Introduction¶

Content automation developers often encounter scenarios where creative directions require manual alignment, scripts and media are generated via different tools respectively, and the final publication involves manual steps. Every step in between requires writing glue code manually, particularly when calling external MCP endpoints, which often necessitates hand-written HTTP requests or scripting glue.

The philosophy of DeepSeek Harness (DSH) is “everything is a plugin,” making such workflows ideal for encapsulation and reuse. Below is an introduction to lussey820’s aitoearn-dsh-plugin, which ports the AiToEarn content creation suite into a DSH plugin to complete the full workflow of “creative guidance → script/image/video generation → publish to TikTok” within a DSH conversation.

What is it¶

aitoearn-dsh-plugin is a port of the AiToEarn content creation suite for DeepSeek Harness, maintained by lussey820 under the MIT license. The installation ID is dsh-aitoearn. The package.json shows version 0.1.0, and it is pure ESM with no build steps required.

It solves the problem of integrating the content creation process into a single DSH session: the Creative Director Agent first clarifies the direction, followed by execution by script/image/video skills, and finally publishing to TikTok. All capabilities are provided as plugins without modifying the DeepSeek Harness source code.

Core Features¶

The plugin consists of four skills: one unified entry point and three executors:

The data flow is unidirectional: the entry point outputs a brief, which executors consume (either the brief or the script JSON), ensuring clear separation of responsibilities between skills.

Regarding low-level connectivity, the plugin accesses AiToEarn MCP endpoints via streamable-http using @deepseek-ai/dsh-mcp-client upon startup. It registers functions like createImageTextDraft, createVideoDraft, getDraftTaskStatus, getMyCreditsBalance, and getDraftGenerationPricing as native mcp__aitoearn__* tools for direct invocation by DSH.

Installation and Enablement¶

First, install using the official command:

dsh plugin --profile demo add github:lussey820/aitoearn-dsh-plugin

When installing from Git, if pnpm requires build permissions, set allowBuilds.dsh-aitoearn to true in pnpm-workspace.yaml and retry. Since this package is pure ESM with no build steps, you can also install it locally with dsh plugin add ./dsh-aitoearn, or install it via a tarball after running npm pack.

After installation, configure the API key by setting the environment variable:

export AITOEARN_API_KEY=your_AiToEarn_API_Key

You can also add this to the profile’s cordis.patch.yml. This step allows the plugin to connect to the AiToEarn MCP endpoints.

If you need to override default behaviors, such as changing the serverName or session name, refer to the plugin’s built-in cordis.patch.yml and override the settings line-by-line in your user patch layer.

Typical Usage¶

Before publishing to TikTok, you need to prepare the agent-browser CLI and a logged-in AiToEarn session:

npm install -g agent-browser
agent-browser --session-name aitoearn open "https://aitoearn.cn" --headed   # Manual login
agent-browser --session-name aitoearn close

First, globally install agent-browser, then use the GUI to open the AiToEarn website for manual login, and finally close the session.

After completing the above steps, you can input tasks directly in a DSH conversation, for example:

帮我做一个赛博朋克咖啡品牌的图文发抖音
→ 创意指导 Agent 追问理清方向 → 输出 brief
→ 图文 skill 生成图文 → agent-browser 发布抖音（扫码确认）

The entire workflow is completed within a single conversation. The publishing step is executed by agent-browser, and scanning the QR code for confirmation is the point of human intervention.

Applicable Scenarios and Notes¶

Suitable users:

1. Already using DeepSeek Harness and want to add content creation and TikTok publishing capabilities to DSH;
2. Have used the Trae version of aitoearn-creative-suite and want to migrate to the DSH ecosystem.

This project is a DeepSeek Harness port of aitoearn-creative-suite (Trae IDE Skill version), preserving the core follow-up decision tree, Cinematography vocabulary list, and brief JSON structure. The differences are in three areas: MCP calls have changed from hand-written HTTP/PowerShell scripts to DSH native mcp__aitoearn__* tools; the API key has been changed from config.json to environment variable injection; and file I/O and browser automation now use DSH native tools (read_file, write_file, bash).

Notes before use:

1. The plugin runs with the permissions of the current DSH process. You should check the source code and license before installing (this project is MIT);
2. Publishing to TikTok depends on the agent-browser CLI and a logged-in AiToEarn session; if these are missing, the publishing step cannot be completed;
3. Calls on the AiToEarn side will consume credits. You can query balances and pricing using registered tools like getMyCreditsBalance and getDraftGenerationPricing.

Conclusion¶

The value of aitoearn-dsh-plugin lies in consolidating the workflow of “clarify direction → generate content → publish,” which was originally scattered across multiple tools, into a single DSH plugin, while reusing the verified decision tree and vocabulary list from the Trae version. If you are working on content automation, it is worth trying out.

• Directory page: https://www.skillhub.cn/plugins/lussey820/aitoearn-dsh-plugin (Community directory, independent site, no official affiliation with DeepSeek / Hyperbolic)
• GitHub: https://github.com/lussey820/aitoearn-dsh-plugin