Lining up 4 EC malls and fixing 60 items across 3 sites — in half a day

2026-05-14 ／ Vol 10 ／ draft at the time of publishing

In Vol 9, I wrote about sharing Claude's memory across multiple PCs.
That was about tidying up the inside of my environment. Today (2026-05-14) went the other way — outward work, finished in half a day.

Lined up, the day looked like this:

🛒 Rebuilt every setting on Color Me Shop (rooo.shop-pro.jp) from zero
🅱 Did the same for BASE (rooo.base.ec) — full setup including PayPal and 2-factor auth
🇶 Audited every setting on Qoo10 (qoo10.jp/shop/rooo), registered the product, fixed origin country
📦 Listed the same product as on Yahoo! Shopping ("Schick Hydro 5 Custom") on all three malls — same images, same description
🔗 Updated related links on esynet.jp / rooo.pro to include all 4 malls
🔍 Ran parallel audits on 3 owned sites (esynet.jp / rooo.pro / karaha.org) with Claude and fixed 60+ items

All in half a day. Here's the walk-through.

Lining up 4 EC malls

Until today, rooo's listings were basically just on Yahoo! Shopping. The other malls existed as accounts but were empty inside.

Today I filled them all in.

One product (Costco-only model "Schick Hydro 5 Custom"), placed on all 4 malls with exactly the same information: name, price, images, description, JAN code, country of origin, shipping method. Same operating rules across all four.

Turned out the per-platform details are more particular than I expected:

To sell "the same product in 4 places," you have to re-learn the rules in each place once. That's the unglamorous part of solo e-commerce.

Two misses Claude caught

Two mistakes I would have shipped without Claude's pickup:

1. BASE's main product image was actually a "next-day delivery rate" text image
I uploaded the 8 Yahoo! Shopping images in their original order. The first slot ended up showing a delivery rate chart instead of the package shot. I opened the live storefront, noticed it looked off, asked Claude to render each image one by one, and we figured out the ordering. Re-uploaded, re-saved.

2. The Specified Commercial Transactions page said "Rakuten Pay" but I never enabled Rakuten Pay
After reconfiguring Color Me's payment methods, the old template text "Rakuten Pay" was still on the public legal page. Caught it when checking that BASE and Color Me's payment lists matched up.

Neither was Claude's "independent discovery." Both started with me feeling slightly off and asking Claude to confirm.
But the cost of that confirmation is now extremely small. It happens many times a day. The friction of "huh?" is near zero.

"The images were duplicated"

When picking 4 images for Color Me, I initially grabbed Yahoo!'s first 4 (main + 2 crops + back). The user (me) looked at it and pushed back: "The images are duplicated."

I had Claude render each image. The first two were both front-of-package shots — one full, one zoomed on the top half. Clearly redundant.

The reselection logic Claude proposed:

1. Whole package (the "face" of the product)
2. Close-up showing all 17 cartridges (visualizing quantity)
3. Back of package (specs / ingredients / country of origin)
4. Next-day-delivery rate by month (signaling shipping reliability)

Going from "fill the slots" to "each slot has a distinct role" matters more when the cap is tight (4 images). Duplicates show up the moment you optimize for count.

Parallel-agent audits across 3 sites

Once the malls were lined up, I moved to auditing my 3 owned sites with Claude.

Three parallel agents, simultaneously:

🟢 esynet.jp audit — the business' official info site
🟢 rooo.pro audit — this site
🟢 karaha.org audit — the mental health peer community

Reports came back totaling 50+ items. I sorted by priority and dispatched fixes across 9 commits:

The push order: P0 (critical) → P1 (info gaps) → low (spelling / SEO) → mid (orphan pages) → leftovers (accessibility). 5 rounds of commits.

Especially for karaha.org, fixing "option E missing in the photo-consent policy page" came first — a policy-vs-implementation inconsistency in a sensitive-information domain directly affects user trust.

Why a half-day is enough now

Writing this up, I notice the feeling itself — "half a day is enough for this" — has become my baseline.

In Vol 8 I wrote about starting from "one-finger typing in 2019." Six years later.
A day like today no longer feels special.

What's actually doing the work:

1. Run agents in parallel
The 3-site audit doesn't run sequentially — it runs simultaneously. 30 extra seconds writing prompts saves a full day of human time.

2. Claude senses "something feels off here"
Wrong main image, spelling inconsistency, missing structured data, policy text drifted from implementation. Things I'd miss surface as audit findings.

3. "Fix" stays inside the same session
Not "audit, then come back later to fix." Right then, 9 commits split out and pushed. No PR review wait. Cloudflare Pages refreshes within minutes. "Notice → fix → ship" closes in a single session.

The tooling layers line up, so within half a day, "notice → fix → ship" keeps cycling.

The evening: four storefronts side by side

To close: I opened all 4 official stores and looked at them in parallel:

・Yahoo! Shopping rooo
・Color Me rooo
・BASE rooo
・Qoo10 rooo

The same product. The same ¥2,990 tax-included. The same shipping-included. The same description. The same "Yamato Transport Takkyubin / Nekopos" carrier text.

It's the most basic job for a one-person e-commerce shop, but it's the kind of work my 2019 self couldn't have finished in half a day.
That this lineup exists tonight is the outcome of cycling with Claude.

Next week is probably the order-handling side of these 4 malls (confirmation → packing → shipment → notification). Or the lower-priority items I left on the table, like unifying the older karaha.org event-page footers.