nexscope-ai/eCommerce-Skills: the header claims 157 skills and the install command says 142, and a profit calculator in markdown carries no version

The header claims 157 skills and the install command installs 142

Two numbers for one collection, both in the first screenful of the document. The header line advertises 157 free agent skills for sellers on six named platforms. Eleven lines later, under Quick Start, the sentence introducing the command you actually run says install all 142 skills at once. Fifteen skills are advertised and not delivered by the command the document tells you to start with. Nothing in the visible text accounts for the difference, and there is no note saying that fifteen are being held back, excluded for one platform, or coming soon. The second command takes a single skill by name, and the example given is the growth strategy skill, which is also the first entry in the most-popular list, so at least the named example exists. The install mechanism is a package runner invoking a third-party command, not a clone or a copy:

npx skills add nexscope-ai/eCommerce-Skills -g

The global flag is the only option shown on both forms, so there is no documented per-project install. The two commands differ by a single argument, and both fit on one line, which is the strongest claim the document makes about its own usability. What the flag does is not stated, and neither is the behaviour when the collection changes. A user who installed 142 skills has no documented way to learn that fifteen more now exist, and no documented way to reconcile the install against the advertised count without reading the repository.

A profit calculator in markdown, with no version and no release

Four of the skills in the pricing group are calculators, one each for Amazon, Shopify, TikTok Shop and Walmart, and their descriptions promise real numbers rather than guidance. The Amazon one is billed as a complete breakdown of the fees the platform charges for storage and fulfilment, plus referral fees, break-even analysis and a pricing recommendation. The Shopify one covers ad spend, customer acquisition cost, payment processing, third-party logistics and a ratio between lifetime value and acquisition cost. Those fee schedules are not constants. They are published tables that the platforms revise on their own schedule, and a calculator that hard-codes them inside a markdown file is only as current as the date somebody last checked them. This repository offers no way to date that. There are no releases, so no calculator is tied to a version, and the branch was last pushed on 2026-08-26 with no changelog to say what was revised. The pricing group does contain two skills that acknowledge the problem in a different register: one is described as guardrailed, with floors and ceilings, rule matrices, simulations, approvals and rollback, and another as evidence-bounded, with unit economics, candidate scenarios, experiment design and rollout gates. Those two are the right shape for this problem. The four calculators are the wrong shape. The margin claim is the one that carries money and the one with the most moving parts. A break-even figure needs a referral percentage, a fulfilment or logistics charge, a storage charge that varies by season and by size band, and whatever the payment processor takes, and none of those are published as stable constants. So each calculator is a snapshot of a fee schedule the document never dates, and the absence of releases compounds that rather than merely adding to it. With no tag there is no way to say which fee table a given version used, so an archived copy of one of these files carries no information about what it actually computed.

Three of the skills exist to identify fake reviews and buyer manipulation

The competitor-analysis group opens with three review checkers, one each for Amazon, eBay and Walmart, and all three are described in the same terms: authenticity, fake reviews and fake feedback, time clustering, verified purchase validation, suspicious patterns, buyer manipulation, red flags. On eBay the description names the platform's own badge as a signal to analyse, and on Walmart it names the fulfilment badge. This is the part of the collection with the clearest scope question, and it is worth being plain about it. Review authenticity analysis applied to your own listing is a quality-control tool, and that is a legitimate and common use. The same skill pointed at a competitor's reviews is an accusation, and the output of a language model asked to find manipulation will find manipulation, because a model asked to identify fake reviews will produce findings whether or not any exist. Nothing in the collection restricts the target, and nothing in it tells the agent whose reviews to analyse. The rest of the cluster is the mirror image: brand protection skills for five platforms covering hijackers, counterfeits, minimum advertised price violations, unauthorised sellers, counterfeit stores, takedowns and domain monitoring, with corresponding monitoring directories in the tree. It is worth noticing that this is the one place in the collection where the subject is other people's behaviour rather than the seller's own business, and the descriptions say so plainly by naming manipulation, suspicious clustering and fake feedback. That is information a legitimate seller needs, because a marketplace where reviews are bought is a marketplace where a seller's own ratings are not worth what they appear to be. The tooling is defensible on that reading. The gap is that the collection ships no target restriction, no evidence standard, and no notion of what a finding has to be backed by before anyone acts on it.

The link that defines the format points at the vendor's own site

The compatibility line in the header names six tools and then adds a seventh category, any agent compatible with the Skills format. That last phrase is a link, and the link does not go to a specification, a schema, an archive or a third party. It goes to the vendor's own hosted skill hub, on the vendor's own domain, with a referrer tag on the query string. So the definition of the format this collection claims to implement resolves to the page selling the collection. The five named tools are third-party projects and are linked plainly by name, which is the honest way to do it. The format link is doing something else. Two consequences follow. A reader who wants to know what a skill file is allowed to contain is sent to a product page rather than to a format description, so the contract is only as stable as the vendor's marketing. And an installer that resolves format compatibility by consulting that page is trusting a page the seller controls, which is a weaker guarantee than resolving it against a published schema. The compatibility claim beside it follows the same pattern. The header says the skills work with a list of named coding agents plus any compatible one, which is a matrix nobody can check from the document, because no version of any of those tools is named and no behavioural difference between them is described. A skill file that is pure markdown should behave identically everywhere. If that is the design, the compatibility list is decorative. If it is not the design, something in the format is tool-specific and a reader has no way to find out what.

