E-comm AI Project Stack

Two brand-agnostic Claude Code project templates for running an e-commerce brand's build and marketing work with AI. Both are stripped of all brand specifics and seeded with a consistent placeholder system, so you can hand either folder to your own Claude Code instance and stand it up for any brand in minutes.

This repo holds the Claude intelligence layer (instructions, subagents, slash commands, hooks, reference frameworks). It is not the live store or theme — it's the operating system that drives the work on top of them.

.
├── README.md                     ← you are here
├── TEMPLATE-CONVENTIONS.md       ← the placeholder system — read this first
├── .gitignore
├── website-build-stack/          ← STACK 1: the dev / site-build workspace
│   └── .claude/                  ← CLAUDE.md, 6 subagents, agent-memory, routing hook, settings
└── marketing-strategist-stack/   ← STACK 2: the creative strategist / copywriter workspace
    ├── CLAUDE.md                 ← the brand brain
    ├── .claude/                  ← slash commands, on-demand reference-router hook, settings
    └── frameworks/               ← strategy + deliverable frameworks (some local-only)

---

The two stacks

1. website-build-stack/ — the build workspace

Drop the .claude/ folder into your store's theme repo and Claude becomes a delegating build lead. A UserPromptSubmit hook enforces mandatory delegation to six specialist subagents:

Domain	Subagent
Shopify theme / Liquid (sections, snippets, OS 2.0, metafields, theme JS/CSS)	shopify-liquid-expert
HTML / CSS / Tailwind / vanilla JS, landing pages, HTML email, a11y/SEO	html-expert
Conversion / CRO audits of a page, copy, or behavioral data	cro-audit-specialist
Checkout Champ funnels, order forms, upsells/downsells, gateways, CRM integrations	checkout-champ-dev
Post-purchase flows: thank-you pages, upsell sequences, review timing	post-purchase-flow-builder
Customer surveys: NPS, CSAT, post-purchase, churn, profiling	crm-survey-builder

Each subagent has its own project-scoped memory under .claude/agent-memory/. The stack also carries a complete Shopify Liquid language reference inside CLAUDE.md.

Default platform: Shopify (themes/Liquid) + Checkout Champ (funnels). The specialist agents apply broadly; if a brand runs a different platform, swap the platform tokens and adjust the agents.

2. marketing-strategist-stack/ — the strategist workspace

Open the folder and CLAUDE.md loads the brand brain: role, compliance hard-stops, brand voice, product line, customer avatars, output formats, and an objection bank. Deeper knowledge lives in frameworks/ and is pulled into context on demand — a hook scans each request against reference-docs.json and tells Claude to read only the framework a request actually touches.

Slash commands for common deliverables:

Command	Does
/setup	Interviews you and fills the entire brand brain (run this first)
/ad	Paid ad copy for cold traffic
/email	Email flow copy
/social	Organic content script (TikTok / IG / YouTube)
/landing	Website / landing page copy
/compliance	Audit copy against your category's hard-stops and rewrite the risky parts
/avatar	Recommend the right avatar for a brief, or recall a profile

---

Quickstart

1. Clone this repo and pick a stack (use one or both).
2. Read TEMPLATE-CONVENTIONS.md. It defines the three markers you'll see everywhere:
   • [[TOKEN]] — a value to swap in (find them all with a search for [[).
   • > 📝 FILL IN — what belongs in a section and what good looks like.
   • > 💡 EXAMPLE (generic, replace or delete) — a non-brand illustration to overwrite or delete.
3. Set up the marketing stack: open marketing-strategist-stack/ in Claude Code and run /setup — Claude interviews you and fills in CLAUDE.md and the committed frameworks. (Or fill the tokens by hand.)
4. Set up the build stack: open your theme repo with the website-build-stack/.claude/ folder in place, then fill [[STORE_DOMAIN]], [[LIVE_THEME_ID]], and the repo-layout callout in CLAUDE.md. Copy settings.local.json.example → settings.local.json and keep/trim the Shopify CLI permission allowlist.
5. Verify nothing's left blank: search the repo for [[, FILL IN, and EXAMPLE (generic.

Requirements

• Claude Code (CLI, desktop, or IDE extension).
• PowerShell 7+ (pwsh) — both stacks use a UserPromptSubmit hook written in PowerShell. On non-Windows, install PowerShell or port the two small .ps1 hooks to your shell.
• Shopify CLI — for the website-build stack's theme pull/push workflow (optional but assumed).

---

What's git-ignored (local-only)

The deliverable-format frameworks sit closest to publishable copy, so they stay on your machine and out of the shared repo:

• marketing-strategist-stack/frameworks/email-campaigns/
• marketing-strategist-stack/frameworks/funnels/
• marketing-strategist-stack/frameworks/video-ads/

Each of those folders keeps a committed README.md so the structure stays documented even though the contents are ignored. Everything else — the strategy frameworks (brand constitution, avatars, product/benefits bible, platform playbook, Meta-algo playbook) — is committed as a skeleton: methodology, structure, fill-in guidance, and generic examples only, never finished publishable copy. Per-user settings.local.json and agent-memory working state are also ignored.

When you fill these templates in for a real brand, your brand's content lives only in your local clone. Keep it that way unless you intend to publish it.

---

Conventions for contributors

Keep this repo brand-neutral. When editing a committed file: use [[TOKENS]] for anything brand-specific, never commit a real brand's name/claims/pricing, and keep committed frameworks at the skeleton-plus-guidance level. See TEMPLATE-CONVENTIONS.md.