AI Cross‑Border Ecommerce Agent Toolkit

What it is – A collection of seven ready‑to‑use AI Agent “skills” that turn a cross‑border ecommerce SOP into a reproducible workflow. Each agent bundles:

• a system prompt defining the role and contract,
• input‑output templates,
• small helper scripts (e.g., market‑metric calculators, compliance checks),
• example data and a checklist for self‑validation.

The toolkit does not provide a SaaS platform; you run the agents locally by feeding the prompts and templates to any LLM (OpenAI, Claude, etc.).

---

Core agents (7)

#	Agent	Core job
01	Market Research Agent	Parses category data, computes concentration, price‑band, and opportunity metrics (via bundled scripts).
02	Product Strategy Agent	Determines positioning, SKU mix, pricing model, profit‑margin matrix.
03	Listing Creative Agent	Generates keyword groups, titles, bullet points, and spec tables; includes character‑count and prohibited‑word linting.
04	Product Image Agent	Plans a 9‑image grid, creates image‑generation prompts, runs pixel‑level compliance checks.
05	Product Video Agent	Writes script & storyboard, builds AI‑video prompts, adds voice‑over and subtitles.
06	Advertising Optimization Agent	Diagnoses funnel stages, calculates break‑even CPC, suggests tiered actions (P0‑P2).
07	Review Analysis Agent	Classifies reviews into eight attribution categories, spots misunderstandings, and proposes listing tweaks.

All agents are marked ✅ v0.4 (stable) and are packaged as SKILL.md files that can be imported into a skill‑based LLM framework.

---

How to run (no API keys needed)

1. Copy the templates/ folder of the desired agent.
2. Fill in your product information (≈10 min).
3. Concatenate the system prompt, task prompt, and the filled template.
4. Send the combined text to any LLM of your choice.
5. Verify the output with docs/04_Result_Checklist.md before publishing.

The repo includes a full end‑to‑end example (electric scooter) that walks through the entire chain from market research to ad/review feedback loops.

---

Compliance focus

The original internal SOP contains prohibited tactics (e.g., fake orders, false claims). The toolkit excludes those actions and provides a 03_Compliance_Guard.md file that lists the red lines and compliant alternatives. A separate 05_SOP_Gap_Register.md documents every omitted item.

---

Platform coverage

Platform	Support level	Key constraints
Shopee (Brazil)	✅ First‑class	120‑char title limit, 9‑image white‑background grid, PT‑BR language
Mercado Livre	✅ First‑class	60‑char title, clean background images, PT‑BR/ES
Amazon	📄 Fields mapped	200‑char title (recommended 80), pure white background
TikTok Shop	📄 Fields mapped	Vertical 9:16 images preferred
Shopify	📄 Fields mapped	No hard limits

Detailed field specs are in docs/02_Platform_Output_Spec.md.

---

Roadmap snapshot

• v0.3 (current): all 7 agents, data‑parsing scripts, compliance guards.
• v1.0: One‑click material‑package generation for direct upload.
• v2: API integration to pull live data from Shopee / Mercado Livre.
• v3: Web UI for visual workflow orchestration.
• v4: Fully automated, scheduled agents with closed‑loop optimization.

---

Who might use it

• Cross‑border sellers who want to codify expert SOPs into repeatable LLM‑driven processes.
• Agencies building internal tooling for multiple client stores.
• Developers looking for concrete prompt‑engineering examples and reusable skill bundles.

---

License

MIT – free to use, modify, and share.

---

Bottom line – The toolkit is a practical, compliance‑first blueprint for turning seasoned ecommerce knowledge into modular AI agents that can be run with any LLM, without needing a hosted service.