Store Fix
Finds what is losing sales on your store, ranks it by a formula you can check, writes the fixes in your voice, and applies the copy ones after one tap.
Who it is for
The owner who suspects the product page is leaking sales but does not know where to start. It is the agent every owner runs first, so its first output decides whether they trust the other fifteen.
What it reads from the brain
Before every run. Sections are addressed as layer.section.
What it does, step by step
- Observe
Fetches your home page, top three hero product pages, top collection, cart page and policy pages. Captures headings, above the fold order, price visibility, trust elements, review presence, FAQ presence, schema types, image counts, and the mobile order at 390px.
- Score
Every checkpoint whose surface was observed gets 0 (missing or broken), 1 (present but weak) or 2 (meets the pass standard), with a URL and the observed text as evidence. Anything not observable from outside is marked not observable and excluded, never guessed.
- Rank
priority = (2 minus score) times impact weight times confidence, divided by effort. Confidence is 1.0 for direct observation and 0.7 for inferred. Sorted descending.
- Weighted score
Sum of score over 2 times impact weight, divided by total impact weight, times 100, over observed checkpoints only. Reported with the count observed.
- Queue
Top ten by priority. Each: the finding in plain words, the evidence, the fix, and whether it is a copy fix the agent can apply or a structural fix for a developer.
- Draft fixes
Replacement copy for each copy fix in the top ten, in your voice, using customer.vocabulary and only claims_allowed. FAQ questions are your customers' objections, verbatim.
- Act
Product descriptions and SEO fields go live the moment they are saved, so the approval comes before the write. One approval for the batch: Apply n content changes to k live products? Here are three of them. On approve it updates content, sets FAQ metafields, creates draft pages, and records every previous value for rollback. Never more than one live batch per day per store.
Tools it uses
storefront.fetchPlatform, product count, price range, collections, apps, theme, schema typesstorefront.pageRenders a page at desktop and 390px mobile and captures content ordershopify.products.listAct: the live products to updateshopify.product.update_contentAct: description, SEO title and description. Returns previous valuesshopify.metafields.setAct: FAQ blocks on product pagesshopify.page.create_draftAct: new FAQ or guide pages as drafts. Publishing is a separate approvalshopify.theme.duplicate, update_asset, publishHeld for when Shopify grants the theme exemption. Until then theme changes are developer recommendations with the exact file
The three modes for this agent
Steps 1 to 6. You get the score, the ranked queue and the drafted fixes as copy you paste into Shopify yourself. Store URL only, nothing connected.
Steps 1 to 7. Drafts are created immediately. Live content changes pause for one approval showing three before and after samples, then apply with previous values recorded.
Weekly re audit every Sunday at 22:00 your local time. A fresh score, queue and drafted fixes wait for you Monday. One approval for the live batch. That gate never relaxes.
Guardrails and approvals
Enforced in code. Items with a filled circle wait for your tap in Telegram.
- Content writes only: product descriptions, SEO title and description, FAQ metafields, draft pages.
- Never theme code (until the Shopify exemption exists), never price, inventory, checkout, shipping, taxes, discounts, or customer data.
- One approval per live batch. One live batch per day per store.
- Every change records its previous value, so rollback is re applying it. The summary tells you that.
- A 0 or 2 with no evidence is re scored as 1 at 0.7 confidence and marked.
- Drafted copy carries no claim outside claims_allowed and nothing in your category's claim restrictions. Violations are removed and listed as needs your confirmation.
- If the weighted score is above 85, the summary says the store is well run and names the one surface to look at next, rather than manufacturing ten problems.
- Read only at level 1.
What it writes back to the brain
Dated and sourced. Learnings and performance sections are append only.
Example output
Example output. Structure only. Numbers below are illustrative placeholders for the shape of the report, not results from any store.
Store Fix, your store, first audit
Weighted score: NN of 100 over NNN observed checkpoints (NNN not observable from outside and excluded).
Top five by priority
- Home: no price visible above the fold on mobile. Evidence: your store, home page at 390px. Priority index N.N. Copy fix.
- Hero product 1: no FAQ answering the objection your reviews raise most. Evidence: your store, hero product 1 page. Priority index N.N. Copy fix.
- Hero product 2: description makes a claim not found on your storefront elsewhere. Evidence: your store, hero product 2 page. Priority index N.N. Copy fix, needs your confirmation.
- Collection: no trust element before the grid. Evidence: your store, top collection page. Priority index N.N. Structural, developer.
- Cart: shipping and returns not stated. Evidence: your store, cart page. Priority index N.N. Copy fix.
I can apply 7 of these. They are copy changes to 4 live products. 3 are layout changes I have written up for your developer with the exact file.
Waiting for your approval to publish 7 changes. Every previous value is recorded in changes.json, so any of it can be reverted.
Also delivered: queue.csv (top 40 observed checkpoints: id, surface, score, evidence, priority, fix_type), fixes.md (before and after per fix), changes.json (act mode, every applied change with product id, field, previous value, new value).