Agentique← Landing pageECORN field manual · 2026 edition
10 repeatable workflows for themes, catalog, CRO, reviews, SEO, inventory, promotions and reporting.
Operating contract
A workflow is not finished until a human can inspect the evidence and approve or reject the next action.
Read this first
This playbook turns repetitive ecommerce work into reviewable outputs: an audit, a prioritized backlog, a code or content diff, a QA report, an operator brief or an explicit decision.
Part I · 60-minute setup
Use the screen brief under each step to create clean, repeatable screenshots without exposing real stores, credentials or customer data.
Use Claude Code’s current native installer, sign in, then confirm the installation before opening a store workspace.
# macOS, Linux or WSL
curl -fsSL https://claude.ai/install.sh | bash
claude --version
claude doctor
claudeConfirm Node.js and Git, install the current Shopify CLI, and work only with a legitimate staff, collaborator or Theme Access account.
node --version
git --version
npm install -g @shopify/cli@latest
shopify versionSeparate always-on instructions, business context, approved inputs, workflow specifications and outputs so Claude reads only what the task needs.
shopify-store-brain/
├── CLAUDE.md
├── .claude/settings.json
├── context/{brand,catalog,customers,cro,operations,metrics,safety}.md
├── theme/
├── data/approved-inputs/
├── workflows/
└── outputs/decision-log.mdPull the theme locally, establish a Theme Check baseline and use a development theme for previews. Do not start with a live publish action.
cd theme
shopify theme pull --store your-store.myshopify.com
shopify theme info
shopify theme check
shopify theme dev --store your-store.myshopify.comUse Shopify’s supported Claude plugin or developer MCP so Claude can consult current documentation and validate Shopify code inside the workflow.
claude plugin install shopify-ai-toolkit@claude-plugins-official
# Alternative developer MCP
claude mcp add --transport stdio shopify-dev-mcp -- npx -y @shopify/dev-mcp@latestPut behavioral rules in CLAUDE.md and technical boundaries in .claude/settings.json. Deny secrets and production publishing; ask before edits.
{
"permissions": {
"allow": ["Read", "Bash(git diff *)", "Bash(shopify theme check *)"],
"ask": ["Edit", "Bash(shopify theme push *)"],
"deny": ["Read(**/.env)", "Bash(shopify theme publish *)"]
}
}Store Brain minimum
Positioning, voice, approved proof and prohibited claims
Products, variants, collections, metafields and merchandising rules
Segments, triggers, objections and anonymized language
Page standards, evidence rules, tests and prior decisions
Inventory, fulfillment, subscriptions, calendar and owners
Definitions, sources, periods, time zones and caveats
Allowed data, sensitive fields, approval actions and rollback owner
Part II · The workflows
Review only by default
Find fragile Liquid, duplicated logic, performance risks and accessibility gaps before they become release problems.
When to use
Use before a theme refactor, after several app installs, during a replatform, or when storefront performance and maintainability are drifting.
Required inputs
Claude Code workflow
Read CLAUDE.md, context/brand.md and context/safety.md before doing anything.
Audit the Shopify theme in ./theme. Do not edit files yet.
1. Map templates, sections, snippets and assets.
2. Run the repository's existing checks, including Shopify Theme Check.
3. Find Liquid errors, deprecated patterns, duplicated logic, render-blocking assets, app residue, accessibility issues and fragile schema.
4. For every finding include file and line, evidence, customer or operator impact, confidence and recommended fix.
5. Rank findings by impact, effort and release risk.
6. End with the smallest safe first branch.
Separate confirmed findings from hypotheses. Do not run a deploy or production theme command.Example · Women’s shapewear brand
The store has accumulated sizing widgets, bundle scripts and two old review-app snippets across three redesigns. Claude maps the render chain and finds duplicate variant listeners, an orphaned stylesheet and a size-guide modal without keyboard focus management. The operator approves one cleanup branch instead of a broad rewrite.
Approval gate
A human reviews the evidence and exact proposed change. No live write, publish, send or mutation occurs from this workflow.
Expected result
Advanced ideas
Review only by default
Turn release checks into a repeatable pass/fail report across PDP, collections, cart, mobile behavior and tracking.
When to use
Use before publishing a theme, launching a new template, shipping a subscription change or releasing a campaign landing page.
