Preface

People who manage repetitive content accounts often encounter an unintuitive problem: the copy itself may not necessarily be bad, but platforms will mark this content as “low creativity” and lower or disable public recommendations. Common triggers include repeatedly using the same blessing text, fixed title prefixes, fixed topic tags, multiple pieces of content with similar structures on the same day, or five pieces of content sharing the same structural skeleton.

Under the plugin-based usage of DeepSeek Harness (DSH), dsh-wechat-mp-studio offers a more engineering-oriented processing path: organizing “anti-homogenization writing”, “visual baseline”, “image generation and OCR acceptance”, “interaction guidance”, and “draft syncing” into a set of executable workflows and scripts. It does not promise traffic; its main purpose is to transform one-off experiences into production steps that are checkable, reusable, and continuously iteratable.

What is this

dsh-wechat-mp-studio is a DSH plugin maintained by FuncWei, with the package name:

dsh-wechat-mp-studio

Its positioning is:

WeChat Official Account content studio for DeepSeek Harness

The license is MIT. It targets WeChat Official Account content production, providing anti-homogenization writing methods, low creativity remediation strategies, visual baselines for blessing cards, scripts for image generation and OCR verification, interaction guides, and draft syncing references based on the measured xiaolvshu (newspic) draft web API.

It is more suitable for accounts that already have fixed content formats, require continuous output, and wish to reduce the risks of structural repetition and copy homogenization.

Core Features

Based on the capabilities listed in the plugin, it mainly includes the following categories:

  1. Rotation writing method
    Used to avoid the structural inertia of “what was written yesterday, writing again today,” emphasizing structure signature rotation and anti-recency selection.

  2. Daily delta test
    Used to check if there are recognizable differences between daily content, reducing the situation where multiple pieces of content cancel each other out.

  3. Low-creativity remediation
    For cases marked as low creativity, provides root cause analysis, rectification directions, recovery rhythm, and other handling methods.

  4. Visual baseline
    Establishes a visual baseline for content such as blessing cards, holiday cards, and greeting images to prevent the same type of images from staying in the same visual perception for a long time.

  5. Image pipeline with OCR acceptance
    Provides a flow for image generation and image text verification. The scripts include generate_images.py and ocr_verify.py.

  6. Interaction guide
    Provides interaction guidance rules, including primary action points, comment guidance, red lines, and closing expressions.

  7. Publishing via measured xiaolvshu (newspic) draft web API
    Provides measured references for the xiaolvshu (newspic) draft web API, used to push content to the draft box.

  8. Structure log and image log templates
    Provides structure log and image log templates for recording the structure and image information of each publication, used for anti-recency selection in the next publication.

Installation and Enablement

The official installation command is:

dsh plugin --profile web add github:FuncWei/dsh-wechat-mp-studio

After installation, the model will load the wechat-mp-studio skill on demand in relevant tasks, such as writing Official Account content, generating images, syncing drafts, and handling low creativity markers.

It can also be explicitly requested in instructions:

use wechat-mp-studio to draft today's post

The package.json declares Node environment requirements:

{
  "engines": {
    "node": ">=18"
  }
}

Therefore, before running, you need to confirm that the current DSH environment meets Node 18 or higher.

Typical Usage

The following explains the process in the order of “write copy, generate images, verify text, sync drafts, record logs.”

1. First determine the account positioning, then produce the copy

First, tell the model the account positioning (e.g., audience, persona, monetization), and then have the workflow produce the copy and structure signature.

The goal of this step is not just to write the body text, but to ensure that every publication has a structure identifier that is recordable and comparable. The subsequent structure logs will rely on this identifier to avoid falling back to the old structure in the next post.

2. Generate cover image or body image

Before generating images, you need to set IMAGE_API_KEY to a gpt-image-compatible endpoint.

Then execute:

python3 skills/wechat-mp-studio/scripts/generate_images.py cover \
  --text "<today's blessing>" \
  --anchor "<subject>"

This command is used to generate a cover image or body image. --text contains the text to be displayed that day, and --anchor contains the main subject of the image.

3. Perform OCR verification on image text

After the image is generated, do not rely solely on visual inspection. The plugin provides an OCR verification script:

python3 skills/wechat-mp-studio/scripts/ocr_verify.py \
  --image cover.png \
  --expected "<today's blessing>"

This command reads the generated cover.png and compares the text in the image with the expected text. For accounts that strongly rely on image text, such as blessing texts, holiday cards, and greetings, this step is used to reduce the risk of typos, missing characters, and inconsistencies between text and images.

4. Push to draft box

Draft syncing references references/xiaolvshu-web-api.md to push content to the WeChat MP draft box.

Please note: This step is drafts only; the actual publishing action is still confirmed by a human. The plugin provides draft syncing and workflow tools and does not replace human publishing judgment.

5. Record structure logs and image logs

After each publication, append a line to the structure log and image log respectively.

When publishing the next time, the workflow will read these logs to perform anti-recency selection. In other words, logs are not archive files, but context participating in the next selection.

File Layout

The main directory structure within the plugin is as follows:

skills/wechat-mp-studio/
├── SKILL.md                              # Entry: eight-step pipeline + red-line checklist
├── references/
│   ├── anti-homogenization.md            # Rotation writing method
│   ├── low-creativity-recovery.md        # Low-creativity remediation playbook
│   ├── visual-baseline.md                # Blessing-image visual baseline
│   ├── gpt-image-2-prompts.md            # Image prompts, anti-typo rules, OCR SOP
│   ├── xiaolvshu-web-api.md              # Measured xiaolvshu draft web API
│   └── interaction-and-hooks.md          # Interaction guide & share hooks
├── scripts/
│   ├── generate_images.py                # Cover/post image generation (templates built in)
│   └── ocr_verify.py                     # On-image text OCR acceptance (dual backends)
└── templates/
    ├── structure-log.md                  # Structure log template
    └── image-log.md                      # Image log template

Structurally, it places methodology, reference documentation, scripts, and log templates in the same skill directory, making it easy to load by task within DSH.

Suitable Scenarios and Notes

Suitable scenarios include:

  1. Repetitive content accounts publishing daily fixed blessings, greetings, holiday cards, maxims, etc.;
  2. Accounts that need to reduce structural, title, and topic tag repetition;
  3. Accounts using the gpt-image-compatible endpoint to generate blessing images and needing to perform OCR verification on the image text;
  4. Accounts that need to push finished products to the WeChat MP draft box but still require human confirmation for publishing;
  5. Accounts that have encountered “low creativity” markers and need to organize rectification and recovery rhythms according to the documentation.

Pre-use notes:

  • The plugin license is MIT, but the plugin runs with the current dsh process permissions. You should check the source code and license before installing.
  • WeChat MP web APIs (such as operate_appmsg) are internal, undocumented, and may change at any time.
  • The xiaolvshu parameters in the plugin are point-in-time measurements; you should re-verify them according to Section 4 of references/xiaolvshu-web-api.md before use.
  • Third-party services required by the scripts, such as gpt-image, MiniMax, and OpenCode Go, require you to provide your own keys/quota.
  • Draft syncing only pushes to the draft box; the publishing action is still confirmed by a human.
  • The plugin provides methodology and tools, but does not guarantee traffic. You should follow platform rules and local laws when using.

Links

GitHub repository:

https://github.com/FuncWei/dsh-wechat-mp-studio

The directory page given in the prompt is:

https://www.skillhub.cn/plugins/FuncWei/dsh-wechat-mp-studio

This directory page URL comes from the plugin prompt and does not appear in the scraped README or package.json. Please confirm it yourself before use.