agentleFS
ReplAi-NowReplAi-Now/shopify-skills/skills/store/conversion-optimiser/SKILL.md

conversion-optimiser

Skill1 starsChanged 20 days agoMITView on GitHub

Audit a Shopify store for conversion problems, then fix them on a theme copy with your OK. Use for "audit my store", "why is my conversion low" or "visits but no orders".

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 6 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Look for saved setup: replai-setup/conversion-optimiser.json in the working folder, or a pasted setup card (a JSON block starting with …

    line 20
  2. Distinctive:

    If not found, send ONE message with these numbered questions.

    line 21
  3. Distinctive:

    Speed: call PageSpeed Insights for the home page and one product page, mobile and desktop (no key needed for light use).

    line 43
  4. Very specific:

    If something cannot be checked, mark it unknown with the reason.

    line 52
  5. Distinctive:

    Ask: "Do you want me to start on the Do first list?" Phase 2 starts only on a yes and only within what setup allows (question 6).

    line 58
  6. Very specific:

    Check the copy at 390 px and desktop, then send the preview link https://<store>.myshopify.com/?preview_theme_id=<id> with a list of what …

    line 68
  7. Very specific:

    After the fixes: rerun PageSpeed and the failed checks, update findings.json and regenerate the report so the merchant sees before and …

    line 79
  8. Very specific:

    Before every write, read and log the current value.

    line 93
  9. Distinctive:

    Never touch payments, payouts, staff accounts, apps' billing or credentials.

    line 95
  10. Distinctive:

    Never invent facts in store content (delivery times, materials, guarantees, review counts) and never add fake urgency.

    line 96
  11. Distinctive:

    Only fetch the merchant's own public storefront, Google PageSpeed and the tools they connected.

    line 97

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Conversion optimiser

by ReplAi (replai.now). Free e-commerce skills pack.

When to use. Audit a Shopify store for conversion problems and then fix them safely with the merchant's OK. Phase 1 (read only) checks speed with Google PageSpeed, the mobile layout at 390 px, product pages (photos, price, size chart, delivery time and cash on delivery info near Add to cart, returns, reviews, sticky Add to cart, honest urgency), trust (contact, WhatsApp, policies, payment icons), navigation and search, cart and checkout friction, Arabic/RTL quality and SEO basics, then gives a prioritised fix list (impact x effort) as an HTML report. Phase 2 makes the fixes on a duplicate theme with a preview link, and content edits through the Shopify connector in confirmed batches with a rollback log. Use when the merchant says "why is my conversion low", "audit my store", "my store gets visits but no orders", "CRO", "speed up my store", "improve my product page", "راجع متجري", or "fix my Shopify theme".

Two phases. Phase 1 audit is read only and ends with a ranked fix list. Phase 2 act changes the store only on a duplicate theme or in small confirmed content batches. Never edit the live theme.

Setup

  1. Look for saved setup: replai-setup/conversion-optimiser.json in the working folder, or a pasted setup card (a JSON block starting with "skill": "conversion-optimiser"). If found, say "Using your saved audit setup for <store>." and continue. "Change my setup" or "redo setup" reruns the questions with current answers pre-filled.
  2. If not found, send ONE message with these numbered questions. The merchant can reply "defaults" or answer some numbers only.
1. Store URL? [read from the Shopify connector if connected]
2. How can I make theme changes? [connector if connected, otherwise manual steps for you] (Shopify CLI / connector / manual)
3. Store languages? [en] (en / ar / both, and which is the default)
4. Main market and currency? [Egypt, EGP] (EG, SA, AE, KW, QA, BH, OM)
5. Anything to focus on first? [everything] (speed / product page / checkout / Arabic / SEO)
6. What may I change in Phase 2? [product descriptions, alt text, SEO titles] (add: theme copy, policies, pages, navigation; or "audit only")
7. Brand voice for any text I write? [friendly, short sentences, no hype]
8. Arabic style, if Arabic? [simple Egyptian Arabic] (Gulf Arabic / Modern Standard Arabic)
9. Report format? [dashboard: one HTML page] (dashboard / spreadsheet / report)
  1. Save the answers in the shape of assets/setup.example.json: write replai-setup/conversion-optimiser.json when there is a file system, otherwise print the setup card as a fenced JSON block and tell the merchant to paste it at the start of the next chat.
  2. Smoke test: fetch the store home page and report its title in one line (no fetch tool: say the audit will use screenshots and exports instead). If a connector is connected, also read the shop name and the live theme name (themes(first: 20, roles: [MAIN])). If the store is password protected, say the audit needs the password page off or screenshots.

Per run (not at setup): which pages to treat as top products (default: best sellers from the connector, else the first 3 on /collections/all?sort_by=best-selling), and any page the merchant is worried about.