Required inputs
Claude Code workflow
Act as the release QA lead for this Shopify storefront. Read the Store Brain and safety rules first.
Scope: [DESCRIBE RELEASE]
Preview: [URL OR LOCAL COMMAND]
Changed files: [BRANCH OR DIFF]
Test homepage to collection to PDP to cart, search, navigation, variant selection, selling plans, discounts, error states and analytics events. Check mobile first, then desktop.
Return a table with: severity, journey, device, steps, expected result, actual result, evidence, likely owner and release recommendation. Mark each item BLOCK, FIX SOON or PASS.
Do not publish the theme. Do not claim a pass for anything you could not test. End with a go/no-go decision and the exact unresolved blockers.Example · DTC running shoe brand
A new fit-finder block is being added before a product launch. The workflow catches that the result does not update when color changes, the sticky add-to-cart covers the final answer on a small iPhone viewport and the selected fit value is missing from analytics. All three are resolved in the preview theme before launch.
Approval gate
A human reviews the evidence and exact proposed change. No live write, publish, send or mutation occurs from this workflow.
Expected result
Advanced ideas
Review only by default
Find missing metafields, inconsistent options, weak titles, media gaps and publishing mistakes without bulk-editing blindly.
When to use
Use after a migration, before a collection launch, when filters are unreliable, or when hundreds of products have inconsistent data.
Required inputs
Claude Code workflow
Read context/catalog.md, context/operations.md and context/safety.md. Analyze [PRODUCT EXPORT PATH]. Do not call Shopify write APIs.
Validate titles, handles, status, vendor, product type, options, variants, prices, compare-at prices, SKUs, barcodes, media, SEO fields, collections and required metafields. Apply the documented catalog rules only.
Output:
1. Dataset profile and quality score by field.
2. Exceptions grouped by root cause.
3. A prioritized correction plan.
4. A proposed CSV containing original value, proposed value, rule used and confidence.
5. A 20-row human review sample.
Never invent product attributes or claims. Put uncertain corrections in NEEDS OWNER INPUT.Example · Multi-location home goods brand
A 1,900-product catalog uses ‘Colour,’ ‘Color’ and free-text finish values, which breaks filters and marketplace feeds. Claude groups the problem into five rule conflicts, proposes normalized values from the approved taxonomy and isolates 74 products that require a merchandiser rather than guessing.
Approval gate
A human reviews the evidence and exact proposed change. No live write, publish, send or mutation occurs from this workflow.
Expected result
Advanced ideas
Review only by default
Combine product facts, customer objections and page evidence into a ranked test backlog instead of generic CRO advice.
When to use
Use when a high-traffic PDP underperforms, after a traffic-source shift, or before a redesign and test-planning cycle.
Required inputs
Claude Code workflow
Act as a senior Shopify CRO operator. Read the relevant Store Brain files before the audit.
Inputs: [PDP], [REVIEWS], [SUPPORT THEMES], [FUNNEL METRICS], [TRAFFIC CONTEXT].
Produce:
- the top conversion frictions with page evidence;
- the customer objection behind each friction;
- the exact copy, proof, layout, image, option, policy or offer change;
- impact, effort and confidence scores;
- direct fix versus A/B test recommendation;
- missing evidence and instrumentation;
- a one-week implementation backlog.
Prioritize mobile. Use only approved product claims. Do not estimate uplift. Do not edit the theme until a test or change is approved.Example · Clinical skincare brand
Paid social sends cold traffic to a treatment PDP with strong clinical proof but confusing usage instructions. Review mining shows irritation anxiety and uncertainty about layering. The backlog prioritizes a tolerance protocol, routine compatibility block and proof hierarchy before cosmetic layout changes.
Approval gate
A human reviews the evidence and exact proposed change. No live write, publish, send or mutation occurs from this workflow.
Expected result
Advanced ideas
Review only by default
Turn customer language into evidence-backed FAQs, proof blocks and message tests while preserving quote traceability.
When to use
Use when refreshing PDP copy, launching ads, improving FAQs or investigating why customers hesitate, return or contact support.
Required inputs
Claude Code workflow
Analyze [REVIEW EXPORT] for [PRODUCT OR CATEGORY]. Read brand, customer and safety context first.
Create a customer-language map covering jobs-to-be-done, desired outcomes, purchase triggers, objections, comparison language, usage confusion and disappointment causes. For each theme show frequency, representative source IDs, confidence and the page element it should influence.
