dsh-scholar: A Research Workbench DSH plugin for DeepSeek Harness

dsh-scholar is a research workbench plugin (DSH plugin) for DeepSeek Harness, maintained by lzszq under the MIT license, last updated in 2026-08. It targets computational research (e.g., machine learning, data science, bioinformatics) by integrating research materials, project conversations, code and data, experiment runs, evidence ledger, and TeX manuscripts into a single resumable project, addressing issues of untraceable research processes, irreproducible experiments, and broken evidence chains. Core capabilities include a governable research workflow (Scope, Idea, Experiment Contract, Evidence, Claim, Release with human gates), controlled experiments (Runner executes frozen plans in local Docker or controlled remote machines), traceable evidence (claims traceable to Runs, Artifacts, and reviewed Evidence), and an integrated workbench (Chat, Workspace, Terminal, Manuscript sharing project context).

What core features does the dsh-scholar DSH plugin support?

The DSH plugin provides Scholar Agent tools, slash commands, Skills, configuration cards, and a dsh Scholar tab that reuses the standalone workbench. Research projects can start from scratch (Init), resume existing ones (Resume), or upload papers/code/data (Upload), with uploads going into isolated Intake and not automatically becoming Evidence. The default governance mode is gate-only; agents cannot approve human gates or forge evidence. Formal experiments must bind immutable snapshots and fixed execution environments.

How to install and build dsh-scholar?

Local experience requires Linux, Node.js 24, pnpm 11.20.0, and Docker Engine. Install command:

dsh plugin --profile web add github:lzszq/dsh-scholar

To build from source, run pnpm install --frozen-lockfile and pnpm run build in this repo, then start the standalone workbench:

bash scripts/start-standalone-ui.sh

Default page is http://127.0.0.1:18610, Research Kernel at 127.0.0.1:17413. On first open, paste the access token from ~/.dsh-scholar-standalone/research-ui-standalone/standalone-token.

How to start the experiment Runner for dsh-scholar?

Without a Runner, projects can still be managed but experiment jobs stay queued. To execute in local Docker, open another terminal:

export DSH_SCHOLAR_KERNEL_TOKEN="$(< ~/.dsh-scholar-standalone/research-ui-standalone/kernel-token)"
export DSH_SCHOLAR_SERVICE_TOKEN="$(< ~/.dsh-scholar-standalone/research-ui-standalone/service-token)"
node workers/runner-gateway/lib/bin/runner.js --kernel http://127.0.0.1:17413 --mode docker

How to integrate dsh-scholar with DeepSeek Harness and configure it?

The current compatibility baseline is @deepseek-ai/[email protected]. It is recommended to install and build the latest DSH source with pnpm, then add the plugin via local path:

cd /path/to/dsh-source
pnpm dsh plugin --profile web add /absolute/path/to/dsh-scholar
pnpm dsh plugin --profile web why @dsh-scholar/research-plugin
pnpm dsh web

After installation, go to DSH Settings → Plugin Config → dsh Scholar to adjust default governance mode (gate-only or full-auto), unattended operation, Standalone address (default http://127.0.0.1:18610/), and new-page shortcut (default Alt+Shift+S).

What are the usage boundaries and notes for dsh-scholar?

DSH Scholar assists research but does not replace researchers in scientific judgment, approval, authorship, or publication responsibilities. Chat, normal stdout, and Interactive Terminal output do not automatically become formal Evidence. The product focuses on computational research and is not suitable for clinical decisions, human trials, wet experiments, or other high-risk research. Standalone address must not contain credentials, query parameters, or URL fragments; tokens should not be placed in URLs.

View more DSH plugins at dsh-plugin.org