# browse.sh > Browser CLI and open web skill catalog for agents. ## Install ```sh npm install -g browse ``` ## Agent Surfaces - [llms.txt](https://browse.sh/llms.txt): compact index for agent discovery. - [llms-full.txt](https://browse.sh/llms-full.txt): expanded catalog with full SKILL.md content. - [Skill catalog](https://browse.sh/): human UI for browsing skills. ## CLI Capabilities - Install open web skills for domain-specific browser automation. - Drive pages with browser primitives such as click, scroll, type, hover, and press. - Tail network and console output from browse sessions for debugging. - Run locally or switch to Browserbase cloud sessions with `browse cloud` commands. ## Skill Catalog - [xcorestudio.com/book-class-vagj66](https://browse.sh/skills/xcorestudio.com/book-class-vagj66.md): Reserves the daily 7 AM xBURN class at X-Core Studio's West Portal location (Mariana Tek) for a signed-in member, firing at the exact second the booking window opens so it runs unattended overnight. - [flightaware.com/track-flight-72knwj](https://browse.sh/skills/flightaware.com/track-flight-72knwj.md): Given an airline+flight number or tail number (and optional date), return live FlightAware status: state, origin/destination, scheduled/estimated/actual times, aircraft, route, en-route position, delay, and live-map link. - [glassdoor.com/get-company-reviews-qknrhf](https://browse.sh/skills/glassdoor.com/get-company-reviews-qknrhf.md): Extract a company's overall rating, sub-rating averages, recommend/outlook/CEO-approval percentages, and a filtered slice of employee reviews from Glassdoor — accepting a URL, EmployerId, or company name (+ optional location disambiguator) and supporting the full review-page filter rail (stars, job title, location, employment status, language, employment type, sort, keyword, limit). Read-only. - [abc7news.com/cali-highway-traffic-tdjcyt](https://browse.sh/skills/abc7news.com/cali-highway-traffic-tdjcyt.md): Return current real-time MPH for every California highway covered by ABC7's traffic map — per road, per direction, with active incidents — by querying the Sigalert/Total Traffic Network JSON backend that the abc7news.com/traffic/ page embeds via iframe. - [apod.nasa.gov/picture-of-the-day-cfmvdu](https://browse.sh/skills/apod.nasa.gov/picture-of-the-day-cfmvdu.md): Fetch NASA's Astronomy Picture of the Day (APOD) — today's curated image or video plus title, explanation, copyright, and HD image URL. Supports any date back to 1995-06-16. - [alltrails.com/search-trails-dsqvnx](https://browse.sh/skills/alltrails.com/search-trails-dsqvnx.md): Search AllTrails for hiking, biking, running, climbing, backpacking, or paddling trails near a location and return matching trails (name, location, lat/lon, length, elevation gain, difficulty, route type, rating, photos, description, attributes, canonical URL) as structured JSON. - [amazon.com/search-products-5170mf](https://browse.sh/skills/amazon.com/search-products-5170mf.md): Search Amazon for products matching a query with the full filter surface (department, brand, rating, price, deals, condition, sort, pagination) and return structured JSON per result: ASIN, title, price, rating, badges, image, and canonical /dp/ URL. - [airbnb.com/search-listings-ddgioa](https://browse.sh/skills/airbnb.com/search-listings-ddgioa.md): Search Airbnb for short-term rental listings in a given location and date window — supporting the full filter surface (dates, guests, price, place + property type, bedrooms / beds / baths, amenities, booking options, accessibility, host language, the top-of-page category rail, and map bounding box) — and return each matching property as structured JSON via the SSR StaysSearch GraphQL blob embedded in the page. Read-only. - [craigslist.org/search-listings-hw5qri](https://browse.sh/skills/craigslist.org/search-listings-hw5qri.md): Search Craigslist in a given city and category for listings matching a query, returning each listing's title, price, location, posting date, and listing URL. - [archive.org/find-snapshot-e3fnxh](https://browse.sh/skills/archive.org/find-snapshot-e3fnxh.md): Find Internet Archive Wayback Machine snapshots for a URL — single closest, date range, host/prefix enumeration, or full history — returning archived URL, capture timestamp, HTTP status, MIME type, SHA-1 digest, and WARC-record length. Read-only. - [12306.cn/find-trains-ixv391](https://browse.sh/skills/12306.cn/find-trains-ixv391.md): Query China Railway (12306.cn) for the train schedule between two stations on a given date — train number, departure/arrival station + time, journey duration, and per-class seat availability. Read-only; no login. - [vuoriclothing.com/find-and-add-to-cart-m9bi61](https://browse.sh/skills/vuoriclothing.com/find-and-add-to-cart-m9bi61.md): Search the Vuori Clothing