Studio Code: treat shop/store/sell requests as ecommerce deterministically

[youknowriad]

The main reason I started this PR is that I noticed that woo commerce was not consistently used when a user asked for a shop. The eval was passing most of the time but not all the time.

Related issues

• Related to the woocommerce-shop-setup eval case in scripts/eval/promptfoo.config.yaml.

How AI was used in this PR

Studio Code (the in-CLI agent) is the subject of the change — this PR tunes the plugin-recommendations skill it loads. The change was developed and verified with Claude: I reproduced the flaky behavior, strengthened the skill, and confirmed determinism by running the agent end-to-end against the WordPress.com AI proxy across multiple consecutive runs.

Proposed Changes

When a user asks the agent to build a shop or to "sell" something, they expect a working storefront — WooCommerce installed with representative sample products — not a few hand-built informational pages. Two intermittent failures caused the opposite:

1. Misclassification. Requests carrying scope/style qualifiers like "small", "just a couple of pages", or "for inspiration" were sometimes treated as informational content sites, so WooCommerce was skipped entirely.
2. Empty catalog under time pressure. Even when WooCommerce was set up, the agent often did theme scaffolding and visual polish first and ran out of its turn budget before creating any products — leaving a shop with nothing to sell.

This keeps the guidance where the WooCommerce logic already lives — the plugin-recommendations skill — without touching the core system prompt:

• The skill's trigger and description now make it unambiguous that selling anything ("I want to sell X", a shop, store, storefront, or ecommerce site, or any design implying products/prices/a catalog/a Shop page) is always an ecommerce request that requires WooCommerce and products — and that size/style qualifiers never downgrade it.
• The skill now tells the agent to front-load WooCommerce setup and product creation: install, configure, and populate the catalog before the slower theme-design and polish phase, and never let image sourcing block products from being created. The catalog is the core of the site, not a finishing touch.

Net effect for users: shop/store/"I want to sell …" prompts deterministically produce a real storefront with products, regardless of how small or stylistically framed the request is.

Testing Instructions

• From scripts/eval, run the regression case (Node ≥22 / npm ≥11):

  npx promptfoo@0.121.4 eval -c promptfoo.config.yaml --filter-pattern "WooCommerce" --no-cache
  

  The woocommerce-shop-setup case asserts WooCommerce is installed and ≥1 product is created.
• Verified on 3 consecutive promptfoo runs (all passed) with the final skill; each run installed WooCommerce and created 6 products, with product creation occurring before the design/polish phase. (An earlier skill-only attempt without front-loading flaked 1-in-3 by timing out after categories but before products — the front-loading change fixes that failure mode.)
• npm run typecheck passes.

Pre-merge Checklist

• Have you checked for TypeScript, React or other console errors?

🤖 Generated with Claude Code

[wpmobilebot]

📊 Performance Test Results

Comparing cef2c42 vs trunk

app-size

Metric	trunk	cef2c42	Diff	Change
App Size (Mac)	2355.89 MB	2355.89 MB	+0.00 MB	⚪ 0.0%

site-editor

Metric	trunk	cef2c42	Diff	Change
load	1717 ms	1802 ms	+85 ms	🔴 5.0%

site-startup

Metric	trunk	cef2c42	Diff	Change
siteCreation	8510 ms	8504 ms	6 ms	⚪ 0.0%
siteStartup	3907 ms	3879 ms	28 ms	⚪ 0.0%

---

Results are median values from multiple test runs.

Legend: 🟢 Improvement (faster) | 🔴 Regression (slower) | ⚪ No change (<50ms diff)

[katinthehatsite]

I think the changes look good 👍

Empty catalog under time pressure. Even when WooCommerce was set up, the agent often did theme scaffolding and visual polish first and ran out of its turn budget before creating any products — leaving a shop with nothing to sell.

I ran into this multiple times, especially with the first page having products but then clicking each product led to empty URLs. This was also a common issue with product categories. I think it is a good direction to prioritize making full products first before setting up everything else to make the user experience more complete 👍