Six highlighted skills route to tracked URLs and the tables route to local paths

Two destinations for the same skills inside one document. The most-popular section links all six of its entries to hosted pages on the vendor's skill hub, and every one of those links carries a referrer parameter identifying the traffic as coming from the repository. The collection tables below link to relative paths inside the repository instead, one row per skill, with no parameters and no host. So a reader who wants the file gets the file, and a reader who follows the highlights lands on a product page with the affiliate tag already attached. Nothing is hidden, and the six highlighted names all appear again in the tables, which is the redeeming detail. But the choice of which six to highlight and how to link them is a marketing decision made inside a document that otherwise reads as a catalogue. The pattern also explains why the vendor wants both: the tables serve the agent and the repository, and the highlights serve the site. The tracked links also encode which skills the vendor treats as its front door. The six highlighted entries are the growth strategy skill, cross-border expansion, a marketing strategy builder, a profit calculator, a pay-per-click planner and a product description generator. That is a deliberate ordering: diagnose the business, expand it, market it, price it, advertise it, write the copy. It is a sensible funnel, and it is also a reading of the collection's priorities that the tables underneath do not convey.

Single skills and skill families sit at different depths in the tree

The top-level listing is a wall of directories, one per skill or per family of them, and the naming is not consistent between the two cases. A standalone skill gets a directory named after itself, so cross-border-ecommerce, dynamic-pricing-ecommerce and competitor-price-analysis are each a single directory. A family gets a directory named after the family with a subdirectory per platform inside it, so the brand protection group is one directory containing five, and the tables link into the second level. That gives the collection two path depths for the same kind of artefact, which is awkward for anything that wants to enumerate the collection: a flat scan finds thirty-odd skills and misses a hundred that are nested. The same split appears in the link syntax in the tables, one level for the singletons and two for the families. Alongside the directories the root holds very little: the readme, the licence, a gitignore and one image file. There is no manifest, no plugin descriptor and no package file, so whatever tool performs the install has to discover the collection by walking the tree rather than by reading a declaration. The listing also mixes three kinds of thing under those names. Some are single skills, some are families, and some are named after an activity rather than an artefact: brand monitoring, domain monitoring, competitor price tracking, customer feedback analysis, returns management, and a directory for dropshipping product research, which is a category rather than a skill. A monitoring directory implies something that runs and reports, and a skill file cannot run. So either those directories hold scripts, which would contradict the plain-markdown description, or they hold instructions for an agent to perform a monitoring task on request, which is a different product from a calculator. The visible material does not settle it.

A markdown-only collection whose detected primary language is Python

The repository metadata reports Python as the primary language, and the document describes the opposite. The what-are-these section is explicit that e-commerce skills are plain-text instruction files, that there are no binaries, no API keys and no setup friction, and that they are just markdown any model can read. So either the tree contains Python that the description does not mention, or the language detection is picking something up that is not part of the deliverable. The top-level listing is cut off partway through the e-named directories, so this cannot be settled from what is here, and it is exactly the sort of thing worth checking before you rely on the description. The status column makes a related point. Every skill in the visible tables is marked available, with the key defining that as production-ready and a second state reserved for beta items described as functional but still being improved. Calling a markdown instruction file production-ready is a strong claim about a document that nobody has executed, and no visible row uses the weaker label.

A second repository holds 51 more of the same skills

Partway down the document, after the most-popular list and before the collection tables, there is a cross-promotion for a sibling repository: fifty-one skills for a single marketplace, covering its fulfilment programme, pay-per-click work, listing optimisation and keyword research. So the ecosystem is one hundred and fifty-seven skills in this repository plus fifty-one in the other, two hundred and eight in total under one vendor, and the split is not arbitrary. This repository is the cross-platform half and the other one is the Amazon half. That is a defensible organisation, and it is also the reason the header's platform list reads as broad while the deepest per-platform coverage sits elsewhere. A seller who installs everything here and then needs marketplace-specific depth for one platform is being walked to a second repository partway through the first one's front page, which is a reasonable funnel and worth naming as one. The directory names show where the depth actually sits. Etsy alone contributes eighteen visible directories, running from offsite advertising and multi-shop through print-on-demand, review strategy, seasonal strategy and shop analytics. That is the deepest single-platform coverage in this repository, and it is Etsy rather than the platform the sibling repository is dedicated to, which is a small irony worth noticing before concluding the split was made by importance.

Editorial conclusion

This is a prompt library rather than software, and that framing is the right one for reading it, because a markdown instruction file cannot be versioned by anything except a commit and cannot be tested except by reading it. Four things to check before you trust any of it. The headline count and the install command disagree, so know which number you actually installed. There are no releases, and the profit calculators hard-code platform fee schedules, which change; find out when the tables were last checked before you act on a margin number. Three of the skills exist to identify fake reviews and buyer manipulation, which is a defensive tool on your own listings and a different act on someone else's, so decide which use you are installing for. And the link that defines the format points at the vendor's own hosted site, as do all six highlighted skills, each carrying a referrer tag, while the collection tables link to local paths instead. For a seller who wants a checklist to argue with, the collection is a fast way to find the questions they have not asked. For anyone whose pricing or margin decision depends on one of these documents, treat it as a prompt to think with, not as a calculation.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. nexscope-ai/eCommerce-Skills on GitHub
4. README