Introduction

DeepSeek Harness (DSH) uses a plugin-based architecture, allowing developers to integrate skills with specific capabilities into the main framework. For e-commerce operations, customer service leads, or store operators, the most time-consuming daily tasks are handling negative review responses, writing competitor listing copy, and calculating listing profits. If general-purpose AI chat tools are used directly, they often suffer from context loss, repeated entry of store policies, and unstable profit calculation formulas.

dsh-shop-assistant is a workflow plugin designed for this scenario. It leverages DeepSeek Harness capabilities to encapsulate file reading, spreadsheet processing, and fixed calculation formulas into workflows. It supports Chinese-language environments and aims to provide reproducible outputs and reduce repetitive operations.

Plugin Overview

dsh-shop-assistant is a workflow plugin customized for e-commerce scenarios, maintained by developer pengzhou267-ai, and released under the MIT license.

It mainly addresses three core problems:
1. Bulk negative review replies: Reads CSV review data and generates replies based on store policies.
2. Competitor listing copy generation: Extracts information from public competitor pages and helps generate listing copy.
3. Profit scoring: Calculates profit margin and recommendation level using a fixed formula, avoiding arbitrary AI guesses.

Core Features

Based on verified facts, the plugin has the following features:

  • Bulk negative review replies: Supports reading CSV review data, generates replies by group, and strictly follows after-sales policies.
  • Listing copy generation: By reading public competitor page URLs, automatically summarizes page information and drafts titles, bullet points, and FAQs.
  • Pre-listing profit check: With inputs for cost, selling price, and scores, uses a fixed formula to calculate profit margin and recommendation conclusion.
  • Reproducible scoring: Provides a reproducible product scoring mechanism, ensuring the same result when the same parameters are entered multiple times.
  • Store policy knowledge base: Supports uploading after-sales policy files as a knowledge base to ensure compliant response wording.
  • Chinese skills: Natively supports Chinese interaction and data processing.

Installation and Enablement

Before using the plugin, ensure DeepSeek Harness is installed and running.

  1. Run the DeepSeek Harness Web UI (for example, using npx @deepseek-ai/dsh web).
  2. Run the installation command in the terminal:
dsh plugin --profile web add dsh-shop-assistant
  1. After installation, restart the Web UI or start a new session.
  2. In the chat interface, select a “workspace folder”. The plugin will only read files inside that folder.

Typical Usage

1. Bulk Negative Review Replies

Operators usually need to export a review CSV file from the merchant backend and paste each entry into an AI tool for replies. This can lead to overly long context and easily missed policy details.

Steps:
1. Place the exported review CSV file (must be UTF-8 encoded) into the workspace folder and name it reviews.csv.
2. Place the store’s after-sales policy document (such as after-sales-policy.md) in the same folder.
3. Send a prompt in the DSH chat box asking the assistant to read the spreadsheet and reply by policy group.

The workspace has reviews.csv and after-sales-policy.md.

Please use the "read review spreadsheet" feature to open reviews.csv.
Do not ask me to paste the table into chat.

Then:
1) Group bad reviews by reason;
2) Write paste-ready replies for each group;
3) Strictly follow the policy file.

2. Competitor Page Copy Generation

Copying titles and selling points from competitor pages is inefficient and may miss key information.

Steps:
1. Open a public competitor detail page in the browser and copy its URL.
2. Paste the URL in DSH.
3. Instruct the assistant to fetch page information and generate copy.

First, fetch information from this public product page (title, description summary, visible price clues).
URL: [粘贴竞品公开链接]

Then output:
1) 5 title options;
2) five bullet points;
3) 5-8 FAQs.

3. Profit Check and Scoring

When deciding listing prices, manually calculating profit margins is error-prone, and results may be inconsistent across different times.

Steps:
1. Prepare input data: cost, selling price, competitor price (optional), and scores for demand, competition, operations difficulty, risk, timing, etc. (1-5 points).
2. Send a prompt to invoke the fixed profit scoring feature.

Please use the "profit scoring" feature (fixed formula) to explain:
unit profit, margin rate, total score, and recommendation (strongly recommend / caution / not recommend).

Cost 35, sell price 99, competitor 109;
demand 4, competition 3, ops difficulty 2, risk 2, timing 4.

Notes and Ecosystem

  • File permissions: The plugin only reads files inside the current workspace folder and does not access other paths.
  • Encoding requirement: Imported CSV files must be UTF-8 encoded, otherwise they may not parse correctly.
  • Advanced configuration: To customize the knowledge base path, set kbRelativeDir in the bundle config.
  • License: The plugin is released under the MIT license and is open source and commercially usable.

As part of the DeepSeek Harness ecosystem, this plugin is published through the SkillHub directory (https://www.skillhub.cn/plugins/pengzhou267-ai/dsh-shop-assistant). It has no direct affiliation with the official DeepSeek or High-Flyer entities and is a skill contributed by a community developer.