dsh-yingnao

A local disk is the most honest digital archive you own, but tens of thousands of files scattered across hundreds of folders break semantic continuity. An agent can only recite dashboard numbers and never actually search, organise, or dispatch governance jobs. dsh-yingnao plugs the local index of Yingnao into DeepSeek Harness so the agent moves from read-only to hands-on: surveying dark zones, searching files, expanding along semantic neighbours, inspecting file quality, and dispatching governance tasks—each step with graceful degradation and a clear boundary.

The six tools are designed around the cognitive shape of the model, not as a mirror of the server endpoint list. related is the most signal-stable capability in a 76k-file benchmark: anchor one document, then radiate outward to surface drafts and earlier versions you never thought to query. govern has a hard line—actions that cost money (OCR re-extraction, LLM labelling) or move actual user files (inbox archiving) are queued and wait for desktop-client confirmation rather than executing silently. Everything else (quality checks, codepoint normalisation) runs immediately and is reversible, with a full backup manifest written before any change.

Best for folks with local document libraries, meeting notes, or creative asset folders who want their agent to manage material instead of merely reading metrics. Paired with dsh-deepbrain, a single agent holds both evidence-chained judgments and the raw source material behind them, closing the loop on writing, review, and knowledge-organisation workflows.