores hides weak matches and manufactures false confidence in the answer. ## Core Features & Use Cases - Top-k similarity retrieval: Embed a query the same way the corpus was embedded, score every passage, and return the top k (default 5) ordered by score descending. - Score-transparent reporting: Every hit carries its numeric score on one shared scale, with weak hits labeled weak and near-duplicates flagged rather than double-counted. - Corpus-gap honesty: When the top score falls below a stated floor, the skill reports that the corpus does not contain the answer instead of dressing up the least-bad passage. - Use Case: Ask what your team's internal documentation says about a deployment rollback policy and receive a ranked list of the five most similar passages with scores, sources, and a verdict on whether the corpus actually answers the question. ## Quick Start Ask what the local corpus says about a specific question and request the top five matching passages with their similarity scores and sources.
