Claude Code for Ecommerce: The Missing Operations Manual

Claude Code for ecommerce has been a gap, not a category. Free agent repos cluster around engineering; the only adjacent store tooling you find is a couple of schema fragments in a general SEO pack. EcomKit fills the gap with 20 commands, 3 skills, and 2 read-only specialist agents covering the entire operational spine a store actually runs: catalog, PDP, lifecycle email, paid ads, and marketplaces — 16,464 measured tokens. A bootstrap skill writes durable STORE.md, CATALOG.md, and CUSTOMER.md context files that every other command reads, so brand voice, SKU structure, and customer segments stay consistent across PDPs, emails, and ad angles. This is the operations manual.

Why was ecommerce the last gap in Claude Code tooling?

Free Claude Code repos cluster around engineering and general marketing because that is where the contributor base lives. Store operations — taxonomy debt, PDP copy at SKU scale, Klaviyo flow architecture, ad creative derived from reviews, Amazon byte limits — is operationally specific work that generic "write me copy" prompts handle badly.

The result was a near-empty space. AI-driven retail orders grew 15x from January 2025 to January 2026 (Salesforce Commerce Cloud, Q1 2026), and AI referral traffic to ecommerce sites is up 393% year-over-year. Stores are already getting found through AI channels. The tooling to actually run those stores inside Claude Code did not exist until EcomKit.

EcomKit was built to own the whole operational spine rather than a single slice of it. The design principle is that each stage produces structured artifacts the next stage consumes — so a listing audit feeds into listing copy, listing copy feeds into lifecycle flows, and review language feeds back into ad creative. That loop is not possible with disconnected one-off prompts.

What does EcomKit actually include?

EcomKit ships 20 commands, 3 skills, and 2 read-only specialist agents in 16,464 measured tokens (tiktoken-compatible, counted on install via ck tokens ecom).

Commands

The full 20-command set also covers: ecom-audit, ecom-page-audit, ecom-flow-audit, ecom-discount, ecom-restock, ecom-support, ecom-ugc-brief, ecom-list-clean, ecom-ppc-trim, and ecom-ad-iterate.

Skills (auto-loaded knowledge)

Three skills load context automatically into relevant commands:

1. ecom-context — store knowledge (brand voice, SKU model, segments, margins) read from STORE.md, CATALOG.md, and CUSTOMER.md
2. ecom-benchmarks — AOV-band benchmarks, category conversion rates, lifecycle timing data used by /ecom no-sales for gap analysis
3. ecom-copy — platform-specific copy rules: Shopify vs Amazon byte limits, PDP structure, email subject line formats, ad character counts

Agents (read-only specialists)

Both agents are read-only auditors/researchers. They produce findings and hand off — they do not block commands or loop outputs through a quality gate.

• ecom-flow-audit agent — audits lifecycle flows against send timing, segment logic, and copy quality; produces a ranked findings report
• ecom-page-audit agent — audits PDPs and collection pages against CRO criteria; outputs severity-ranked issues with the fix command named for each

What is the flagship command and why does it matter?

The flagship is /ecom no-sales — a store triage command that benchmarks your store against AOV-band data and surfaces the gaps with the highest revenue impact.

The typical operator starting point is: "sales are slow, I don't know where to start." /ecom no-sales answers that. It reads your STORE.md context, loads the ecom-benchmarks skill with AOV-band conversion rates and lifecycle timing norms, and produces a gap report sorted by estimated revenue delta. Is your cart abandonment rate above benchmark? Is your average order value below the floor for your category? Are you missing a post-purchase sequence entirely? It tells you, ranked, and names the command that closes each gap.

The output is a structured out/ecom/no-sales-report.md — not a reviewer that asks "are you sure?", but a report you act on. That is the v2 design principle across all of ClaudeKit: commands end with evidence (report, diff, verified file), not a gate.

How do you install and run EcomKit?

# Install via CLI (global, available across all projects)
ck auth <your-license-key>
ck install ecom
 
# Or install locally for a single project
ck install ecom --local
 
# Check what's installed and token counts
ck list
ck tokens ecom
 
# Run the flagship diagnostic
/ecom no-sales
 
# Build a Klaviyo lifecycle program from scratch
/ecom flows all --platform klaviyo
 
# Generate Amazon listing from STORE.md context
/ecom amazon --asin B0XXXXX

Alternatively, install through the plugin marketplace: /plugin marketplace add Madni-Aghadi/claudekit-ecom. The token ledger prints on every install so you know the exact context cost before you run anything.

EcomKit requires claudekits v0.1.3 on npm. If the install fails, run ck doctor — it diagnoses auth, token cache, and context file issues.

How does the store spine chain together?

The spine is a loop, not a line. Each stage produces artifacts that feed the next.

1. Foundation — /ecom no-sales reads STORE.md/CATALOG.md/CUSTOMER.md (built by ecom-context skill on first run). Everything else reads from these files.
2. Catalog and PDP — /ecom listing and /ecom page-audit produce on-brand copy using the brand voice and SKU structure from context.
3. Lifecycle — /ecom flows designs the full Klaviyo-style flow architecture and writes welcome (3 emails), abandoned-cart (3), browse-abandon (2), post-purchase (4), and winback (3) sequences. Timing and segments come from ecom-benchmarks.
4. Reviews and ads — /ecom reviews mines VOC, maps objections, and produces review-to-angle artifacts. /ecom ads consumes those angles to write platform-native creative. Ads built from real review language consistently outperform ads built from imagination.
5. Retention loop — post-purchase flows include review requests, which generate reviews, which feed back into /ecom reviews next cycle. The loop compounds.

