Introduction

DSH’s philosophy is “everything is a plugin”: encapsulating tools, processes, and context into composable capabilities. The community directory is an independent site with no official affiliation to DeepSeek or Fangfang, nor is it an official app store. For scenarios where you need to operate the SCNet supercomputing cluster within DSH, SSH configuration, Slurm jobs, cluster discovery, and compute node diagnosis are often scattered across manual steps. lql341/dsh-scnet packs these capabilities into a DSH bundle for dsh to invoke.

What is dsh-scnet

dsh-scnet is a community DSH bundle maintained by lql341, used to operate the SCNet supercomputing cluster. It belongs to the networking tool plugin category, providing profile-based SSH, Slurm job generation, cluster discovery, and compute node diagnosis capabilities.

It is an independent community project compatible with DeepSeek Harness, but it is not an official DeepSeek product, and it does not imply endorsement, cooperation, or authorization. The license is MIT.

Core Capabilities

The capabilities included are introduced below:

  • scnet-hpc skill: Provides profile-based operation guidance for SSH, Slurm, offline compute nodes, and accelerator verification.
  • scnet_list_clusters: Lists the packed cluster profiles.
  • scnet_show_cluster: Reads the selected profile before reporting resource constraints.
  • scnet_generate_job: Generates accelerator or pure CPU Slurm scripts adapted to the profile.
  • scnet_setup_ssh: Configures local SSH keys and host entries after explicit user confirmation.
  • scnet_probe_cluster: Generates an initial profile based on discovery from read-only login nodes.
  • scnet_refresh_cluster: Refreshes profile fields to ensure freshness; compute discovery is optional.
  • scnet_run_compute_probe: Submits a minimal compute node capability discovery.

Installation and Enabling

First, install the bundle to the web profile via dsh:

dsh plugin --profile web add dsh-scnet

After installation, check if the bundle layer appears in the current profile configuration:

dsh --profile web --dump-config

These two steps complete the installation and confirmation.

Runtime Environment

To use it, the following conditions must be met:

  • Node.js 22.19 or higher
  • DeepSeek Harness dsh
  • Native Linux or macOS shell execution
  • Windows requires WSL2; bundled Bash scripts do not support native Windows
  • Remote operations require an SCNet account and cluster credentials

Local Sync and Validation

If you need to sync scnet-hpc related content from a local path, you can execute:

./sync.sh --src ../scnet-hpc

This step is used to sync content from the specified directory to avoid manually maintaining duplicate materials.

When doing local package checks, you can execute:

npm install --no-package-lock --ignore-scripts
npm run validate
npm pack --dry-run

These commands are used to install dependencies, validate the package, and perform a dry-run packaging check.

Typical Workflow

You can first understand the usage path in the following order:

  1. Use scnet_list_clusters to list the packed cluster profiles.
  2. Use scnet_show_cluster to read the selected profile, then judge resource constraints.
  3. Use scnet_generate_job to generate accelerator or pure CPU Slurm scripts adapted to the profile.
  4. Use scnet_setup_ssh to configure local SSH keys and host entries after explicit user confirmation.
  5. Use scnet_probe_cluster to generate an initial profile based on discovery from read-only login nodes.
  6. Use scnet_refresh_cluster to refresh profile fields; if compute discovery is needed, enable the optional compute discovery.
  7. Use scnet_run_compute_probe to submit a minimal compute node capability discovery.

Security Boundaries

When using these capabilities, pay attention to the following:

  • The repository and npm package must not contain private keys, tokens, usernames, private endpoints, or local discovery caches.
  • SSH configuration, remote discovery, and Slurm submission are state-changing operations requiring clear targets and user authorization.
  • Generated Slurm files may contain the local username and are ignored by Git.
  • Cluster specifications and scheduling policies vary by profile and need to be verified against the target environment.
  • The MIT license requires retaining the copyright notice and the MIT license notice upon redistribution; software is provided “as is”.

Since it executes within the current dsh session, it is affected by the permissions of the current dsh process. It is recommended to check the source code, dependencies, license, and target profile before installation.

Use Cases

It is suitable for:

  • Operating SCNet clusters in DSH
  • Generating Slurm jobs adapted to profiles
  • Viewing cluster profiles and performing login node discovery
  • Submitting minimal compute node capability discovery

It is not suitable for:

  • Expecting official DeepSeek product endorsement
  • Running bundled Bash scripts directly on native Windows
  • Executing with fixed configurations without verifying target cluster specifications

Links

Community Directory Page: https://www.skillhub.cn/plugins/lql341/dsh-scnet

GitHub: https://github.com/lql341/dsh-scnet