The core philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” When developing and running DSH, you often need to manage AI employees, knowledge bases, skills, MCP servers, and the plugins themselves simultaneously. staff-os/dsh-workbench provides a centralized workbench that brings these five domains into a unified management interface.

What Is This

This is an enterprise-grade workbench plugin for DeepSeek Harness, maintained by staff-os and licensed under MIT. It does not depend on external backends or databases. All state is stored as local files under the $DSH_HOME directory, and it can run standalone.

Core Features

The plugin provides tools for five domains through the ctx.tools mount point:

  1. Skill Management

    • Local skills are managed directly as directories, with each directory containing SKILL.md.
    • Supports installing skills from marketplaces. A DSH restart is required after installation for changes to take effect.
    • Provides security scanning based on regex rules and character-set detection to prevent potential risks.

  2. AI Employees

    • Manages agent presets and supports creating new employees by copying existing presets.
    • Presets with system-level trust cannot be edited or deleted.
    • Knowledge base binding is declared only and does not automatically trigger retrieval.

  3. Knowledge Base

    • Supports local chunking and BM25 keyword retrieval, without relying on vector models.

  4. MCP Servers

    • Configuration is written to the patch file of the current profile.
    • Uses the import_json interface to convert environment variable references (such as ${VAR}) into JS execution expressions (!!js process.env.VAR).
    • Configuration changes take effect after the next DSH startup.

  5. DSH Plugins

    • Forwards operations to the dsh plugin CLI.
    • It does not involve pnpm or directly modify dsh.profile.bundles; plugin status is determined through listing.

Installation and Enabling

Run the following command in the terminal to install it:

dsh plugin --profile web add @staff-os/dsh-workbench

For local development, you can use an absolute path link:

dsh plugin --profile web add link:/path/to/dsh-workbench

After installation is complete, restart DSH.

Configuration

The plugin is configured through environment variables and patch files.

Environment Variables

Configure them in $DSH_HOME/.env or in the .env file in the same directory where dsh is invoked:

  • DSH_PROFILE: Specifies the profile for MCP and plugin management. The default is web.
  • CLAWHUB_REGISTRY: Root URL for a ClawHub-compatible marketplace. If not set, the built-in public endpoint is used.
  • CLAWHUB_API_KEY: Marketplace credential. It is used only as a reference name; the actual value is resolved by the credential service or the startup environment and is not written inline into configuration.

Patch Layer Configuration

Edit the profile’s own patch file $DSH_HOME/profiles/<name>/cordis.patch.yml. Configuration entries must be fully rewritten (not deep-merged):

- id: workbench
  config:
    profile: web
    dshHome: ''
    registries:
      - id: skillhub
        name: Intranet skills
        url: http://10.0.14.55
        flavor: clawhub
        apiKeyEnv: SKILLHUB_REGISTRY_API_KEY
    registryTimeoutMs: 15000
    toolTimeoutMs: 20000
    networkToolTimeoutMs: 120000
    dshExecutable: dsh
    pluginToolTimeoutMs: 300000

The flavor option determines the marketplace interface style: clawhub (default) or skillhub.

Marketplace Source Management

Skill marketplace sources can be added or removed in the UI. Configuration is written to $DSH_HOME/workbench/market.json and takes effect in real time.

Runtime and Notes

  • Environment Requirements: Node.js ^22.19.0 || >=24.0.0 is required.
  • Destructive Operations: Operations such as deletion or overwriting must explicitly set confirm: true in the tool parameters; otherwise, the operation will be rejected.
  • State Synchronization: Skill writes are synchronized immediately; MCP server changes require a restart.

Applicable Scenarios

This plugin is suitable for users who need to centrally manage AI employees, knowledge bases, skills, MCP servers, and plugins within DeepSeek Harness. It provides a complete management loop from CLI to UI, with all data maintained as local state.

References

  • Plugin directory: https://www.skillhub.cn/plugins/staff-os/dsh-workbench
  • Source code: https://github.com/staff-os/dsh-workbench