Foreword¶

The common pain point in creating technical videos or public account content isn’t “not being able to write scripts,” but rather that the entire workflow is fragmented: topic notes are in memos, scripts are in documents, final edits are in export directories, subtitles and thumbnails require separate tools, and publication status relies on mental notes. DeepSeek Harness (DSH) connects agents, file tools, and a plugin ecosystem, with many community extensions for specific scenarios. If you want “a video from topic to publication” to stay local, with AI advancing based on the actual artifacts in the folder, dsh-oil-creator is worth a look—it’s an open-source client-side plugin by maintainer oil-oil, categorized under “client” in the SkillHub plugin library, currently with around 127 stars and 25 forks, under the MIT license.

Note: Community directories like SkillHub are maintained by third parties and have no official affiliation with DeepSeek / High-Flyer; DSH itself adheres to the principle “everything is a plugin.” Before installation, it’s recommended to read the repository source code and license yourself. Plugins run with the current dsh process permissions, so please assess according to your own security boundaries when they involve directory read/write, external APIs, and optional publishing capabilities.

What This Is¶

dsh-oil-creator is positioned as an AI-assisted local creator workbench for DeepSeek Harness. It doesn’t build a closed content database but adheres to the principle “one video, one folder”—content and artifacts remain as ordinary files that any editor and AI file tools can read and write directly; the plugin only saves workbench state, such as project bindings, publication markers, synced playback data, and generation tasks, in ~/.dsh-oil-creator/overlay.json, without duplicating and hiding the content.

The current version in the README indicates compatibility with Node.js 22.19+ and DeepSeek Harness 0.1.0-rc.6 / 0.1.0-rc.7. The core media library can be used independently; capabilities like Screen Studio editing, subtitles, thumbnails, public account articles, and multi-platform publishing can be installed on demand—only install what you need.

A typical directory structure (excerpted from the official README) is as follows:

~/Movies/Video Projects/
└── 2026-08-19_DeepSeek Harness Getting Started/
    ├── topic.md
    ├── script.md
    ├── DeepSeek Harness Getting Started.mp4
    ├── DeepSeek Harness Getting Started.srt
    ├── DeepSeek Harness Getting Started_subtitled.mp4
    ├── DeepSeek Harness Getting Started_3x4.png
    ├── DeepSeek Harness Getting Started_4x3.png
    ├── DeepSeek Harness Getting Started_16x9.png
    ├── publish-package.json
    └── Public Account Article/

Core Features and Highlights¶

The workbench breaks down a piece of content from topic to data collection into collaborative stages, with AI responsible for organizing files and state, and key decisions still confirmed by humans:

Building on this, the plugin also provides these verified capabilities:

1. Local Media Library: Scan directories by date_readable title, display stage, final edit, subtitles, thumbnails, articles, and publication status.
2. Conversational Context: Use @current details, content search, or /current content to pass the target folder to AI, avoiding repeated path descriptions each time.
3. AI Bootstrap Configuration: Built-in creator-workbench Skill; on first run, it can call read-only oil_creator_setup to discover content directories and check optional capabilities, requiring preview and confirmation before writing.
4. Long-Term Script Rules: Save tone, structure, taboos, and target audience, then reuse when writing or modifying script.md.
5. Long Task Tracking: After starting subtitles, thumbnails, and burning, it immediately returns control while the workbench continues to monitor file artifacts and task status.
6. Directory Organization: Preview and correct old folder names; by default, it does not execute or delete files.
7. Optional Publishing Loop: Prepare platform drafts, then humans finalize publication, and finally sync plays, likes, comments, and work links.

The workbench doesn’t pretend to complete recording, editing, or final publication for humans; it’s responsible for placing the files, state, and next actions needed at each step within the same context.

Installation and Enabling¶

In the SkillHub directory page and GitHub README, the official installation command is consistent:

npx @deepseek-ai/dsh plugin --profile web add github:oil-oil/dsh-oil-creator

After installation, restart the web profile to use it. The plugin registers to the configuration, and generally, no manual modification of Harness configuration is needed. If dsh is already installed globally, you can remove npx @deepseek-ai/dsh from the command.

