dsh-web-search-pro

Inside DeepSeek Harness, default web retrieval is often limited to a single engine, lacks cross-session memory, and struggles with login-gated platforms or dynamically rendered pages. dsh-web-search-pro bundles multi-engine search, persistent caching, platform search, and Playwright rendering into 11 directly callable tools, giving the model a more reliable and deeper way to access web content instead of starting from scratch on every query.

Its core capabilities include multi-engine search with RRF fusion, automatic fallback, engine cooldown, and a two-tier memory/SQLite cache; platform-specific search across 20 services such as GitHub, Bilibili, YouTube, Xiaohongshu, and Reddit, with support for authenticated sessions; readable fetching with browser-rendering fallback; and persistent per-site extraction rules plus custom platforms, allowing users to adapt to site changes or add new sources without touching code.

The plugin works closely with dsh-browser to offer layered automation—from read-only scripts and structured recipes to external userscripts—while approval policies and resource budgets prevent over-calling in standard mode. It is especially suitable for developers, researchers, and automation users who need in-depth web research, platform data collection, or long-term knowledge accumulation inside DeepSeek Harness.