dao-zang-skill

Finding a single passage or concept in the Daoist Canon has long meant wading through tens of thousands of scrolls in the Zhonghua Daozang or the Zhengtong Daozang. dao-zang-skill consolidates 285,117 scripture chunks and 3,152 source Markdown files into one local workspace so that every search and original-text extraction runs entirely offline with zero network dependency.

The default text engine, built on ChromaDB full-text filtering and TF/IDF ranking, needs no extra packages and returns keyword hits in milliseconds. Switching to the semantic engine, a local bge-m3 ONNX model produces 1024-dimension cosine vectors identical to those in the database, so retrieval stays free of any external embedding API. Each hit is traced back to the raw Markdown file with special hit markers and line numbers, and a file-name filter lets you narrow the search to a curated subset of sources.

Designed for Daoist scholars, Chinese NLP developers, and anyone who needs fast, private, offline access to Daoist scripture inside a DeepSeek Harness environment. Three distribution packages cover direct-copy, download-and-offline-rebuild, and clone-and-download workflows, all released under the MIT license.