dsh-learn-wiki: A Learn-by-Doing Knowledge Base Plugin

`dsh-learn-wiki` is a CRAG-style corrective retrieval plugin in the DeepSeek Harness ecosystem for filling knowledge gaps after static retrieval failures. It uses a three-tier architecture: L1 (Markdown knowledge base), L2 (Hindsight), and L3 (Web). When retrieval misses, the plugin automatically performs rate-limited background web supplementation based on struggle-detection signals, stores the new knowledge in a staging area, and then uses explicit tools to review and commit it to the knowledge store. Implemented in pure JavaScript with no build step required, the plugin supports explicit retrieval, conversation history extraction, and knowledge provenance.