SESSION-0002.0 — Shopify BDD analysis bootstrap

Date: 2026-05-31 Driver: Ben Stull + Claude (Sonnet 4.6) Repos touched: wiggleverse-ecomm-meta (session record), Shopify Analysis workspace (~/projects/wiggleverse/ecomm/Claude/Shopify Analysis/ — not a git repo; flat file workspace)

Honesty note on this artifact. This is a session record authored by the assistant from the conversation — faithful and uncurated as to decisions, sequence, and wrong turns, but not the raw byte-for-byte transcript. OHM publishes raw transcripts via ohm-infra's publish-transcript.sh; ecomm has no such tooling yet. Building a transcript-dump path is a future roadmap item. Until then, records like this stand in — labeled as what they are.

Purpose

Kick off the Shopify feature-analysis workstream. The goal: walk every corner of the Shopify admin for a fresh store, document all functionality as Gherkin BDD scenarios, and build a navigable reference that drives decisions about what to build in the ecomm alternative. Session was scoped to analysis + spec generation only — no implementation.

What happened, in order

Access setup — wrong display, then blocked MCP. Ben granted Chrome access. The Claude-in-Chrome MCP attempted to navigate to admin.shopify.com from within its own tab group — blocked by the extension's org policy. Separately, the first computer-use screenshot landed on the wrong monitor (LG ULTRAGEAR 2, showing LinkedIn). Used switch_display to find LG ULTRAGEAR 1, where the Shopify admin was already open. Chrome remained read-only (tier: "read") for the rest of the session — all Shopify exploration was done via screenshots and training knowledge, not live clicking.
Domain map. Catalogued 18 functional domains from the admin nav (Home, Orders, Products, Customers, Marketing, Discounts, Content, Markets, Finance, Analytics, Online Store, Point of Sale, Apps, Settings — plus sub-domains). Produced shopify-domain-map.md with a session-breakdown table (~20 sessions, ~525+ scenarios estimated) and a "start here" recommendation. Saved to the Shopify Analysis workspace.
Style calibration. Ben asked to start with Settings & Configuration. Before writing the full set, produced two sample Feature blocks (General Settings + Staff Accounts) and surfaced five calibration questions: altitude, persona naming, error-scenario placement, Feature scope, and Scenario Outline usage. Ben approved the style as-is ("This is fantastic") without changes.
Settings & Configuration — 15 files written in parallel. Three subagent groups ran concurrently (Group A: general/plan/billing/ users-and-permissions/brand; Group B: payments/checkout/shipping-and- delivery/taxes-and-duties/locations; Group C: notifications/customer-data/ policies/metafields/domains). All 15 files confirmed present; line counts verified via bash. ~120 scenarios, 1,564 lines total.
- Minor inconsistency surfaced: general.md used third-person steps ("When the Store Owner saves…") while all other files used first-person ("When I save…"). Flagged to Ben; left uncorrected pending his call.
"Are you saying it's all done?" Ben questioned the completeness claim. Prompted a verification pass — ran ls + wc -l on all 15 files, then spot-read locations.md and taxes-and-duties.md in full. Confirmed complete and on-style. The question was a healthy catch; the earlier announcement had been too confident without the verification step.
Products — 6 files written in parallel. Two subagent groups: (A) products + product-variants; (B) collections + inventory + purchase-orders; (C) gift-cards. ~95 scenarios, 1,406 lines. Notable: products.md became the largest single file to that point (437 lines, 36 scenarios) owing to the product form's breadth. Key edge cases captured: draft orders don't reserve inventory; automated collections block manual product add/remove; the three inventory policy modes.
Orders — 5 files written in parallel. Three subagent groups: (A) orders; (B) fulfillment + returns-and-refunds; (C) draft-orders + abandoned-checkouts. ~175 scenarios, 1,230 lines. orders.md hit 410 lines (72 scenarios) across 7 groups. Exchange asymmetry (replacement costs more → payment request; less → refund) and the post-fulfillment editing block were explicitly captured — both are common re-implementation traps.
Memory and index written. Ben asked whether a new session would have enough context to continue. Checked the memory directory — the project memory file existed but MEMORY.md index was missing. Wrote both: updated project-shopify-bdd.md with full session progress and remaining domain queue; created MEMORY.md with the pointer entry. Ben confirmed this was sufficient to hand off.
Session record requested. Ben directed the record to SESSION-0002.0-TRANSCRIPT-2026-05-31.md with the pre-assigned filename. Requested access to the sessions folder (outside the connected workspace) via request_cowork_directory; Ben approved. Read SESSION-0001.0 to match house style; wrote this record.

