Cross-Border E-Commerce AI Knowledge Base

Read it as a book. Install it as agent capability. Run it as operations.

Every figure sourced, dated or marked as an estimate · Every prompt ends with a self-check · Every guide states where it stops working — all checked in CI

🇺🇸 English · 🇨🇳 中文 · 🇯🇵 日本語  |  📖 Read Online  |  📦 Install for Agent

Commerce Agent OS on a seeded demo tenant — verified evidence, pending human approvals, and the agents that produced them, on one screen.

In Claude Code — installs the 9 skills; only their names and descriptions load until one is used:

/plugin marketplace add kangise/ecommerce-ai-skills
/plugin install ecommerce-ai-skills@ecommerce-ai-skills

Run the operations UI locally:

pip install "ecommerce-ai-skills[mcp] @ git+https://github.com/kangise/ecommerce-ai-skills"
opc-ecommerce demo-seed --db ./demo.sqlite && opc-ecommerce demo --db ./demo.sqlite --port 8788
# then open http://127.0.0.1:8788/app

What This Is

An AI operations knowledge base for cross-border e-commerce. One source, three uses:

Use	What it is	Entry
Read it	69 chapters, sourcing to growth, complete in zh/en/ja	Online site
Install it for your agent	Knowledge + domain model + guarded capabilities: a Claude Code plugin, or one MCP config line for Claude Desktop / Cursor	dist/
Run it	Commerce Agent OS — an operations runtime on real store data, with multi-agent review and human approval	opc-ecommerce demo

All three are held to the same CI gates. Gates fail, nothing ships.

Run It — Commerce Agent OS

The first two uses hand knowledge to an AI. The third puts it to work — the same ontology and skills inside a runtime with approval, audit, and failure recovery.

pip install "ecommerce-ai-skills[mcp] @ git+https://github.com/kangise/ecommerce-ai-skills"

opc-ecommerce demo-seed --db ./demo.sqlite     # isolated demo tenant
opc-ecommerce demo --db ./demo.sqlite --port 8788
# open http://127.0.0.1:8788/app

The demo tenant is marked DEMO DATA throughout and is physically isolated from real data.

What it does today

Capability	Detail
Real data in	Amazon SP-API / Amazon Ads / Shopify connectors; Business Report, Ads, FBA, returns, and listing CSV/XLSX import once as durable Evidence
Multi-agent review	Weekly Ops Council: an evidence agent and one specialist per marketplace review in parallel, then a cross-platform controller and store manager set priorities
Human approval gate	Writes go through proposal → approval → execution; advertising actions carry an extra capability gate and are blocked without authorization
Durable execution	Four worker classes (job / schedule / report-sync / daily-ops) resume after interruption, with idempotency leases and recovery
Full audit trail	Every conclusion must cite its input source; tenant isolation, API key rotation, operation audit log
Operations UI	/app with seven views — briefing, agents, evidence, approvals, connections, automation, audit; persisted zh/en/ja preference, light and dark themes