DeepSeekCreativeWritingAI score · 76/100AI-assisted edit · no obvious prompt-injection riskGitHub

DeepSeek Harness WeChat Article Plugin: From Topic Selection to WeChat Drafts via Chat

Install a plugin in DeepSeek Harness and use natural-language chat to handle topic selection, writing, review, layout, image generation, and publishing prep for WeChat Official Account articles

ExperiencedStepsPromptTip
Who
Content operators or developers already using or willing to deploy DeepSeek Harness locally who need to produce WeChat Official Account content
You get
After installing the plugin and configuring the WeChat Article Skill, describe your article needs in chat to generate Markdown drafts, styled HTML, cover images, and inline images locally; with credentials configured, push directly to the WeChat drafts box
You need
DeepSeek

Constraints

Prerequisites: macOS or Linux terminal, Node.js 22.19+ (or 24+), pnpm 11, git. Currently RC stage; install DeepSeek Harness and the plugin from source. After installation, configure account positioning, content strategy, writing style, visual style, and optional WeChat publishing credentials via the Web UI. All articles and assets stay in your local project directory.

What they did

aiworkskills developed the aws-wechat-article plugin for DeepSeek Harness, bringing WeChat article production into Harness via Skills. Users install the plugin and the separate wechat-article-skills repository from source, configure account info in the Web UI, then drive the full pipeline—from topic selection to publishing prep—through natural-language chat.

How it went

Author report: after setup, users initiate writing requests in natural language; the agent advances through Skill rules and pauses at key decision points for confirmation. The output includes a Markdown final draft, HTML layout file, cover image, and inline images, all saved in the local project directory. If WeChat Official Account credentials are configured, the article can be pushed directly to the WeChat drafts box.

Worth knowing

1) DeepSeek Harness is at 0.1.1-rc.2; the plugin requires source installation and comfort with terminal commands. 2) The WeChat Article Skill is maintained independently from the plugin and can be used in other Skill-compatible tools; the methodology is open in the wechat-article-skills repo. 3) The config tool at aiworkskills.cn/config is provided by the author, not this site. 4) The README does not detail how to obtain or enter WeChat appId/appSecret; users must prepare these credentials separately. 5) This site has not installed or run the plugin; generation quality and WeChat draft push behavior are unverified.

Steps

  1. Install the plugin from source

    Run the following commands in your terminal to clone DeepSeek Harness and the aws-wechat-article plugin, install dependencies, build both, register the plugin, and start the web server.

  2. Open a WeChat project workspace

    In the DeepSeek Harness Web UI (default http://127.0.0.1:3081/), create or open a session and select a local folder as the workspace. This folder will store the project’s WeChat config, writing assets, and article files. Click the WeChat icon at the bottom of the left sidebar to open the WeChat workspace panel.

  3. Install the WeChat Article Skill

    On first use, click Install WeChat Article Skill. The plugin pulls aiworkskills/wechat-article-skills from GitHub (9 Skills defining the full pipeline from topic selection to publishing). This is a one-time install; later it shows Update Skill. If skillSourceDir is preconfigured in your deployment, this step is skipped and Update Skill appears directly.

  4. Complete WeChat account configuration

    Fill in five categories of settings and save: 1) account positioning, target readers, default author; 2) content strategy (article types, headline directions, structure); 3) writing guidelines (tone and content requirements); 4) layout, cover, and inline image style; 5) WeChat Official Account publishing info (optional, for auto-push to WeChat drafts). You can later adjust settings via Skill Config in the top-right, or use online config, the standalone config site, or the local config folder.

  5. Write articles in chat

    After configuration, type your writing request directly into the main chat. You do not need to click Start New Article in the right panel. The agent follows the Skill flow and pauses at decision points. If you exit midway, you can ask it to read existing article files and continue.

    Execution prompt · Source text。DeepSeek Harness main chat input

    写一篇介绍 DeepSeek Harness 插件体系的公众号文章,面向有 TypeScript
    经验的开发者。先给我几个选题方向,确认后再继续写。
  6. Review and revise article files

    The Creation panel on the right lists articles by most recent update. Expanding an article shows the Markdown draft/final, HTML layout file, cover image, and inline images. Click a file to preview; click the quote button to bring a draft or image into the main chat, then request changes.

    Execution prompt · Source text。DeepSeek Harness main chat input (after quoting an inline image)

    重新生成这张图片,保持原来的内容,但改成更简洁的科技媒体风格。
  7. Manage writing assets

    Use Writing Assets to create content directions (e.g., WeChat Skills, DeepSeek Harness), upload documents and images, and support drag-and-drop, batch import, preview, rename, and delete. When writing, tell the agent which content direction to use; it will read related materials per Skill rules and prefer images from that direction for inline images.

Prompt you can copy

Reference prompt · Editorial suggestion。Confirm the destination, input variables and output format before use.

This block is the prompt. Ads are never placed here.

No prompt needed — just follow the steps.

After installation, simply describe your article needs in the Harness main chat to start the workflow.

Source

在 DeepSeek Harness 里装公众号写作插件,对话里先出选题再写成稿、改配图 · aiworkskills · GitHub · 2026-08-25T23:49:58Z

Comments & feedback

Tried this method? Share results, problems or improvements. Comments appear after moderation. Please avoid sensitive information.

Loading…

    Related