The core philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” For agent developers who need to handle multimedia content, integrating fragmented model calls, prompt engineering, and workflow orchestration into a stable, automated workflow typically requires writing a large amount of glue code. dsh-media-gen is designed to address this issue. It directly transforms media briefs into model selection, production-ready prompts, and (optionally) executable Atlas Cloud workflows.

Introduction

This is a standalone DSH configuration package maintained by AtlasCloudAI, covering image, video, audio, and 3D workflows. After installation, three core skills become available, along with optional Atlas Cloud MCP bridging capabilities. The plugin mainly addresses the following challenges:

  • Model selection: Identify the exact model IDs, APIs, and schemas that Atlas Cloud should use for image, video, audio, 3D, ASR, or LLM tasks.
  • Video planning: Use the Seedance skill to plan controllable and consistent video shots and reference relationships, outputting storyboards and continuity plans.
  • Prompt generation: Transform a general video brief into a cross-model general prompt specification, as well as model-specific compiled versions.
  • Execution bridging: Use MCP bridging to enable DSH to directly invoke Atlas Cloud operations such as model queries, media uploads, generation submissions, and account usage checks.

Installation and Configuration

  1. Install the plugin:
    Install the plugin into your designated DSH profile (for example, web):
    dsh plugin --profile web add 'github:AtlasCloudAI/dsh-media-gen#v0.2.0'
    dsh --profile web --dump-config
  1. Configure credentials:
    Plugin installation and skill discovery do not require Atlas Cloud credentials, but executing generation operations do. Obtain an API key from the Atlas Cloud console and set it in the process that starts DSH:
    export ATLASCLOUD_API_KEY="<your-key>"
  1. Enable MCP bridging (optional):
    The MCP bridge is disabled by default for security reasons because it is a trusted subprocess that runs outside the sandbox. To enable it, add the following line to $DSH_HOME/profiles/<profile>/cordis.patch.yml in the target profile directory:
    - id: atlascloud-mcp
      disabled: false

After restarting the profile, DSH will expose tool names such as `mcp__atlascloud__atlas_list_models`.

Typical Usage

In a DSH session with this plugin enabled, you can directly enter an instruction to let the plugin handle a media brief. For example:

Plan an 8-second coffee product video. Compare Seedance with one available alternative, recommend a model, and produce a three-shot storyboard, final prompt, and exact parameters. Do not submit generation; wait for my confirmation.

DSH will use its built-in skills to analyze the brief, compare model options, and output a detailed plan including shot storyboards, final prompts, and parameters, without automatically submitting generation requests.

Notes

  • License: This project follows the MIT license.
  • Compatibility: The DSH configuration package uses dsh.bundle.patch and Cordis MCP lines, which use different host manifests than OpenAI/Codex plugins. Although skill content can be reused, host integrations are not directly interchangeable; you must use package.json and the corresponding DSH configuration format.
  • Security: MCP servers are disabled by default. It is recommended to enable them only in trusted environments to avoid potential security risks.
  • Verification: It is recommended to review the source code and license before installation.

Summary

By encapsulating Atlas Cloud capabilities as DSH skills, dsh-media-gen enables workflow automation from requirement analysis to execution. For more technical details and source code, refer to its GitHub repository or skill catalog page.