Introduction

The plugin-based architecture of DeepSeek Harness (DSH) allows developers to extend agent capabilities. When an agent needs to operate a remote OpenKapsel workspace, dsh-openkapsel provides a bridge layer. It turns the remote workspace into tools visible to the model, invoked through skill-driven tool calls, instead of directly exposing host-local permissions.

Plugin Overview

This is a remote-only OpenKapsel workspace bridge for the DeepSeek Harness.

Maintainer: zzzmmmnn
Category: Online tool
License: MIT

Core Features

The plugin invokes remote resources through skill-driven tool calls and provides the following capabilities:
- File, Shell, context, memory, sharing, and preview operations for remote OpenKapsel workspaces
- A unified kapsel_* tool interface
- Reuse of Python helper functions from the skill/openkapsel-rest skill (openkapsel_http.py and openkapsel_config.py)

Installation

The plugin is installed through the DSH package manager and is installed into the web profile by default.

dsh plugin --profile web add github:zzzmmmnn/dsh-openkapsel

After installation, the plugin automatically installs a preset named OpenKapsel Remote. This preset contains only the OpenKapsel bridge tools and does not include host file system or Shell tools.

Usage

  1. Select preset: In the DSH mode selector, select the OpenKapsel Remote preset.
  2. Configure connection: Use the kapsel_config tool to provide the URL and control token for the remote workspace. The URL must end with /w/<READ_TOKEN>.
  3. Execute operations: The model uses the provided kapsel_* tools to operate the remote workspace.

Permissions and Security

The plugin strictly separates local and remote permissions:

  • Local permissions: Under the OpenKapsel Remote preset, the model has no host file system or host Shell tools. Local access is limited to running a fixed set of Python helper scripts through the DSH Shell service, and run_code is unavailable.
  • Credential storage: Connection credentials are stored under the user DSH state directory (default: ~/.dsh/state/dsh-openkapsel).
    • Unix systems: file permissions are set to 0600, and directory permissions are set to 0700.
    • Windows systems: directory ACLs are relied upon (chmod does not enforce Unix permissions on Windows).
  • Remote permissions: The model’s operations on the remote workspace are limited by token authorization. read-only in DSH policies rejects remote modifications and Shell operations; workspace-write and danger-full-access remain subject to remote token limits.
  • Credential management: An automatic renewal mechanism may replace stored credentials.

Environment Requirements

  • DSH: DSH 0.1.2-rc.1 or later is recommended (web profile); note that 0.1.5-rc.2 updated the preset persona field.
  • Node.js: Node.js >= 18 (tested with 22 and 24).
  • Python: Python 3.10+, must be available in the system PATH (use python on Windows, and python3 on macOS/Linux).

Notes

  • This is a community plugin, not an official DeepSeek product.
  • Do not add the OpenKapsel Remote preset to the standard or minimal profiles, as they expose host-local tools.
  • The plugin reuses code from the openkapsel-rest skill; the model cannot select scripts or use this service as a Shell tool.