Decisions citable later

Analysis-first, not implementation. This workstream produces a spec library; no code was written.
File structure locked: one folder per Shopify domain section, one .md file per sub-section/nav-item, flat within each domain folder. Naming: kebab-case matching the Shopify nav label.
BDD style rules locked (Ben approved in step 3):
- Business-behavior altitude — no UI clicks
- First-person steps; Background block per page context
- But for negative assertions; H2 headers for groups within a file
- No Scenario Outlines unless a table genuinely adds clarity
- Persona: "Store Owner" (full admin) / "Staff Member" (restricted)
Starting order: Ben overrode the recommendation (Products first) and chose Settings first. Both are valid; the domain map explains the dependency rationale.
Parallel subagents for file generation worked well at this granularity — style brief was tight enough that three concurrent agents produced consistent output.
Verification step is required before announcing completion. Lesson from step 5: ls + wc -l + spot-reads should be routine, not prompted by the user.

Open / carried forward

general.md style inconsistency — third-person steps vs. first-person in all other files. Flagged but not fixed; Ben to decide.
Chrome MCP blocked on admin.shopify.com — all future Shopify exploration sessions will rely on screenshots (read-only computer use) + training knowledge unless the extension's domain allowlist changes.
9 domains remaining (in queue order): customers, marketing, discounts, content, online-store, analytics, finance, markets, sales-channels.
~390 scenarios written; ~135+ estimated remaining to complete all 18 domains at current density.
No git history on the Shopify Analysis workspace — it is a flat folder, not a repo. If versioning is wanted, that's a future housekeeping item.

Next session

This is the next session of the ecomm Shopify BDD analysis project.

READ FIRST:
- Your memory for this project (project-shopify-bdd.md via MEMORY.md).
- ~/projects/wiggleverse/ecomm/Claude/Shopify Analysis/shopify-domain-map.md
  — the full domain list and session breakdown.
- ~/projects/wiggleverse/ecomm/git/wiggleverse-ecomm-meta/sessions/SESSION-0002.0-TRANSCRIPT-2026-05-31.md
  — especially "Decisions citable later" for the locked style rules.

CONTEXT:
Three domains are complete: settings/ (15 files), products/ (6 files),
orders/ (5 files). Running total: 26 files, ~390 scenarios.

TASK: Continue writing Gherkin BDD scenarios for the next domains in queue.
Work through as many as the session allows, in this order:

  1. customers/    — customer list, customer profile, customer segments
  2. marketing/    — campaigns, automations, UTM tracking, channel integrations
  3. discounts/    — discount codes, automatic discounts, Buy X Get Y

For each domain:
- Create a new folder under ~/projects/wiggleverse/ecomm/Claude/Shopify Analysis/
- Write one .md file per sub-section (match the Shopify admin nav labels,
  kebab-case filenames)
- Follow the locked style rules exactly (see SESSION-0002.0 "Decisions
  citable later" — business-behavior altitude, first-person, Background
  block, But for negatives, no Outlines unless needed)
- Use parallel subagents per the pattern from Session 2 — one agent group
  per domain, brief them with the full style rules, verify output with
  ls + wc -l + spot-reads before announcing completion

After completing each domain, update project-shopify-bdd.md in memory to
mark it done and update the running total.

DONE WHEN: all three domains above are written and verified, memory is
updated, and the session record is written to:
~/projects/wiggleverse/ecomm/git/wiggleverse-ecomm-meta/sessions/SESSION-0003.0-TRANSCRIPT-<date>.md

Session 2 closed here.

9.0 KiB Raw Blame History