dsh-research-kit

Scientific writing demands clear steps, material requirements, and quality boundaries, yet a general chat box is built for one-off questions. Peer review, drafting an introduction, running a systematic literature search, designing an analysis plan—these tasks repeat week after week, and a single vague instruction rarely yields consistent output. What you actually need is to see the full prompt before it ships, edit it freely, and trust that no reference, data point, or conclusion has been fabricated. DSH Research Kit turns those recurring scientific workflows into reusable, parameterized, pre-send reviewable prompt templates so you can focus on the science itself.

The kit ships 349 human-reviewed research workflows spanning 28 process families—from manuscript writing and literature research to bioinformatics, clinical trials, neuroscience, ecology, and astrophysics; 109 skill entries covering methodological guidance modules (scientific writing, statistical review, citation verification, reproducibility checks) and host-capability catalogs (bulk RNA-seq, Nextflow, Benchling); and 129 scientific data sources, 11 of which support direct in-plugin queries returning candidates with stable identifiers like DOI or PMID. Every workflow is a parameterized, editable template with built-in anti-fabrication guardrails. After launching one, the workbench replays the assembly step by step so you can verify every anchor in the generated prompt before sending. An interactive evidence graph links resources, workflows, query sources, and saved evidence into a traceable network with zoom, path modes, and sanitized exports; 31 MCP tools let your host agent invoke search, verification, and review capabilities directly without touching the UI.

If you are a graduate student, postdoc, or domain researcher who runs the find-materials → structure → draft → verify loop on a regular cadence, and you want every step visible, editable, and auditable, Research Kit packs that discipline into a single workbench inside your DSH session. It does not make judgments for you or fake search results—it assembles the task, marks the boundaries, and hands execution back to your agent and, ultimately, to you.