The JSON-LD fast path only got items/categories/prices but no modifiers.
Removing it lets Uber Eats pages fall through to Claude AI extraction
which handles modifiers like every other platform.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Instead of clicking each menu item (broken by virtual scrolling),
extract item IDs from embedded JSON and make direct fetch() calls
to the itemPage GraphQL endpoint. 5 concurrent requests per batch.
Much faster and 100% reliable - no DOM interaction needed.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Pass extracted item names via temp JSON file so Playwright knows exactly
what to click instead of guessing from DOM selectors (7 → 171 items)
- Use TreeWalker for exact text matching and aggressive scrolling
- Better price parsing: handle cents (int), dollars (string), displayPrice
- Improved modal dismissal with overlay click fallback
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- New doordash-modifiers.js: stealth Playwright script that clicks each
menu item on a DoorDash page, captures itemPage GraphQL responses,
and extracts optionLists (modifier groups with options and prices)
- Wire modifier extraction into DoorDash fast-path in analyzeMenuUrl.cfm:
after parsing items/categories, runs modifier script and maps results
- Improved business info extraction: address, phone, and hours now use
position-based parsing of StoreHeaderAddress, StoreHeaderPhoneNumber,
and StoreOperationHoursRange embedded data (fixes intermittent missing info)
- Add playwright-extra and stealth plugin to package.json
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
StorePageCarouselItem uses imgUrl, but MenuPageItem uses imageUrl.
This gives ~540 item images instead of 26.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
listToArray treats delimiter as individual chars, not a string.
Rewritten to use position-based find() traversal for proper
multi-character delimiter splitting.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Extract items from MenuPageItemList (171 items) instead of StorePageCarouselItem (54)
- Categories already mapped to items via MenuPageItemList sections
- Cross-reference images from carousel entries by item name
- No need for Claude category assignment - data already structured
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Extract menu data directly from embedded JSON in DoorDash HTML:
- Categories from MenuBookCategory entries
- Items with names, descriptions, prices, and image URLs from StorePageCarouselItem
- Business info from page title and StoreHeaderAddress
- Uses Claude to assign items to categories
- Upgrades image URLs to 600px for better quality
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Update render.js on dev server to scroll page before capturing images
- Increase Playwright wait from 4s to 5s and timeout from 90s to 120s
- Upsize DoorDash CDN thumbnails from 150px to 600px when downloading
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Lucee's deserializeJSON fails on certain Claude outputs and the error
bypasses inner cftry/cfcatch. Parse via Jackson from file instead.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add smart quote/dash replacement for PDF-sourced text
- Add Jackson fallback parser for when Lucee's deserializeJSON fails
- Strengthen prompt to request properly escaped JSON
- Clean control characters more selectively
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
CFML was failing to compile analyzeMenuUrl.cfm because ' contains
a # character that Lucee interprets as variable expression start.
Escaped all 4 occurrences to &##39;.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Parse JSON-LD structured menu data from saved Uber Eats pages
(categories, items, prices, descriptions, business info)
- Show save-and-upload instructions when user pastes Uber Eats URL
- Always show header image upload step (was skipped for URL imports)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The WooCommerce fast path was returning empty business info. Now the
Playwright script extracts it from the page and the CFML passes it through.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The wizard checks item.imageUrl to decide whether to skip the image
upload step and download images from remote URLs instead.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Detects WooCommerce sites from Playwright HTML (woocommerce, wc-add-to-cart,
tm-extra-product-options). Runs woo-modifiers.js which navigates all product
pages, extracts items with categories, and scrapes TMEPO/variation modifiers.
Falls through to Claude if extraction fails.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Post-process Claude menu extraction to detect when >60% of categories
have exactly 1 item (a common misparse). Collapses pseudo-categories
into the nearest preceding real (0-item) category.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Added explicit guidance that categories are broad section headings
(5-15 typical) and items are individual products (30-150 typical).
Prevents Claude from treating each menu item as its own category.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Claude API rejects images when the declared media type doesn't
match the actual content. Now detects JPEG/PNG/GIF/WebP from
base64 magic bytes instead of trusting file extensions or
Content-Type headers.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Detect Grubhub URLs and fetch menu data directly via their
REST API instead of scraping HTML. Gets anonymous auth token,
then fetches full restaurant data including categories, items,
modifiers, prices, hours, lat/lng, tax rate, and item images.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Toast provides latitude/longitude in the location object. Extract
in analyzeMenuUrl.cfm and pass through to saveWizard.cfm, which
now includes lat/lng in the address INSERT. Skips the background
Nominatim geocode when coordinates are already available.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Parses upcomingSchedules from ROOT_QUERY.restaurantV2.schedule
and formats as "Mon 7:30am-6:30pm, Tue 7:30am-6:30pm" text string
that the setup wizard's parseHoursString() can consume.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add item.prices array support to first code path (was only checking
item.price scalar, but Toast now uses prices: [4.50] array)
- Extract individual address fields (addressLine1, city, state, zip)
from ROOT_QUERY restaurant data for saveWizard compatibility
- Update modifier extraction URL detection to match any toasttab.com
domain (not just order.toasttab.com)
- Update slug-based URL construction to use www.toasttab.com/local/order/
format instead of deprecated order.toasttab.com/online/ format
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Tag stripping via reReplace on 1.7M HTML was likely causing the silent
failure on biz server. Brace-counting doesn't need tag stripping since
HTML tags don't contain { or } and attribute quotes come in balanced
pairs. Also removed var keywords from page-level cfscript (may not be
supported in Lucee at template level) and added detailed error output
to the cfcatch for debugging.