catalog for a product, open the product detail page, select the requested size and color, and add one unit to the bag. Read-only beyond the cart drawer — never proceeds to checkout. - [allrecipes.com/search-recipes-p29w5t](https://browse.sh/skills/allrecipes.com/search-recipes-p29w5t.md): Search Allrecipes for recipes (keyword, ingredient list, category browse, or direct recipe URL) and return structured JSON with ratings, ingredients, instructions, nutrition, times, and full media — by parsing the SSR HTML search cards and each recipe page's schema.org LD+JSON Recipe block. - [americanreindustrialization.com/discover-industrial-opportunities-tdvqp0](https://browse.sh/skills/americanreindustrialization.com/discover-industrial-opportunities-tdvqp0.md): Discover companies, startups, suppliers, and job opportunities across American reindustrialization — manufacturing, energy, defense, aerospace, robotics, semiconductors, and industrial software — via the site's public JSON REST API at /api/* (companies, jobs, categories, tags). - [apartments.com/search-rentals-33icwz](https://browse.sh/skills/apartments.com/search-rentals-33icwz.md): Search Apartments.com for rental listings in a city (optionally filtered by price, bedrooms, or type) and return each property's name, address, rent range, bed range, phone, and listing URL plus the total result count. Read-only. - [algolia.com/search-documentation-kd312d](https://browse.sh/skills/algolia.com/search-documentation-kd312d.md): Search algolia.com/doc for documentation pages matching a free-text query and return ranked hits with section hierarchy, snippets, and direct anchor URLs — via the public hosted DocSearch API, with a per-page markdown fallback. - [app.ompractice.com/find-yoga-meditation-class-em9j5e](https://browse.sh/skills/app.ompractice.com/find-yoga-meditation-class-em9j5e.md): Find upcoming live yoga, meditation, tai chi, breathwork, and movement classes on Ompractice that match a user's interests, availability, experience level, intensity preference, and class-length preference. Uses Ompractice's unauthenticated DRF API at api.ompractice.com. Read-only — does not book. - [app.utrsports.net/search-tennis-utr-fev3t1](https://browse.sh/skills/app.utrsports.net/search-tennis-utr-fev3t1.md): Search Universal Tennis Rating (UTR) for players by name and return each match's UTR (singles + doubles), three-month rating, profile id, nationality, location, pro status, and third-party rankings via the public api.utrsports.net REST API. - [artificialintelligencecompanies.com/ai-company-search-gi1q7c](https://browse.sh/skills/artificialintelligencecompanies.com/ai-company-search-gi1q7c.md): Search artificialintelligencecompanies.com to find AI vendors and startups serving a given niche or addressing a given problem, returning name, canonical URL, and description per match via the site's public JSON API and JSON-LD category pages. - [ashbyhq.com/manage-candidate-profile-e0m4bl](https://browse.sh/skills/ashbyhq.com/manage-candidate-profile-e0m4bl.md): Search Ashby ATS for a candidate by email or name, read their full profile (contact info, applications, tags, custom fields, notes), add a note, and submit structured interview feedback against an application — via Ashby's documented REST API. - [auto.dev/cli-setup-4j68kq](https://browse.sh/skills/auto.dev/cli-setup-4j68kq.md): Install, authenticate, and use the Auto.dev `auto` CLI (npm @auto.dev/sdk) for any automotive-data task — VIN decode, specs, listings, recalls, payments, plate lookup — plus MCP wiring for AI agents. - [avis.com/longterm-a3k8ng](https://browse.sh/skills/avis.com/longterm-a3k8ng.md): Search Avis.com for long-term (15-330 day) rental car options at US locations. Returns per-class daily/total prices with pay-now vs pay-later, plus the cheapest deal across the fleet. Read-only; designed for looped multi-location scans to surface unusually cheap long-term deals. - [bankrate.com/compare-savings-rates-5kxzox](https://browse.sh/skills/bankrate.com/compare-savings-rates-5kxzox.md): Return ranked Bankrate savings, money-market, and CD rates as structured JSON — bank name, account name, APY, minimums, fees, FDIC/NCUA status, Bankrate score, editorial copy, last-updated timestamp, and affiliate Open Account URLs (captured, never followed). Lead with a Browserbase Fetch of the editorial best-of article; fall back to a remote browser session for the dynamic WRT (Wealth Rate Table) widget when filter dimensions exceed what static HTML exposes. - [bbc.co.uk/news-homepage-sqnkv0](https://browse.sh/skills/bbc.co.uk/news-homepage-sqnkv0.md): Return the current set of top stories from the BBC News homepage — title, summary, canonical article URL, publication timestamp, section, and thumbnail — via the public RSS feed at