dsh-shop-assistant

中文 | English

For shop owners, CS leads, and operators — you do not need to write code.

You do not need to know English tool names. Follow the cases step by step.

In one minute

After you install this DeepSeek Harness (dsh) plugin, you can do three jobs in chat with plain language:

1. Batch bad-review replies — put an exported review spreadsheet in a folder; get many paste-ready replies that follow your return policy.
2. New listing copy from a competitor page — paste a public product URL; the assistant summarizes the page, then drafts titles, bullets, and FAQs.
3. Go / No-Go before listing — give cost, price, and 1–5 scores; a fixed formula computes profit and a recommendation (not a made-up guess).

This is not “just another chatbot.” Versus pasting into a web AI chat, you get whole-table handling, stable policy wording, reproducible math, and less copy-paste.

---

Install

1. Run DeepSeek Harness (e.g. npx @deepseek-ai/dsh web).
2. Install this plugin:

dsh plugin --profile web add dsh-shop-assistant
# or
dsh plugin --profile web add github:pengzhou267-ai/dsh-shop-assistant

3. Restart the Web UI or open a new session.
4. Pick a workspace folder (next section), then chat.

---

Before you start: where do files go?

What is the “workspace”?

It is the folder you select when you start dsh Web chat.
The assistant reliably reads tables and docs inside that folder only.

Suggested layout:

my-shop-files/
├── reviews.csv           ← your exported reviews
├── after-sales-policy.md ← your return rules
└── (optional) products.csv

Try without your own data first?

Copy samples from this package into the workspace:

File	Use
examples/reviews.csv	Fake reviews for case 1
examples/products.csv	Fake products
examples/score-inputs.csv	Numbers for scoring
kb/sample/售后政策.md	Sample return policy (edit before real use)

Header formats: examples/README.zh.md (Chinese; table headers are the same).

---

Case 1: Batch bad-review replies (daily)

How people usually do it

Copy reviews one by one from the seller console → paste into ChatGPT / DeepSeek web → re-explain return rules every time → paste replies back. Long threads blow up; wording drifts.

Prepare

1. Export reviews from Taobao / Pinduoduo / etc. Save as CSV UTF-8 if needed.
2. Prefer columns like: order id, rating, review text, date, SKU (see examples/reviews.csv).
3. Put the file in the workspace, e.g. reviews.csv.
4. Put return rules in after-sales-policy.md (start from kb/sample/售后政策.md).

Steps

1. Open dsh Web; set workspace to that folder.
2. Confirm you can see reviews.csv and the policy file.
3. Paste and send:

The workspace has reviews.csv and after-sales-policy.md (or 售后政策.md).

Please use the “read review spreadsheet” feature to open reviews.csv
(use the Taobao-style column mapping if headers look like a Taobao export).
Do not ask me to paste the table into chat.

Then:
1) Group bad reviews by reason (shipping delay, color mismatch, damage, size, …);
2) Write paste-ready replies for each group;
3) Strictly follow the policy file — no promises that are not written there.

…