Then draft: five FAQ entries, three proof-block concepts, three headline directions and two comparison-section structures. Preserve the customer's meaning without fabricating quotes. Mark medical, performance or regulated claims for verification.
Do not expose names, emails, order numbers or other personal data.Example · DTC pet supplement brand
Reviews praise convenience but repeatedly mention uncertainty about dosage by weight and how long results take. Claude proposes a weight-based usage module, a transparent timeline FAQ and new headline directions. The compliance owner reviews every health-related phrase before anything reaches the PDP.
Approval gate
A human reviews the evidence and exact proposed change. No live write, publish, send or mutation occurs from this workflow.
Expected result
Advanced ideas
Review only by default
Improve metadata, structured product information, internal links and machine-readable catalog clarity without keyword stuffing.
When to use
Use during a collection restructure, after a theme migration, when organic landing pages decay, or when product data is hard for search and shopping agents to interpret.
Required inputs
Claude Code workflow
Audit [STORE OR URL SET] for technical SEO and machine-readable commerce discovery. Read catalog and brand context first.
Check indexation signals, canonicals, pagination, redirects, titles, descriptions, headings, internal links, product and offer structured data, variant information, availability, policy references and content duplication.
For every issue provide affected template or URL pattern, evidence, likely discovery impact, fix, validation method and owner. Separate confirmed defects from strategic opportunities.
Do not recommend mass-generated doorway copy. Do not invent search volume or ranking impact. End with the ten highest-leverage fixes and the data needed to validate them.Example · Specialty coffee subscription brand
One-time bags and subscriptions share inconsistent titles and schema, while roast-level collections have thin internal linking. The audit separates a theme-level offer-schema defect from a content architecture opportunity and proposes a crawl-based validation plan.
Approval gate
A human reviews the evidence and exact proposed change. No live write, publish, send or mutation occurs from this workflow.
Expected result
Advanced ideas
Review only by default
Find stockout risk, slow movers, broken collection logic and paid traffic sent to unavailable products.
When to use
Use weekly, before a campaign or purchasing decision, and whenever stock position changes faster than merchandising can react.
Required inputs
Claude Code workflow
Read context/catalog.md, context/operations.md and context/metrics.md. Analyze [INVENTORY], [SALES VELOCITY], [COLLECTIONS] and [CAMPAIGN CONTEXT].
Flag projected stockouts, unavailable hero products, slow movers, stranded variants, location imbalances, collections dominated by unavailable items and campaigns pointing to constrained inventory. Show the formula and source timestamp for every alert.
Return: alert, SKU/product, evidence, commercial context, urgency, recommended action, owner and confidence. Separate data-quality failures from real inventory risk.
Do not adjust inventory, reorder products, pause campaigns or change collections.Example · Outdoor equipment retailer
A paid campaign still sends traffic to a bestselling tent whose main variant has five days of cover, while a similar model is overstocked at another location. The brief gives growth and merchandising a joint decision instead of letting Claude move stock or alter ads itself.
Approval gate
A human reviews the evidence and exact proposed change. No live write, publish, send or mutation occurs from this workflow.
Expected result
Advanced ideas
Review only by default
Check offer logic, dates, exclusions, inventory, creative, landing pages and tracking before customers find the conflict.
When to use
Use before seasonal promotions, product drops, bundles, market-specific offers and coordinated email, SMS or paid-media launches.
Required inputs
Claude Code workflow
Act as launch QA lead for [PROMOTION]. Read commercial, operations and safety context. Treat the approved offer brief as the source of truth.
Validate dates and time zones, markets, currencies, customer eligibility, product/collection inclusion, exclusions, stacking, minimums, subscription behavior, inventory, landing-page copy, email/SMS/ads, cart messaging, terms and analytics events.
Create a pass/fail matrix with evidence and exact remediation. Test normal, boundary and ineligible cases. Mark BLOCKERS separately from follow-ups.
Do not create, activate or edit discounts. End with named approvals required before launch.Example · Functional beverage brand
A bundle launch promises 20% off across paid ads, email and the landing page. QA finds that subscriptions are unintentionally eligible, one flavor is excluded by a stale collection rule and the countdown uses the wrong time zone. The launch pauses until the merchant fixes and re-tests the configuration.