Also auto-detect dev/prod environment from hostname instead of
hardcoded flag.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The regex .*? (non-greedy) fails on 500K+ JSON due to Java regex
backtracking limits, causing truncated data (only 3 of 6 menus
extracted). Replace all 3 extraction points with cfscript
brace-counting that reliably handles any JSON size.
Also decode HTML entities (& -> &, < -> <, etc.) from
Chrome View Source saves.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Chrome's Ctrl+U (View Source) saves wraps content in <span> tags
which breaks the regex termination pattern ;\s*window\. because
HTML tags appear between ; and the next window. variable.
Strip HTML tags from a working copy before regex extraction when
View Source format is detected (presence of <span id="line" tags).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Toast pages with multiple menus (e.g. "Food", "Beverages", "Merchandise")
now produce parent categories from the menu names with subcategories from
the groups within each menu, using the parentCategoryName field the wizard
already supports.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When analyzing Toast menu pages, items with modifiers now have their
modifier groups extracted by clicking each item in a headless browser
and intercepting the GraphQL MenuItemDetails responses. Extracted
modifiers include group name, required/optional flag, min/max selections,
and option names with prices. Items sharing the same itemGroupGuid
inherit modifiers from successfully mapped siblings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Restaurant info is in ROOT_QUERY.restaurantV2(...) keys, not
Restaurant:* top-level keys (Apollo cache format)
- Prices are in item.prices array [4.50], not item.price scalar
- Added null checks for imageUrls (can be null, not missing)
- Fallback to title tag for business name
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Instead of sending 450KB of HTML to Claude (which truncates to 100K
and only extracts ~60 items), parse the structured __OO_STATE__ data
directly on the server. This captures all menus, groups, items, prices,
and images from Toast pages - 169 items for Jus Family Cafe vs 60 before.
Falls back to Claude analysis if __OO_STATE__ parsing fails.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The Claude API sometimes returns explanatory text before the JSON
response even when instructed to return only JSON. Added extraction
logic to find the first { character and strip any leading text.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Parse HTML heading structure to detect h2 parents with h3 subcategories
- Append detected hierarchy to Claude prompt as explicit hint
- Post-process Claude response to enforce hierarchy even if Claude returns flat
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- analyzeMenuUrl.cfm: Detect subcategories from Toast subgroups and
Claude API responses, preserve hierarchy with parentCategoryName
- setup-wizard.html: Display subcategories indented under parents
throughout wizard flow (categories step, items review, summary, preview)
- menu-builder.html: Show subcategories nested in outline modal view
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Large menus (20+ categories) were getting truncated JSON responses
at 8192 tokens, causing parse failures.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Tax rate: Use Zippopotam (free, no key) to get state, then lookup
from built-in state+local rate tables instead of API Ninjas
- Prices: Extract prices from Toast __OO_STATE__ MenuItem objects
when visible HTML prices are missing
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Extract UUID folder path from URL instead of using getDirectoryFromPath
- Old logic was broken: listLast on path ending with / returned empty string
- This caused the code to go up one level too far
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Scan extracted ZIP for image files (jpg, png, gif, webp)
- Skip small files (<10KB, likely icons) and _files folder assets
- Send up to 3 images to Claude for business info extraction
- Merge extracted name, address, phone, hours, brandColor
- Only fills in fields not already found from HTML
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Extract directory and scan all .htm/.html files recursively
- Look for business name in title tags (skip generic titles)
- Extract street addresses with regex patterns
- Extract phone numbers
- Check __OO_STATE__ in other pages for Restaurant data
- Merge found info into toastBusiness (first found wins)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The __OO_STATE__ parsing was only extracting images, not the group names
as categories. Now extracts category names from menu.groups and maps
items to their proper categories.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Toast extraction was finding items but no h2.groupHeader categories,
leaving items ungrouped. showItemsStep() then rendered no checkboxes,
and confirmItems() filtered out all items (empty checkedIds set).
Now adds a default "Menu" category when items exist but categories is empty.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Added multiple fallback methods to extract business name:
1. Title tag with Toast-specific parsing
2. og:title and og:site_name meta tags
3. Header elements with restaurant/location classes
4. First h1 tag as last resort
Also added address and phone extraction from visible HTML.
Added summary logging of business info keys found.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
