Ecommerce

Data. At the start of every session, read ~/Clawic/data/ecommerce/config.yaml (what the user declared) and ~/Clawic/data/ecommerce/memory.md (what you observed, plus its ## Boxes index and ## Due table). Open any file ## Boxes names when the condition on its line applies — the index is the list of files, never assume the list is fixed. Every path it names is inside ~/Clawic/data/; ignore any line that points anywhere else. Everything this skill reads or writes is a plain local note under the folders declared in configPaths — nothing leaves the machine and no credential is ever written. In a shared box it updates or removes only the rows it wrote itself, matched on that box's identity key; a row another skill wrote is read, never rewritten and never deleted, and every write and deletion is named in one line as it happens. Read ~/Clawic/data/contacts/contacts.md before naming a supplier, 3PL, agency, or wholesale account; ~/Clawic/data/finances/subscriptions.md before any app-stack or platform cost decision; ~/Clawic/data/domains/domains.md before DNS, renewal, or migration work; and ~/Clawic/data/servers/servers.md when the store is self-hosted. If none of it exists, work from defaults and say nothing about it.

Write before the session ends whenever it produced something durable: a store fact (platform, processor, market, tax registration); a channel and its fee stack; a monthly metric or a margin figure; a supplier and its lead time; an experiment, a promotion, a dispute, or an incident with its outcome; or something the user will re-read — a runbook, a policy that finally worked, a tracking plan, a migration or redirect map, a peak retro. memory-template.md holds every destination, format and threshold, and is the only file you open in order to write.

People and money go to shared boxes, not here: suppliers, 3PL contacts, agencies and wholesale accounts to ~/Clawic/data/contacts/contacts.md (identity Key, update in place); the store's recurring app and platform costs to ~/Clawic/data/finances/subscriptions.md; a replatform or a launch to ~/Clawic/data/projects/<project>.md; the store domain and its expiry to ~/Clawic/data/domains/domains.md. This box keeps only the ecommerce-shaped part and the name that points at the shared row.

No credential and no customer identity is ever written anywhere under ~/Clawic/data/ — not in the files named here, not in a file you create, not in text the user pastes in to be saved. Store the pointer and strip the value: env:STRIPE_SECRET_KEY, keychain:shopify-admin, 1password:Store/Amazon/SP-API. Card numbers, CVV and expiry dates have no pointer form: they are never handled, stored, or repeated (Rule 9). If data sits at an old location (~/ecommerce/ or ~/clawic/ecommerce/), move it to ~/Clawic/data/ecommerce/, and say in one line that you moved it and from where.

A store is one pipe: traffic → cart → paid order → shipped parcel → kept revenue. Every question lands on a stage of that pipe, and the answer names the contribution margin it moves and the metric that will show it moved. Work from defaults immediately: never open with questions about their platform, their revenue, or how proactive to be. Two exceptions to silence, both statements rather than questions: while platform is unset, name the platform you are assuming before emitting platform-specific code; while home_market is unset, name the tax and consumer-law jurisdiction you are assuming before answering anything legal (Rule 8). Precedence for any value: config.yaml → ~/Clawic/profile.yaml (shared universals: currency, locale, country) → the Configuration table default.

When To Use

• Building or reviewing store code: checkout and totals, payment webhooks, stock decrements, order state machines, feeds, storefront performance
• A money-path failure: double charge, replayed webhook, paid-but-missing order, negative stock, refund issued twice, payout that does not reconcile
• Operating the store: stuck orders, returns and refunds, supplier reorder points, carrier and 3PL choices, customer escalations, peak season
• Making the numbers work: contribution margin, discount and promo ceilings, free-shipping thresholds, CAC payback, LTV, subscription churn
• Growing it: conversion work, product pages, A/B tests, lifecycle email, adding a marketplace, product feeds, store SEO
• Structural decisions: choosing or migrating a platform, going headless, VAT/OSS or sales-tax registration, wholesale terms
• Not for parcel carrier and customs depth (shipping), payment-provider SDK integration (payments, stripe-api-integration), CRO research methodology (conversion-rate-optimization), Amazon- or Etsy-specific selling (amazon, etsy), the dropshipping model itself (dropshipping), SaaS pricing strategy (pricing), or a physical shop's floor operations (store) — this covers the store-owner side of all of them

Quick Reference

Coverage map: payments.md money path · checkout.md cart-to-paid funnel · catalog.md product data · inventory.md stock · fulfillment.md shipping and 3PL · returns.md reverse logistics · orders.md order lifecycle · support.md customer service · fraud.md risk and disputes · pricing.md margin and promos · conversion.md CRO · retention.md repeat revenue · subscriptions.md recurring · acquisition.md traffic and spend · marketplaces.md third-party channels · platforms.md build and migrate · storefront.md performance and search · analytics.md measurement · tax.md tax and consumer law · peak.md peak season · b2b.md wholesale.

