FlowCay logo
FlowCay
Ecommerce Agent Browser Skill

Ecommerce Agent Browser Skill

General e-commerce browser automation Skill with a beginner-friendly CC shell UI.

Install command

npx skills add https://github.com/81211860/ecommerce-agent-browser-skill
SkillOpen SourceAll Platforms
Added on Popularity 16

Details

Overview

Ecommerce Agent Browser Skill (by 81211860) is an open-source, general-purpose browser automation skill for e-commerce. Instead of hard-coding a platform's operations, it gives an AI agent a stable, observable, executable and reviewable browser foundation — the model understands pages like a human, picks a path, executes actions, and converts repeatable flows into faster tooling. The code was extracted from the PDD Agent project (Pinduoduo seller backend is the first verified vertical); it's designed to extend to Taobao, JD, Douyin Mall, the PDD consumer side, other seller backends and generic web tasks.

Key Features

  • Persistent browser sessions: persistent Chromium profiles save login state per platform/account (PDD, Taobao, JD, Douyin), isolated so you rarely re-login and never ship login cookies inside installers.
  • Structured page observation: reads DOM, accessibility tree, visible text, forms, lists, tables, popups and scroll state into a lightweight, model-readable page view.
  • Hybrid text + vision perception: combines DOM-based views with VLM (vision-language) analysis and evidence handles for screenshots and region text.
  • Goal-driven execution with feedback: short-id click protocol (reduces model output errors), plus post-action pageChange feedback so the model can continue judging from the actual page change.
  • Scroll exploration & context control: progressive reads keep 1M-context models (DeepSeek, MiMo) efficient without flooding context with duplicate DOM/coordinates.
  • Request capture & CLI-ization: records GET/POST/XHR/fetch via browser context and generates candidate CLI adapters; stable, low-freedom flows get condensed into recipes.
  • Recipe & account management: recipe validation, per-platform/account profile management, and product understanding/listing helpers (product-*.mjs).
  • Beginner-friendly CC shell UI: a Claude Code wrapper desktop app that turns login, asset drag-in, task input, run logs and model config into buttons — no CLI or Playwright knowledge needed for non-technical users.
  • Model guidance: complex long-chain tasks (product publish, price change, SKU edits, cross-page debugging, popup handling) work best with strong reasoning models (e.g. GPT-5.5, Claude Opus 4.7/4.8) at full reasoning; Flash/Pro model routing and diagnostics keep simple tasks on cheap models.

How to Use

  1. Use a Skill-compatible agent such as Claude Code, Cursor, or Codex.
  2. Install from the terminal (Node.js required):
npx skills add https://github.com/81211860/ecommerce-agent-browser-skill

Or paste the install prompt from this tool page into your agent and ask it to install for you. 3. Restart / reload the agent so the new Skill is loaded. 4. Invoke the Skill with its slash commands or by asking the agent to run the workflow described in Overview / Key Features. 5. Review outputs before publishing — treat model-generated copy, images, and data as drafts.

Pricing

  • Free & open source: the Skill and CLI (SKILL.md, Node/Playwright CLI, lib modules, recipes) are free on GitHub. Windows portable builds (e.g. PDD.Agent-v0.1.95-win.zip) are published to GitHub Releases.
  • Cost driver — LLM API: you supply your own model keys; costs follow the model provider's token pricing.

Best For

Cross-border and domestic e-commerce operators using AI coding agents (Claude Code, etc.) who want to automate seller-backend work — product publishing, on/off shelf, price and SKU/inventory changes, store/product-list monitoring, competitor sales observation, customer-service message reading/replying, order/after-sales/logistics/report queries and backend menu exploration — especially Pinduoduo seller workflows, with a beginner UI and a CLI for power users.

FAQ

What is Ecommerce Agent Browser Skill?

A general browser-automation skill layer for e-commerce agents: it perceives web pages, executes goal-driven actions, and consolidates repeatable flows into recipes/CLIs — not a scripted per-platform bot.

Which platforms does it support?

Pinduoduo seller backend is the first verified vertical (pdd-operator + PDD recipes); Taobao, JD, Douyin Mall, PDD consumer side and other seller backends are the planned extensions.

What does it cost?

The Skill is free and open source. You only pay for your own LLM API usage; strong reasoning models are recommended for complex long-chain tasks.

Does it work with weaker/cheaper models?

Yes — DeepSeek/MiMo are optimized for 1M context, lightweight page views, short-id clicks and Flash/Pro routing, but they're best for simple, cost-sensitive or already-experienced flows rather than long-chain autonomous exploration.

How do I run it?

Node.js + Playwright; either use the CLI (scripts/pdd-operator.mjs etc.), the Windows portable package, or the beginner-friendly CC shell UI. Dependency check/install is via scripts/setup.mjs.