A 38-page ops manual, distilled into seven prompt workflows

What it does

The repo decomposes an internal 38-page Shopee Brazil operations manual into seven agent modules covering the full listing lifecycle: market research, product strategy, listing copy, images, video, ad optimization, and review analysis. Each module ships a system prompt, input/output templates, helper scripts, and a real worked example. It’s not software or a SaaS — you paste the prompt plus a filled template into any LLM and check the output against a checklist.

The interesting bit

The philosophy is “let scripts compute, let the model interpret”: market concentration, price bands, break-even CPC, and review attribution counts are all calculated deterministically, with the agent only doing judgment work. The loop also closes — agents 06 and 07 feed ad and review data back into revising the listing, which is what separates this from one-shot generation tools.

Key highlights

• Seven agents form a complete chain: analysis → strategy → creative → launch → data-driven revision
• Deterministic scripts for metrics (pricing matrices, title linting, pixel-level image compliance checks)
• Ships the original unredacted 38-page SOP PDF as source material, alongside a compliance guard and gap register documenting which gray-area tactics were deliberately not implemented
• Skill-packaged (SKILL.md entries) so the whole folder can be installed as an agent skill
• No API key or environment setup required — works with any LLM

Caveats

• The bundled PDF is an unedited internal manual that includes platform-prohibited practices (review manipulation, fake listings parameters); the repo documents compliant alternatives but you’re working next to raw gray-area material
• v0.3 beta status; examples center on Shopee Brazil and Mercado Livre, so transfer to other markets is untested
• Value depends heavily on the quality of your input data — the templates are only as good as what you fill in

Verdict

Worth a look if you run cross-border listings on Shopee or Mercado Livre and want to systematize operator judgment into repeatable prompts. If you’re hoping for a runnable product or a general-purpose ecommerce agent framework, this is methodology and glue — well-organized glue, but glue.