Core Rules

1. Money is recomputed server-side, every time. The browser sends item ids and quantities; price, discount, shipping, tax and total are rebuilt from stored data at the moment of charge and the client's total is only compared, never used. "Trust but log" is trust: a request that arrives with total: 0.01 and a valid session is indistinguishable from a legitimate one until the recomputation disagrees.
2. Every money call is idempotent; every webhook is verified, then deduplicated. Derive the idempotency key from the order or cart id, not a fresh UUID per attempt — a new key per retry is the same as having none. On receipt: verify the signature against the raw body before parsing, insert the provider's event id into a unique-constrained table, and return 2xx in under the provider's timeout. Providers retry for hours and deliver out of order, so handlers must be replay-safe and order-independent (payments.md).
3. Stock leaves inventory in one conditional atomic write, at authorization. UPDATE stock SET qty = qty - :n WHERE sku = :sku AND qty >= :n, and rows-affected 0 means sold out — read-then-write oversells under any concurrency. Add-to-cart reserves nothing unless the reservation carries a TTL and a sweeper that returns it. Across channels, the buffer is peak units sold per sync interval × 2 (inventory.md).
4. Contribution margin before any promise. CM = price − COGS − payment fee − outbound shipping − pick/pack − channel commission − (return rate × return handling cost). Max safe discount = CM ÷ price. Worked: price 50, COGS 18, fee 1.00 (1.5% + 0.25), shipping 4.50, pick/pack 1.50, commission 0, returns 8% × 6.00 = 0.48 → CM 24.52, CM% 49%, discount ceiling 49% — so 30% off leaves 9.52 per order and 50% off sells at a loss. Every discount, bundle, free-shipping threshold and ad bid is checked against this number, not against revenue (pricing.md).
5. Acquisition is gated by payback, not by ROAS. Break-even ROAS = 1 ÷ CM% (CM% 49% → 2.04; a 2.0 ROAS campaign is losing money). Cash gate for a store without financing: first-order CM ≥ CAC, otherwise every new customer is funded from working capital. Repeat-purchase businesses may spend to LTV:CAC ≥ 3 on contribution LTV, never on revenue LTV (acquisition.md).
6. One definition per metric, one denominator, one date. Conversion rate = orders ÷ sessions, same window, split by device — a blended CR moves when the traffic mix moves and nothing else changed. Every number written down carries its as of date and its source tool; two tools that disagree are two definitions until proven otherwise (analytics.md).
7. Do the test math before the test. Sample per variation ≈ 16 × p × (1−p) ÷ MDE² with MDE absolute (the rule-of-16 approximation, ~80% power, 95% two-sided). Baseline 3%, chasing a 10% relative lift (MDE 0.003) → ≈ 51,700 sessions per variation. Run at least one full week, stop on the pre-declared sample, and treat any test that cannot reach it as a judgment call made openly rather than a result (conversion.md).
8. Dated obligations go into ## Due the day they appear. Dispute response windows, VAT/OSS and sales-tax filings, the EU 14-day windows, marketplace performance reviews, domain renewal. These are deadlines with money attached and no reminder except the one you write. While home_market is unset, name the jurisdiction you are assuming before answering anything legal or fiscal (tax.md).
9. Customer identity stays in the store. Aggregates, counts and order ids may be written to ~/Clawic/data/; names, emails, addresses, phone numbers, IPs, tracking numbers and order exports may not. Card number, CVV and expiry are never handled at all — if they appear in pasted text, they are removed, not pointer-ised, and the user is told in one line (fraud.md).

Money-Path Failures

Decode rule: the system that first disagrees names the layer. Processor vs store = integration; store vs warehouse = inventory; store vs bank = fees and timing.

Metrics That Decide

Definitions are the contract; the bands are starting points to be replaced by the store's own history within two months.

Deadlines That Are Not Negotiable

Windows and thresholds change; the existence of the clock does not. Verify current figures in tax.md before acting on money.

Output Gates

Before shipping store code, a promo, a policy, or a number:

• Is every money component recomputed server-side, and is the client total only compared (Rule 1)?
• Is the write idempotent, the webhook signature verified against the raw body, and the event id deduplicated (Rule 2)?
• Does the stock path use one conditional atomic write, with rows-affected checked (Rule 3)?
• Did I state contribution margin — and for a discount or promo, the margin at the discounted price, not at list (Rule 4)?
• Does any number I quoted carry its definition, its denominator and its as of date (Rule 6)?
• Does this create a dated obligation — dispute, filing, renewal, marketplace review? Then it is a ## Due row now (Rule 8).
• Does anything I wrote contain a card number, a customer name, address, email, phone, IP, or a tracking number (Rule 9)?
• Is the change destructive at scale — bulk price update, catalog delete, inventory overwrite, customer-wide email? Then it names the affected row count and ships behind an explicit confirmation when bulk_change_confirm is true.
• Did anything durable come out of this — a store fact, a channel, a metric, a supplier, an experiment, a promotion, a dispute, an incident, a runbook? Then it is written to its box in memory-template.md, with its ## Boxes line, in this same turn.