The key design choice: ecom-context skill writes STORE.md, CATALOG.md, and CUSTOMER.md on first run, and every subsequent command reads those files rather than re-prompting for context. Brand voice does not drift across 20 runs.

How does EcomKit compare to other Claude Code ecommerce options?

There are three realistic alternatives: free skill repos, general marketing kits, and building your own prompts.

MarketingKit (/marketing) covers email and content well but is not store-specific — it does not know about AOV bands, Amazon byte limits, Klaviyo flow architecture, or SKU-level catalog structure. EcomKit was purpose-built for those workflows.

The real cost of free alternatives is not zero. We measured free skill repos in a controlled test and found they average 2.3x the token cost of equivalent EcomKit commands for the same output quality, because they lack the structured context model. See The Real Cost of Free Skills for the full breakdown.

What does the Amazon workflow look like?

Amazon is a first-class citizen in EcomKit, not an afterthought. The /ecom amazon command covers:

• Titles — keyword-first, under 200 bytes, brand style from STORE.md
• Bullets — 5 feature-benefit bullets, under 1,000 bytes each, keyword-seeded
• Description — narrative copy under 2,000 characters
• A+ content — module-level copy for Enhanced Brand Content
• Backend keywords — byte-limit-aware, deduped against visible copy, search-terms field aware

The byte-limit awareness is the part that matters. Amazon's limits are character-counted in bytes, not characters, which creates edge cases with special characters. /ecom amazon handles that transparently — you get compliant output without counting bytes manually.

We cover the Amazon workflow in depth in AI Amazon Listing Optimization.

Why do commands end with evidence instead of a reviewer gate?

EcomKit v2 drops the blocking reviewer/quality-gate agent architecture from v1. In the old design, a cro-auditor agent would score copy 0-100 and loop back up to three times before releasing output. In practice this added latency, consumed tokens on iteration, and produced false confidence — a 90/100 from an AI auditor is not validation.

The v2 design is simpler: commands end with evidence. /ecom no-sales ends with a gap report you can read and act on. /ecom listing ends with the listing copy and a structured JSON-LD block you can inspect. /ecom ads ends with the creative matrix and the review-mined angles it came from. You can see the reasoning, not just the verdict.

The two read-only agents (flow-audit, page-audit) produce structured findings reports — they are researcher/auditor roles that run when you explicitly call them, not gatekeepers that block other commands. This is the architecture described in Claude Code Agents vs Skills vs Slash Commands.

What does EcomKit cost?

EcomKit is available on all ClaudeKit pricing tiers:

• $14.99/month — single kit (EcomKit only)
• $29.99/month — Pro (any 3 kits, swap 1 per cycle)
• $49.99/month — All-Access (all 5 kits)
• Annual plans — $119/$239/$399/year (equivalent to ~2 months free)
• $99 one-time — lifetime license per kit, as shipped (no future updates)

All plans include 3 devices and a 14-day refund window. See pricing for full details.

The other four kits: EngineerKit (25 commands, 20,413 tokens), MarketingKit (20 commands, 16,714 tokens), VideoKit (17 commands, 12,602 tokens), SEOKit (19 commands, 16,004 tokens). Total across all 5 kits: 101 commands, 19 skills, 13 agents, 82,197 measured tokens.

FAQ

Does EcomKit work with Shopify and Amazon at the same time?

Yes. The ecom-context skill and STORE.md context file work across both platforms. Commands branch on platform context — /ecom listing produces Shopify-optimized PDP copy by default; add --marketplace amazon and it switches to byte-limit-aware Amazon listing copy. The same brand voice and SKU structure drive both.

How do I start if I have no STORE.md file yet?

Run /ecom no-sales first. The ecom-context skill detects missing context files and walks you through creating STORE.md, CATALOG.md, and CUSTOMER.md before running the triage. It takes 5-10 minutes and the files persist in your .claude/ directory for every future command.

How much does a typical store workflow cost in tokens?

The full kit loads at 16,464 tokens. Individual commands load a subset: a listing command is roughly 4,000-6,000 tokens; a flows command that writes 15 emails across 5 sequences is closer to 12,000-14,000 tokens. Run ck tokens ecom to see the current measured count, or ck tokens ecom --command flows to preview a specific command's cost before running it.

Is EcomKit useful for smaller stores or only at scale?

It is most useful once you have a real store with real SKUs and real customers — at least a few months of transaction data and some reviews to mine. The benchmarks in ecom-benchmarks skill are most meaningful when you can compare your actual metrics against them. If you are pre-launch, /ecom listing and /ecom amazon still produce solid copy from a STORE.md, but the diagnostic commands need data to diagnose.

Does EcomKit replace Klaviyo, Postscript, or other ESPs?

No. EcomKit writes the architecture and copy; you implement it in your ESP. The /ecom flows command produces a flow map and all message copy in a structured format you can paste into Klaviyo's template editor or export as a brief for your developer. The ecom-flows skill includes Klaviyo-specific formatting rules (block types, conditional content syntax) to make that handoff clean.

How is EcomKit different from using Claude directly without a kit?

Three differences: shared context, command chaining, and measured token cost. Without EcomKit, every Claude session starts cold — you re-explain your brand, margins, and customer segments every time. With STORE.md and the ecom-context skill, that context loads automatically. Commands chain whole workflows (reviews → angles → ad copy) rather than requiring you to manage the handoffs manually. And every command has a measured token cost you can check before running, which matters when you are running 20 commands across a store build.

If you run a store and want to stop pasting context into Claude on every session, EcomKit is the place to start. The no-sales diagnostic alone — benchmarking your store against AOV-band data in a single command — tends to surface one or two changes that pay for the kit in the first week.