DeepSeek Harness (DSH) uses a plugin architecture. For users who need fine-grained content management, version control, and channel publishing, the native interface is often too generic. dsh-wemedia-workbench, as a workflow plugin, aims to address these issues.
What Is This¶
This is a local-first self-media workbench for DeepSeek Harness, maintained by Harzva. It provides a set of tools for AI-assisted content organization, revision management, evidence review, and channel tasks.
Core Capabilities¶
- Local Content Library: Read-only discovery of articles, images, and videos, with support for filtering, source mapping, and an independent write root.
- Articles and Media Drafts: Immutable revisions, asset binding, diff previews, and unsaved change protection.
- Review Evidence: Supports four categories of evidence—facts, editing, images and formulas, and mobile visuals—and binds evidence to specific revisions and file bytes.
- DSH Collaboration: Native tools and PTC share application services and inherit the model and tool permissions of the current session.
- Persistent Tasks: Provides preflight checks, one-time operation intents, native approvals, Job status, and target readback. Unknown results are verified first rather than being blindly resent.
- Batch WeChat Drafts: Sends selected articles or all pending articles in the library. The workflow first confirms the list, then assigns it to the current Agent for per-article approval and delivery. Per-article results can be viewed, and the queue can be stopped.
- Multi-Channel Interfaces: Provides adapter contracts and workbench action entry points for WeChat Official Accounts, Zhihu, Xiaohongshu, and X. Runtime bridges, login environments, and accounts are configured separately by the user.
- Asset Helpers: Provides deterministic formula rendering and PDF chart cropping scripts.
Installation and Build¶
Requires Node.js >=22.19.0 and pnpm 10.16.1.
git clone https://github.com/Harzva/dsh-wemedia-workbench.git
cd dsh-wemedia-workbench
pnpm install --frozen-lockfile
pnpm build
Installing from source does not automatically configure publishing accounts. The channel bridge code does not imply that real-account posting has been verified; the default configuration does not include content directories, credentials, or enabled remote adapters.
Typical Workflow¶
- Organize content from authorized local assets and paper sources, preserving citations.
- Save revisions through the native workflow; source files remain read-only.
- Check facts, wording, images, and formulas; actually review the mobile preview; record evidence for the current version.
- Freeze the content, assets, and report; preflight-check each article and preview the intended actions.
- Approve the exact actions natively, then execute and track the same Job.
- Read back the target and the current revision.
Use Cases and Considerations¶
Use Cases:
Suitable for developers and operators who need to handle multi-channel content publishing and emphasize version control and evidence preservation.
Considerations:
* Early development: The project is in an early-development source-code state, currently pinned to DSH 0.1.1-rc.2 / Cordis 4.0.1, and does not promise compatibility with other versions.
* Unofficial product: This project is not an official DeepSeek product.
* No unified CLI: There is no unified wemedia CLI. Batch WeChat drafts are provided through the workbench and the native tools from the same source; it is not an independent CLI, nor a batch publishing command without approval.
* Permissions and configuration: After installation, you must configure content directories, credentials, and remote adapters yourself. AI assistance in content organization should be clearly disclosed in the articles.
Summary¶
dsh-wemedia-workbench provides DeepSeek Harness with a local-first media workflow solution. Through immutable revisions, review evidence, and native approval mechanisms, it fills the gap between content management and channel publishing.