omp packages

cx-listing-playbook0.3.0

A pi package for the work on an Amazon listing: how it performs and why, a log of each change, and how to judge one.

GitHub Actions Published 7 hours ago MIT ~238 downloads/mo

Untested on omp

This is a pi package. No one has reported whether it works on omp yet. If you try it, add your result below.

Did it work for you?

Result

One report per person. A new report replaces your old one.

Reports 0

No reports yet.

Readme

cx-listing-playbook

A pi package for the work on an Amazon listing: creating it, how it performs and why, a log of each change, and how to judge whether a change worked. It is written for the assistant in a ComChatX seller's box, and reads the seller's Amazon account through the box's cx command.

It is text and one script. Its changes are in CHANGELOG.md. Every sentence of the skills ends with the research item it rests on, and the research is in research/.

What it holds

What the model reads
skills/listing-review The funnel and which element serves each stage, what Amazon's own answers show about a listing, the signals that something changed, which file carries each figure, how a write is validated (references/writing.md), and scripts/report.sh
skills/listing-change-log One file per listing change: before, after, what was expected and from which file, the words bet on, the seller's decision, the request as it would be sent, Amazon's receipt id and the observation window
skills/listing-create Creating a listing in the stores the seller names: new item or existing ASIN, the product type and each store's schema, identity, the request built from the seller's data, the preview (sent as a read, since it persists nothing), the write on the seller's word, the read-back until buyable and discoverable, and variation families
skills/featured-offer How Amazon picks the Featured Offer (the Buy Box), how to read whether the seller's offer holds it and why not, which causes a Listings Items write reaches and which sit in Seller Central, the Brand Registry portal or Amazon's own choice, and what the Support API reads
skills/effect-review Experiments, baseline and observation windows, controls, confounders, and the three verdicts: held, not held, cannot tell
prompts/listing-create.md /listing-create <marketplaces> <your words, and where the product's data is>: hands the seller's words to cx, e.g. /listing-create US,DE 新品上架,数据在 /tenant/products/new-router/,FBA 发货
prompts/listing-review.md /listing-review <ASIN>: the listing and its issues, the market since last time, a proposal with its reasons

Each skill's detail is under its references/.

/listing-watch now lives in its own package, @comchatx/cx-listing-watch, which uses this one. Guarding a listing's Featured Offer is @comchatx/cx-featured-offer.

Files in the seller's tree

Place Holds
/tenant/listings/<ASIN>/watch.md The standing instruction /listing-watch (@comchatx/cx-listing-watch) writes
/tenant/listings/<ASIN>/changes/ One file per listing change; changes made together are one change
/tenant/reports/<reportType>/<ask>/<n>/ One report of that ask: receipt.json with the report id, Amazon's status and createdTime, each reply as Amazon gave it, and the document

The report script

scripts/report.sh <reportType> --origin <origin> --marketplace <id>[,<id>...] [--start <ISO 8601>] [--end <ISO 8601>] \
                  [--option <name>=<value>]... --wait <seconds> [--every <seconds>]

It asks Amazon's Reports API through cx raw amazon-sp-api: createReport, getReport until a terminal status or the --wait ceiling, getReportDocument and the document, decompressed, landed beside the receipt. Each run asks for a report of its own, except that it carries on the ask's report not yet landed: one still queued at the ceiling ends the run with its id and status (exit 3) for the next run to carry on. Earlier reports stay as they landed. It has no rate limiter and no retry of its own. Report types, their options and their periods are the amazon-sp-api skill's (references/door.md). It needs bash, jq, curl and gzip, which the box has.

What it needs

pi, and in the box the cx command with the Amazon SP-API source; Keepa where a skill names a Keepa field. It reaches no network itself except the one document link Amazon hands the script.

Install

pi install npm:cx-listing-playbook        # for you, in ~/.pi/agent/settings.json
pi install -l npm:cx-listing-playbook     # for one project, in .pi/settings.json
pi remove npm:cx-listing-playbook

On ComChatX, a seller installs it from the Plugins page.

Develop

npm ci
npm run typecheck
npm test               # pi at the pin loads the package and its tarball and expands every template; every text line cites a research item;
                       # the script runs against a stub cx

TENANT_TREE moves the script's tree from /tenant for a run outside the box, as the tests do.

Licence

MIT. See LICENSE.