Make creating WooCommerce shop sites easy with Studio Code - #3741
Conversation
📊 Performance Test ResultsComparing e5393f8 vs trunk app-size
site-editor
site-startup
Results are median values from multiple test runs. Legend: 🟢 Improvement (faster) | 🔴 Regression (slower) | ⚪ No change (<50ms diff) |
Teach the Studio Code agent to reliably build real ecommerce stores when a user asks for a shop, instead of producing a static catalog mockup. - Add a WooCommerce runbook to the plugin-recommendations skill: treat a shop/store request as ecommerce by default, install WooCommerce, disable "Coming soon" mode, create contextual products with imported images, and add a mini-cart to the header. - Have the site-spec summary commit to a "Functionality & plugins" line so the plugin decision happens up front rather than being skipped. - Polish every page (user + WooCommerce), looping only the home page. - Render the Skill tool call as a single compact row in the transcript. - Add a minimal eval asserting a shop request installs WooCommerce and creates products. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The agent was skipping the site-spec skill when a build prompt already provided the site name, which meant the "Functionality & plugins" decision (and thus WooCommerce setup for shop requests) never happened. Make loading site-spec unconditional for new-site builds, and clarify that skipping discovery only skips the questions — the site spec is still produced. Verified with the woocommerce-shop-setup eval, which now passes: the agent loads site-spec, installs WooCommerce, and creates products. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
71b9131 to
ae10111
Compare
The agent sometimes read "small" as a signal to skip WooCommerce. Make
explicit that size/scope words ("small", "simple", "just a few pages")
describe how big the site is, not whether it sells - a small shop is still
a shop and still gets WooCommerce.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
sejas
left a comment
There was a problem hiding this comment.
Nice! Thanks for improving the creation of WooCommerce websites.
I tested it by running npm run eval -- --filter-pattern "shop request sets up WooCommerce" and also npm run cli:build && node apps/cli/dist/cli/main.mjs code with the following prompt: Create a small shop of surfing boards top brands, keep it small only 2 pages. The name of the shop is CanarySurf
The shop looked good and it contained 4 products.
The eval failed to install WooCommerce and create the products. It seems we might need to tweak the skill description a bit more.
| # Regression: a "shop"/"store" request must set up WooCommerce WITH products, | ||
| # even when framed with scope/style qualifiers ("small", "for inspiration") |
|
I think the latest skill update might fix that eval issue. |
The renderer tweak to collapse the Skill tool call to a single row did not have the intended effect, so revert it and leave tool-result rendering as-is. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
| wp_cli option delete _transient__wc_activation_redirect | ||
| ``` | ||
|
|
||
| 3. Configure store basics. Match the currency and base location to the user's context when known, otherwise use sensible defaults. Also turn off "Coming soon" mode, which fresh WooCommerce installs enable by default - otherwise the storefront shows a "Something big is brewing! Our store is in the works and will be launching soon!" placeholder instead of the real site: |
There was a problem hiding this comment.
otherwise the storefront shows a "Something big is brewing! Our store is in the works and will be launching soon!" placeholder instead of the real site:
is this needed?
The WooCommerce pages were phrased as an illustrative "such as" example, so the agent could treat the single-product/Cart/Checkout pages as optional. Make them a firm checklist (Shop, single-product, Cart, Checkout, My Account) so every storefront page gets polished. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Drop the verbatim placeholder message from the "Coming soon" step; the instruction to disable it is enough. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
The unit test failure seems like the one being worked on separately and failing on trunk. So I'm merging this. |
Related issues
How AI was used in this PR
Built with Claude Code. The skill/prompt changes were driven by analyzing real Studio Code session transcripts of an agent repeatedly failing to build a proper shop ("MyPet") — each fix targets a specific failure observed in those transcripts (trigger never firing, then firing but misclassifying the shop as informational, then the cart link never landing). All changes were reviewed by the author.
Proposed Changes
When a user asks Studio Code to build a shop, the agent used to produce a static catalog mockup — no WooCommerce, no products, no cart — even for an explicit "pet shop." This makes ecommerce the reliable default and gives the agent a complete happy path for standing up a real store.
Testing Instructions
npm run cli:buildnpm run eval(look for thewoocommerce-shop-setupcase).Pre-merge Checklist
🤖 Generated with Claude Code