dsh-researchcraft

Scientific work rarely lives inside a single tool. Literature review happens in a browser, experiment notes scatter across notebook apps, analyses run in a terminal, and manuscripts are assembled in a separate editor. dsh-researchcraft collapses that fragmented pipeline into one DeepSeek Harness conversation so a single AI agent can handle literature search, experiment logging, code execution, and manuscript formatting side by side, eliminating the constant context switching.

The plugin vendors more than 700 domain skills drawn from six open-source projects, spanning chemistry, genomics, bioinformatics, statistics, machine learning, and scientific writing. A living lab notebook is writable, readable, and exportable across the entire subagent delegation tree in both Markdown and ZIP. Academic search is wired through four connectors (Parallel, Firecrawl, Scite, Consensus) and layered on top of a local hybrid BM25-plus-vector workspace index for fast on-device semantic retrieval of local documents. The toolkit also includes scientific file inspection (molecular structures, mass-spec traces, AnnData), LaTeX-to-PDF compilation, PDF-to-Markdown conversion, remote GPU or CPU offload via Modal and RunPod, generative scientific figures, and roughly 330 workflow templates that guide every step from question framing to final verification.

It is built for interdisciplinary graduate students, independent researchers, and small labs that want to drive a full research workflow from a chat interface without stitching together a stack of separate tools. Once the ResearchCraft preset is selected, the complete set of skills, connectors, and utilities is available immediately.