15 Claude Skills for Shopify

The problem this pack exists for

Shopify reports forty-one orders for the week. Meta reports thirty-eight purchases. Google reports twenty-two conversions. Nobody in this chain is lying. Shopify counts orders placed in the store. Meta counts purchases it believes it influenced, inside its own attribution window. Google counts conversions it can match to a click or a signed-in user, inside a different window again. On the day someone has to decide whether to raise the budget, the decision depends on which of the three numbers gets trusted, and most stores never wrote down a rule for that choice.

That gap is the smaller problem. The bigger one happened earlier, inside the product data, and it never announced itself as an error. A variant with no barcode does not get rejected with a message; it simply stops being served into Shopping placements. A product title written to read well on a collection page does not match what a shopper actually types into a search box, so it never gets the impression in the first place. An unfilled metafield is not a warning, it is a filter, and the product that needed it fails silently against every rule that checks for it.

None of this shows up as a number going the wrong way in the ads dashboard, because the ads dashboard only reports on inventory that made it into the auction. Half a catalog is not losing bids. Half a catalog never entered a single one.

The question "should we put more budget behind Shopping" does not have an answer inside the ad account. It has an answer inside the Products CSV export, in the metafields that were left blank, in the variants that quietly collapsed into duplicates of each other, and in the collections and tags that decide how spend gets grouped in the first place.

This pack starts there, in the feed, before it ever gets to a bid or a headline. Every skill in it reads something Shopify already exports: the product catalog, the orders export, the Shopify Analytics reports, the Customer Events and Web Pixels API output, the sales channel connections, the discount codes, the checkout and Shop Pay settings. None of it needs write access to the store, and none of it needs to guess.

What is in the pack

Fifteen skills, in four groups. Each one is a markdown file with required inputs, a workflow, decision rules with thresholds, an output format, a worked example and guardrails.

Product data and the feed

• Product Title Rewrite For Feed. Rewrites titles from the Products CSV export against what shoppers actually search for, instead of what reads well on a collection page.
• Variant And Duplicate Audit. Shows which variants enter the feed as separate rows, which are the same product counted more than once, and where identifiers are missing.
• Metafield Gap Check. Names the missing attributes and points to the Shopify metafield meant to carry each one.
• Collection To Campaign Map. Turns collections and tags into a campaign structure and feed labels, so budget can be split by margin rather than by alphabet.
• Merchant Center Disapproval Triage. Takes pasted diagnostic output and maps every disapproval back to the specific Shopify field causing it.

Measurement and signal quality

• Purchase Event Audit. Checks whether the purchase event fires once, at checkout, with a value, and what the Customer Events layer does with it.
• Orders Versus Ad Platform. Reconciles the orders export against the numbers in the ad panel and says which part of the difference is attribution window and which part is a real fault.
• Refund And Subscription Adjustment. Subtracts refunds and cancelled subscriptions before anyone calls a campaign profitable.
• Attribution Window Read. Reads the traffic source report from Shopify Analytics against what each platform claims and names exactly where the two stop agreeing.

Product pages built for paid traffic

• Product Page Paid Traffic Check. What has to be visible to someone who arrived from an ad and has never heard of the brand.
• Ad To Product Page Match. Whether the promise made in the ad is actually on the page the ad sends traffic to.
• Theme Speed And Checkout Friction. Reads a speed measurement and the checkout settings for how much paid traffic drops out before the cart.

The store and the ad accounts

• Sales Channel Connection Check. What the sales channel apps are actually still syncing, and what has quietly stopped.
• Discount And Promotion Sync. Whether the discount showing in the store is the same discount the ad is promising.
• The Weekly Shopify Readout. What to fix, what to measure and where to add budget, written in language a client reads without a glossary.

The three rules that do the most work

Orders and platform conversions disagree by default, and a gap under the threshold is not a fault. Attribution windows overlap and none of the three platforms is counting the same event the same way. The reconciliation only flags a real problem once the gap moves past what the attribution window alone explains, and it says which window it used to make that call.

Refunds and cancelled subscriptions come off before the word profitable gets used. A campaign that looks rentable on gross orders can be underwater once returns and cancellations land, and returns land later than the order that triggered the ad spend. The readout will not report a margin figure that has not had this subtraction applied.

A product missing a required attribute is treated as a product that is not available, not as a product with a minor gap. A blank metafield or a missing identifier does not get partial credit in the feed, and this pack will not pretend it does either; it reports the product as functionally out of the auction until the field is filled.

What Claude Skills for ecommerce refuse to do

Several of the fifteen decline to answer under stated conditions rather than return a number the export cannot carry. The purchase event audit refuses to certify a value-passing event when the Customer Events output does not include a value field, even if the event fires. The orders reconciliation refuses to call a gap a tracking bug before it has isolated how much of it is explained by attribution window overlap alone. The metafield gap check refuses to guess at a missing attribute's value; it names the gap and stops there. The disapproval triage refuses to map a disapproval to a field when the diagnostic text does not name one clearly enough to be sure. This is the part worth keeping even for a store that never installs any of it: a check that always returns something is not a check, and most feed and measurement failures in ecommerce are not wrong numbers, they are real numbers answering a question nobody actually asked.

What changes for a dropshipping store

A dropshipping store does not control the product data at the source. The supplier owns the title, the images, the attributes and the barcode, and the store only holds a copy. That changes what a fix is worth. A rewritten title, a filled metafield or a deduplicated variant lives entirely in the store's own layer, and every one of those fixes gets erased the moment the supplier's catalog resyncs, often without any notice that it happened. The pack's product data skills are written to be rerun on a schedule for exactly this reason, not run once and filed away. The second change is competitive: when several stores resell the same supplier catalog, near-identical listings compete for the same search terms and the same Shopping placements, which moves the real fight from the feed to the product page and the margin, since undercutting on price against a store selling the identical item has an obvious floor.

How the numbers work

Every threshold in the pack lives in one file and is cited by name from the skills, so a number cannot drift between them. The helper scripts read their values out of that same file at run time rather than carrying their own copies, which is the only version of this claim that survives someone changing a number. The pack ships a validator that fails the build if a skill restates a threshold instead of citing it, cites a key that does not exist, or writes a worked example that never declines anything, and a self-test that recomputes the worked example by an independent route and asserts the scripts agree with it.

Each threshold carries a tag. Platform guidance means Shopify, Google or Meta state it themselves, such as a field length limit in Merchant Center or a required field in the Products CSV schema. Heuristic means a practitioner starting point to recalibrate against your own store. Most of this pack is heuristic and says so plainly rather than dressing a guess up as a rule.

How to install

The pack is on GitHub under an MIT licence. Clone it and copy the skill folders into your Claude skills directory, either inside a project or at user level.

git clone https://github.com/mardab96/shopify-claude-skills.git
mkdir -p ~/.claude/skills
cp -r shopify-claude-skills/*-shopify \
      shopify-claude-skills/references \
      shopify-claude-skills/scripts \
      ~/.claude/skills/

The references folder is not optional: it holds every number the skills cite, and without it the thresholds do not resolve. Start a new session in that project and the skills activate when what you are asking about matches one of them. If you use Claude somewhere other than Claude Code, the files are plain markdown and you can paste the relevant one into the conversation with your export underneath.