Phase 1: audit (read only)

  1. Collect pages: home, one top collection, 3 top product pages, cart. Use the agent's browser or fetch tool. With neither (chat apps with uploads only), ask for the phone screenshots and exports listed in references/data-sources.md, which also has the export fallbacks.
  2. Speed: call PageSpeed Insights for the home page and one product page, mobile and desktop (no key needed for light use). Save each response and run: python3 scripts/psi_summary.py psi-home-mobile.json psi-home-desktop.json psi-product-mobile.json psi-product-desktop.json (home first: S1-S5 use the first mobile file) It returns scores, Core Web Vitals and ready-made S1-S5 checks. Request format, limits and fallbacks: references/pagespeed.md.
  3. Walk the checklist in references/checklist.md (50 checks: speed, mobile at 390 px, product page, trust, navigation and search, cart and checkout, Arabic/RTL, SEO). For each check, look at the real page:
    • Mobile: open pages at 390 px wide; measure document.documentElement.scrollWidth; look for pop-ups, button size, Add to cart position, sticky bar.
    • Product: count photos (/products/<handle>.js), find delivery time, COD, returns line, size chart, reviews near the button.
    • Checkout: add one item, open cart and the first checkout step, read what is required, leave. Never type personal data, never submit anything.
    • Arabic: switch language, check direction, leftovers in English, font.
    • SEO: read <title>, meta description, alt attributes, H1 count; fetch internal links on audited pages and note 404s. If something cannot be checked, mark it unknown with the reason. Never mark pass without looking.
  4. Write findings.json (shape in assets/findings.example.json). Every fail or partial needs evidence.url, evidence.seen (what exactly you saw, with numbers), impact, effort, fix (the exact change: which setting, which text, which block) and change_type (theme, content, setting, app, manual). Paste the psi_summary output under "speed" and its suggested_checks into checks.
  5. Prioritise and report: python3 scripts/prioritise.py findings.json --html conversion-report.html --csv fixes.csv --lang en Priority = impact x (4 - effort), from 9 (high impact, low effort) down to 1. Buckets: Do first (6-9), Next (3-4), Later (1-2). The health score is the weighted share of checks passed. If the script lists errors, fix findings.json and rerun. Without code, follow references/method.md.
  6. Tell the merchant in their language: the health score in one line, the top 5 fixes (what, evidence, fix), what needs them (photos, settings), and what could not be checked. Then attach or show the report. Use --lang ar for an Arabic report frame.
  7. Ask: "Do you want me to start on the Do first list?" Phase 2 starts only on a yes and only within what setup allows (question 6).

Phase 2: act (with confirmation)

Group the approved fixes by change_type and handle one group at a time.

Theme changes (blocks, settings, snippets, code). Full commands and fallbacks in references/theme-workflow.md.

  1. Record the live theme ID in the change log (it is the rollback).
  2. Duplicate the live theme: shopify theme duplicate --theme <live id> --name "ReplAi CRO <date>", or GraphQL themeDuplicate(id:, name:), or manual: Online Store > Themes > ... > Duplicate.
  3. Make the changes on the copy only: theme editor clicks, or shopify theme push --theme <copy id> --only <file>, or GraphQL themeFilesUpsert (needs a Shopify exemption most connectors lack). Never --live, never --allow-live.
  4. Check the copy at 390 px and desktop, then send the preview link https://<store>.myshopify.com/?preview_theme_id=<id> with a list of what changed.
  5. Publish only after an explicit yes for publishing this copy, after checking the live theme was not edited in the meantime. Tell the merchant how to switch back.

Content changes (descriptions, SEO titles, alt text, policies, pages, navigation) through the connector. Queries, mutations and the log format in references/content-edits.md.

  1. One area per batch, up to 25 changes. Read the before values.
  2. Show the exact list (item, field, before, after) and wait for a yes for that list.
  3. Apply, read back, report errors, and append the batch to replai-cro-change-log.json (or print it in chat apps).
  4. To undo: python3 scripts/rollback.py replai-cro-change-log.json --batch N lists what to restore; that is a new batch with its own yes.

Settings, apps and manual work: give exact admin clicks (for example Settings > Checkout > Company name: Don't include) or the brief (for example "2 on-model photos for the denim jacket"). Do not install apps or change settings yourself unless the merchant asks and confirms.

After the fixes: rerun PageSpeed and the failed checks, update findings.json and regenerate the report so the merchant sees before and after. Suggest comparing conversion rate in Shopify Analytics over the next 2-4 weeks. Never promise a number.

Outputs

  • conversion-report.html (dashboard, default): health score, top fixes, Next and Later lists, speed table with plain explanations, passed and unchecked items. Works offline and at 390 px. Ends with Made with the free ReplAi skills pack.
  • fixes.csv (spreadsheet): ranked fixes with evidence and fix text.
  • report: a short message leading with the 3-5 decisions, built from the script's JSON.
  • findings.json and replai-cro-change-log.json: the audit record and the rollback log.

Safety

  • Phase 1 is read only. No store writes, no form submissions, no personal data typed anywhere (checkout is looked at, not filled in).
  • The live theme is never edited. Theme work happens on a duplicate, is previewed, and is published only on the merchant's explicit yes for that publish. The previous theme stays in the library for rollback. Never delete a theme.
  • Content changes go in small batches; each batch needs its own explicit yes for the exact list shown. Approval never carries to the next batch.
  • Before every write, read and log the current value. Never delete products, pages, policies or menu items; menuUpdate replaces the whole menu, so always send every existing item back.
  • Change only the areas allowed in setup question 6. Ask before touching anything else.
  • Never touch payments, payouts, staff accounts, apps' billing or credentials. Do not install or uninstall apps without the merchant's yes.
  • Never invent facts in store content (delivery times, materials, guarantees, review counts) and never add fake urgency. Estimates are labelled with their assumption.
  • Only fetch the merchant's own public storefront, Google PageSpeed and the tools they connected. No personal data in URLs.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.