Approval gate
A human reviews the evidence and exact proposed change. No live write, publish, send or mutation occurs from this workflow.
Expected result
Advanced ideas
Review only by default
Turn store and marketing data into exceptions, evidence, decisions, owners and next actions—not a dashboard recap.
When to use
Use for weekly leadership reviews, cross-functional operating meetings and agency client reporting.
Required inputs
Claude Code workflow
Create the weekly Shopify operator report for [DATE RANGE]. Read context/metrics.md and the latest decision log first.
Use only the supplied data. Cite the source and period for every material claim. Cover revenue, orders, conversion, AOV, acquisition, retention, refunds, inventory, promotions and customer voice where available.
Structure the report as:
1. What changed.
2. Material exceptions and evidence.
3. Likely drivers, clearly labeled as confirmed or inferred.
4. Risks and opportunities.
5. Decisions required with owner and due date.
6. Actions from last week and their status.
7. Data gaps.
Do not fill space with generic recommendations or claim causality without evidence.Example · Subscription supplement brand
Revenue is flat, but the report shows new-customer conversion down while subscription renewal revenue offsets the decline. The likely driver section connects the drop to an out-of-stock entry SKU and a landing-page change, then asks growth and merchandising for two explicit decisions.
Approval gate
A human reviews the evidence and exact proposed change. No live write, publish, send or mutation occurs from this workflow.
Expected result
Advanced ideas
Review only by default
Keep every store’s context, credentials, rules, outputs and decision history isolated while standardizing delivery.
When to use
Use when an agency serves multiple Shopify clients, a group operates several brands, or internal ownership changes frequently.
Required inputs
Claude Code workflow
Prepare a governed handoff for [STORE / CLIENT]. Read only this store's Store Brain and access matrix.
Audit the workspace for context completeness, environment separation, credentials handling, workflow versions, open decisions, approved outputs and unresolved risks. Confirm that shared templates contain no client data.
Return:
- workspace and access checklist;
- workflows installed and last validated;
- current owners and approval gates;
- outstanding decisions and risks;
- deliverables with source paths;
- exact next-run instructions;
- items the receiving team must accept.
Stop and report a boundary violation if another store's data or credentials are visible.Example · Shopify agency managing eight brands
The agency keeps shared workflow specifications in one template repository and a separate private Store Brain for each client. Before a strategist changes accounts, Claude produces an acceptance checklist, identifies two stale data exports and catches a path that accidentally points to another client’s asset folder.
Approval gate
A human reviews the evidence and exact proposed change. No live write, publish, send or mutation occurs from this workflow.
Expected result
Advanced ideas
Part III · Results
Business impact often lags. First prove that the workflow is faster, more complete, accurate, safe and useful to the human making the decision.
Minutes from approved inputs to a reviewable output
Recommendations accepted, edited or rejected by a human
Confirmed issues versus false positives and unsupported claims
Unapproved writes, exposed secrets and skipped approval gates
Defects found before release and decisions closed on time
The store metric connected to the specific workflow
Decision log · one row per recommendation
| Date | Workflow | Recommendation | Evidence | Decision | Owner | Measure date | Outcome | Store Brain learning |
|---|---|---|---|---|---|---|---|---|
| YYYY-MM-DD | 04 · PDP CRO | Add usage protocol | Review theme R-07 | Approved | Ecommerce lead | +14 days | Pending | customers.md |
Part IV · Advanced roadmap
Level 1
Run one workflow manually with approved inputs until the output is easy to review.
Level 2
Move the proven procedure into .claude/skills with an explicit invocation and output schema.
Level 3
Add validated JSON or CSV beside the human report without removing caveats or confidence.
Level 4
Schedule stable audits and reports. Alert a human; do not take action automatically.
Level 5
Require an allowlist, dry run, diff, approver, rollback, audit log and post-action verification.
Level 6
Feed accepted, rejected and measured decisions back into the Store Brain.
Part V · 30-day rollout
Week 1
Install the stack, build Store Brain, configure permissions and choose one workflow.
Week 2
Use a narrow scope, review false positives, improve the schema and approve one low-risk recommendation.
Week 3
Re-run it, compare time and accepted findings, then record the decision and outcome.
Week 4
Turn the stable procedure into a skill and document owner, cadence and approval gate.