Preface¶

When conducting computational research with DSH, the common practice involves keeping conversations in Chat, scattering code and data across local directories, manually organizing experiment logs, and drafting papers in a separate editor. Progress relies on memory, and there lacks an auditable binding between Scope, experiment Contract, Evidence, and Claims, making it difficult to restore a complete project state from a single conversation.

Below, we introduce dsh-scholar (GitHub: lzszq/dsh-scholar, maintainer lzszq). It is a research workspace plugin for DSH that consolidates project conversations, research materials, code and data, controlled experiment runs, Evidence, and TeX manuscripts into a single recoverable project. You can start from a new question or continue advancing existing work.

What It Is¶

dsh-scholar is categorized under “Connected Tools” in SkillHub, with 25 GitHub stars. The package name is @dsh-scholar/research-plugin (version 0.1.0, MIT license), targeting computational research in fields like machine learning, data science, and bioinformatics. It is not intended for high-risk scenarios such as clinical decision-making, human trials, or wet lab experiments.

The plugin shares the same Research Kernel (127.0.0.1:7412) and project data directory (~/.dsh/research-kernel) with the standalone workspace. Upon installation in DSH, it adds the Scholar tool, slash commands, Skills, settings, and a compact dsh Scholar tab. Unbound conversations can be linked to existing projects or create new projects with just a name, while bound conversations display the current stage, next steps, and execution summary. The full interface opens via “Open in new page” or a keyboard shortcut.

Core Features¶

Stage-Aware Research Guidance¶

Chat supports natural language, Grill Me project inception Q&A, file uploads, explicit slash commands, and authoritative NextAction prompts tailored to the current research stage. Regular narratives are interpreted based on the project’s current stage; explicit commands serve as deterministic, advanced entry points.

Governable Research Workflow¶

Decisions such as Scope, Idea, Contract, Evidence, Direction, and Release remain explicit, revisable, and auditable. The default governance mode is gate-only: the Agent cannot impersonate a Human subject, fabricate accepted Evidence, or bypass research Gates. full-auto only auto-passes Scopes, Ideas, Contracts, and Budget Gates that are whitelisted within registered FixtureProfiles, and currently, the only standardized action executor is survey_run. Release, Direction, Intake, Evidence, etc., remain under Human control.

Controlled Execution Environment¶

Runner Profiles describe local, local Docker, or remote SSH environments, allowing you to fix container images and declare NVIDIA GPU capabilities. Formal Jobs will not execute until the Profile, target, snapshot, Protocol, and budget are ready. If a Runner is missing, the target is offline, a SecretRef is unavailable, or the Contract is incomplete, the interface shows preparation or blocking items instead of marking it as runnable.

Integrated Workspace¶

Project-level Chat, editable files, session-bound Web terminals, run logs, artifacts, TeX source, compilation diagnostics, and PDF previews share the same context. Protocol revisions, run classifications, synthesis requests, review results, knowledge package activations, and Claim-Evidence linkages persist as research state. Trajectory and Topology views allow you to inspect sub-agent relationships, states, subsequent tasks, and outputs.

Installation and Enabling¶

The local workspace requires Linux, Node.js 24, pnpm 11.20.0, and Docker Engine for controlled experiments, TeX compilation, and clean-room reproduction. The @dsh-scholar/* packages are not yet published to npm; you need to build from source and install via local paths.

1. Build the Standalone Workspace (Optional but Recommended)¶

After cloning or downloading the repository, run the following commands in the repository root:

pnpm install --frozen-lockfile
pnpm run build

Start the standalone UI:

bash scripts/start-standalone-ui.sh

Open http://127.0.0.1:18610 in your browser, read the token from the following file, and paste it:

~/.dsh-scholar-standalone/research-ui-standalone/standalone-token

2. Install the DSH Pre-release Version¶

npm install -g @deepseek-ai/dsh@next
npm ls -g @deepseek-ai/dsh --depth=0

Note the actual installed version for troubleshooting compatibility issues.

3. Add the Plugin to the DSH Web Profile¶

After completing the build in the repository root, register the plugin using the absolute path:

cd /absolute/path/to/dsh-scholar
pnpm install --frozen-lockfile
pnpm run build
dsh plugin --profile web add /absolute/path/to/dsh-scholar
dsh plugin --profile web why @dsh-scholar/research-plugin
dsh web

To update, rebuild the same checkout and add the same path again. To uninstall:

dsh plugin --profile web remove @dsh-scholar/research-plugin

4. Configure Execution Environment and Plugin Settings¶

Open Settings → Execution environment in the standalone workspace or DSH, and select a Runner Profile (local, local Docker, or remote SSH). In DSH, open Settings → Plugin config → dsh Scholar; restart DSH for changes to take effect after saving.

The Standalone URL must not contain credentials, query parameters, or fragments. Access tokens can only be copied explicitly from a loopback DSH instance; the page does not display tokens or expose the Kernel, Runner, Provider, or SSH keys.

Typical Usage¶

Three Entry Points¶

1. New Research: Provide only the project name and answer Grill Me questions in Chat to complete the Brief.
2. Open Existing Project: Resume previously persisted stages, conversations, files, tasks, runs, and methodology history.
3. Upload / Join: Attach papers, code, data, images, or logs to an existing stage; uploaded materials first enter isolated Intake and do not automatically become Evidence.

Common Workflow¶

Create or join → Grill Me → Scope → survey → Ideas → Baseline → Contract
→ controlled Runs → classification and synthesis → Evidence and Claims
→ TeX writing and review → private bundle → Human Release Gate

Common Slash Commands¶

/new  /status  /survey  /ideas  /ideas generate 3  /ideas select <idea_id>
/gates  /contract  /run  /evidence  /claims  /write  /review
/release-bundle  /release

/run only executes when the snapshot, Protocol, Runner, target, and budget are all ready. Formal experiments must bind immutable code and data snapshots, a Protocol that is frozen when needed, and an explicit Runner Profile; Chat text, ordinary stdout, and interactive terminal output do not automatically become formal Evidence.

Use Cases and Considerations¶

Who It’s For: Developers or researchers who need auditable, reproducible computational research within DSH; those who wish to chain project inception, experimental Contracts, runs, Evidence, and TeX writing into a traceable pipeline, rather than aligning stages manually across multiple tools.

Boundaries: The plugin assists research but does not assume responsibility for scientific judgment, approval, authorship, or publication. It is not suitable for clinical decision-making, human research, wet experiments, etc. full-auto has strict Fixture and whitelist boundaries; Release remains under Human control.

Security and Permissions: The plugin runs with the current dsh process permissions, accessing local files and invoking configured execution environments and network capabilities. Before installation, read the source code and MIT license to ensure Runner, SSH SecretRef, and remote target configurations comply with your security policies. The DSH ecosystem follows “everything is a plugin”; SkillHub (https://www.skillhub.cn/plugins/lzszq/dsh-scholar) is an independent community directory with no official affiliation to DeepSeek / High-Flyer.

Conclusion¶

dsh-scholar connects the DSH conversation layer with a governable research workflow, controlled execution, and manuscript output within the same project data directory, making it suitable for computational research scenarios requiring clear stages and traceable evidence. The directory page and source code can be found on SkillHub and GitHub, respectively:

• Directory page: https://www.skillhub.cn/plugins/lzszq/dsh-scholar
• GitHub: https://github.com/lzszq/dsh-scholar