If pnpm explicitly prompts that the build was blocked during installation, you can retry with --allow-build:

npx @deepseek-ai/dsh plugin --profile web add --allow-build=dsh-oil-creator github:oil-oil/dsh-oil-creator

When installing from source, clone the repository and run pnpm install --frozen-lockfile, pnpm build, then add it with the local path add "$PWD". After installing, uninstalling, or updating the configuration, you need to restart dsh web; do not manually modify ~/.dsh/profiles/web/package.json, and do not copy the project’s cordis.patch.yml to the user profile.

Uninstallation command:

npx @deepseek-ai/dsh plugin --profile web remove dsh-oil-creator

Typical Usage Examples¶

1. Let AI Complete the First-Time Configuration¶

It’s recommended to use Harness’s standard or code Agent preset (the minimal preset doesn’t include Skills and file tools, so it’s not suitable for first-time configuration). You can directly tell the AI:

Check and configure the content workbench, find a suitable content directory, and tell me which capabilities are still missing.

oil_creator_setup will look for existing content directories, check optional capabilities like Screen Studio / subtitles / thumbnails / Ego Browser, and only report whether credentials are configured without reading API keys back into the conversation, and preview before saving. When a candidate directory doesn’t exist, it will first display the full path it plans to create, then write after confirmation.

2. Create the First Piece of Content¶

After configuration, you can start working like this:

Today, let’s do an episode on DeepSeek Harness installation and getting started. Create a new content directory, write the topic into the notes, and give me a script draft.

Then continue saying “bind the Screen Studio project from earlier,” “wait for the final edit to generate subtitles and thumbnails,” or “what’s missing for this video,” and the workbench will advance the stages based on the actual artifacts in the folder.

The settings entry is located in Settings → Plugins → Content Workbench. The content directory can be changed to any absolute path, with each direct subfolder representing a piece of content; enabledPlatforms defaults to including Xiaohongshu, Douyin, Bilibili, and WeChat Channels, and disabled platforms won’t participate in AI publishing or data synchronization.

Optional Capabilities and Dependencies¶

The core media library and script management do not depend on the tools in the table below; when something is missing, just disable the corresponding stage.

API keys are saved through Harness’s official credential service; the settings page only displays “configured / not configured” and does not show plaintext.

Applicable Scenarios and Precautions¶

Who it’s for:

• Those who use DSH for technical videos, tutorials, or self-media content, hoping that local files are the truth and AI can advance based on folders;
• Those who already have Screen Studio, subtitles, thumbnails, or publishing Skill ecosystems and want to connect them in one workbench;
• Teams that need long-term script rules, media library views, and multi-platform draft preparation but insist that “humans press the final publish button.”

Precautions:

• The plugin does not automatically upload content; uploading only occurs after the user explicitly calls the publishing Skill and stops before the final publication.
• Subtitles, thumbnails, and platform synchronization access their respective external services; they won’t be enabled if not installed or configured.
• Capabilities like Screen Studio editing, Ego publishing, and data collection have macOS limitations; Linux users can prioritize using the media library, script, and subtitle/thumbnail pipeline (you need to configure the corresponding Skills yourself).
• Directory creation, configuration saving, and batch renaming follow the principle “preview first, confirm, then execute”; before installing the plugin, please read the MIT license and repository implementation.

For a more complete list of tools and file conventions, see docs/usage.md, docs/files.md, and docs/implementation.md in the repository.

Conclusion¶

If you’re tired of copying paths and states between multiple tools, dsh-oil-creator turns DeepSeek Harness into a local creation workbench with one video per folder: the core media library is ready to use, subtitles, thumbnails, editing, and publishing are added on demand, AI handles context and workflow, and key nodes remain in your hands. Why not start with the installation command and run through the entire process with a small topic?

• Plugin directory page: https://www.skillhub.cn/plugins/oil-oil/dsh-oil-creator
• GitHub repository: https://github.com/oil-oil/dsh-oil-creator