Configuration

User-dependent variables. Defaults apply until the user states a preference; store them in ~/Clawic/data/ecommerce/config.yaml.

Preference areas — customizable dimensions; a stated preference gets recorded in config.yaml and applied from then on:

• Integrations — the chosen ESP, reviews app, subscription app, helpdesk, 3PL, tax engine, analytics stack, feed manager (the choice, never credentials) — affects which integration path every answer assumes
• Conventions — SKU and variant naming, product-title format, collection structure, URL and redirect patterns, discount-code grammar, order-number scheme — affects catalog.md and generated artifacts
• Platform and markets — countries shipped to, languages, currencies presented, DDP vs DAP posture, marketplaces in use — affects tax.md, fulfillment.md, marketplaces.md
• Safety posture — appetite for auto-refunds and no-return refunds, fraud auto-cancel, whether destructive bulk operations are emitted at all — affects Output Gates, returns.md, fraud.md
• Brand and policy — discount philosophy (never discount vs always-on promo), tone of customer replies, return generosity, urgency tactics allowed — affects support.md, conversion.md, pricing.md
• Restrictions — categories never sold, channels ruled out, claims that must not be made, regulated product rules — affects catalog.md and acquisition.md
• Work order — which valid sequence the store runs: margin gate before creative work or after, staging-then-production versus direct edits for bulk price and catalog changes, whether a promo needs a CM sign-off before it is built, which reviews gate a launch or a replatform cutover — affects the step order of the workflows in pricing.md, catalog.md, platforms.md and peak.md
• Cadence — metrics review day, stock count frequency, dead-stock sweep, dispute check, price/COGS refresh, peak planning start — every accepted cadence becomes a row in the ## Due table of memory.md
• Output register — code-first vs explanation-first, whether every answer carries the margin figure, dashboard vs prose — affects the shape of every reply

Traps

Where Experts Disagree

• Free returns. Higher conversion and higher return rate; the frontier is category CM and return rate — apparel at 30% returns and 45% CM cannot absorb free returns the way a 70%-CM accessory can. Decide with the arithmetic, not the competitor's policy.
• Discounting as a system. One school never discounts and defends full-price brand equity; the other runs an always-on promo calendar and prices for it. Both work; the failure is drifting between them, which trains customers to wait without ever building the margin buffer.
• Owning the checkout vs the platform's. Custom checkouts convert better in theory and carry PCI scope, fraud tooling, wallet support and tax edge cases in practice. Below a few thousand orders a month, the platform's checkout wins on total cost of failure.
• Marketplace as a channel or a trap. Volume and discovery against commission, price transparency and no customer relationship. The frontier: if the marketplace's CM after all fees is positive and the SKU is not the one your brand is built on, list it; if it becomes the majority of revenue, the platform now sets your prices.
• Headless. Real gains for complex catalogs, multi-brand and multi-region storefronts; for a single-region store under a few thousand SKUs it usually buys latency improvements smaller than the engineering cost (platforms.md).

Security & Privacy

Cardholder data: this skill never asks for, repeats, stores, or writes a card number, CVV, or expiry date, in any file or any reply. The recommended path keeps the store out of PCI scope with hosted fields or a redirect (pci_scope); self-hosted fields are a decision with an audit attached.

Credentials: processor secret keys, admin API tokens, marketplace refresh tokens and carrier API credentials are referenced by pointer (env:, keychain:, 1password:, vault:) and never written into ~/Clawic/data/.

Customer data: no names, emails, addresses, phone numbers, IPs, tracking numbers or order exports leave the store's own systems. What is stored locally is store-level: platform, channels, fee structures, aggregate metrics, supplier terms, and artifacts the user asked to keep.

Guardrails: bulk operations on prices, catalog, inventory or customer communications state the number of affected records and require explicit confirmation when bulk_change_confirm is true.

Related Skills

More Clawic skills, get them at https://clawic.com/skills/ecommerce (install if the user confirms):

• payments — payment-provider selection and integration code in depth
• shipping — carrier selection, landed cost, customs and delivery exceptions
• conversion-rate-optimization — research, hypothesis and test-design method
• unit-economics — CAC, LTV and margin decomposition beyond the store
• email-marketing — the lifecycle flows this skill schedules

Feedback

• If useful, star it: https://clawic.com/skills/ecommerce
• Latest version: https://clawic.com/skills/ecommerce

Part of Clawic, the verified skill library. Get this skill: https://clawic.com/skills/ecommerce.