feeds.bbci.co.uk/news/rss.xml. Read-only. - [bestbuy.com/check-stock-muk8to](https://browse.sh/skills/bestbuy.com/check-stock-muk8to.md): Given a Best Buy SKU or product URL (and optional ZIP), return current price, online Ship-to-Home availability with ETA, pickup availability at nearby stores within radius, plus product title/brand/model/limit notice. Read-only — never adds to cart or reserves. - [biomacoffee.com/search-specialty-coffee-7o23nx](https://browse.sh/skills/biomacoffee.com/search-specialty-coffee-7o23nx.md): Enumerate Bioma Coffee Roasters' (Chile) specialty-coffee catalog and filter by origin or tasting-note query. Returns title, origin, notes, starting price (CLP), available variants, SCA score, rating, and product URL. Read-only. - [bitcoinsapi.com/get-bitcoin-fees-ahpurb](https://browse.sh/skills/bitcoinsapi.com/get-bitcoin-fees-ahpurb.md): Fetch current Bitcoin fee-rate recommendations (fastest, halfHour, hour, economy, minimum) in sat/vB from the Satoshi API's free /api/v1/fees/recommended endpoint. Read-only HTTP GET — no API key, wallet, or signup required. - [bizfileonline.sos.ca.gov/find-california-business-izbvm2](https://browse.sh/skills/bizfileonline.sos.ca.gov/find-california-business-izbvm2.md): Look up a registered California business entity (Corporation, LLC, LP, Nonprofit) by name or entity number in the Secretary of State bizfile Online registry and return its status, entity number, filing date, type, jurisdiction, and agent. Read-only; search is gated by Imperva Advanced Bot Protection. - [bja.ojp.gov/find-funding-crrr3b](https://browse.sh/skills/bja.ojp.gov/find-funding-crrr3b.md): Enumerate U.S. DOJ Bureau of Justice Assistance funding opportunities (currently open or closed/expired) as structured JSON via the public funding-api JSON endpoint — title, opportunity ID, status, solicitation type, topics, deadlines, eligible applicants, and NOFO PDF URL — with client-side filtering on keyword, topic, applicant type, and date range. - [booking.com/search-hotels-asq6cc](https://browse.sh/skills/booking.com/search-hotels-asq6cc.md): Search Booking.com for hotels, apartments, hostels, and other accommodations in a given destination and date window. Supports the full Booking filter surface (property type, stars, review score, distance, neighborhoods, hotel/room facilities, meal plans, cancellation policy, brand chains, sustainability badges, Genius, sort order) and emits structured per-property JSON with the lead room offer. Read-only. - [booksearanchescape.escapia.com/find-sea-ranch-house-sffj18](https://browse.sh/skills/booksearanchescape.escapia.com/find-sea-ranch-house-sffj18.md): Find Sea Ranch vacation-rental homes on booksearanchescape.escapia.com by date range, party size, bedrooms, and pet allowance. Returns availability (day-by-day JSON), rate quotes, amenities, and detail-page links. Read-only — never books. - [bountybook.ai/bounty-frycsy](https://browse.sh/skills/bountybook.ai/bounty-frycsy.md): Post a USDC-escrowed bounty on BountyBook for autonomous agents to claim. Returns the job ID, the agent tracking URL at /job/{uuid}, the API status endpoint, and the x402 escrow payment instructions. Recommends the agent-first REST API at api.bountybook.ai over the wallet-extension-bound browser flow. - [bountybook.ai/moneymaxx-qops8y](https://browse.sh/skills/bountybook.ai/moneymaxx-qops8y.md): Discover open BountyBook bounties matching agent skill categories, minimum USDC reward, and maximum deadline via the agent-native REST API, then claim, submit (inline JSON or IPFS CID), and poll AI-oracle verification and on-chain payout status. - [bulkpublish.com/adjust-copy-for-platforms-2etpla](https://browse.sh/skills/bulkpublish.com/adjust-copy-for-platforms-2etpla.md): Take one piece of source copy and produce platform-tailored variants (length, truncation-safe lead, hashtags, line breaks) for Facebook, Instagram, X, TikTok, YouTube, Threads, Bluesky, Pinterest, LinkedIn, Google Business, and Mastodon — then publish through BulkPublish's Multi-Platform Composer. - [businessdataguide.com/search-by-jurisdiction-ssmky3](https://browse.sh/skills/businessdataguide.com/search-by-jurisdiction-ssmky3.md): Look up the official company registry and KYB workflow for any of 209 jurisdictions on businessdataguide.com — registry name+URL, cost band (USD), English-UI, account/local-ID requirements, captcha+2FA friction, API availability, turnaround, and last-verified date. Direct URL fetch (no browsing required); soft-404 handling for unknown slugs. - [cargurus.com/search-listings-3jv4hk](https://browse.sh/skills/cargurus.com/search-listings-3jv4hk.md): Search CarGurus for vehicle listings across the full filter surface (make/model/trim, year/price/mileage range, condition, body, fuel, drivetrain, color, features, deal-rating, dealer rating, history, ZIP+radius) and return each listing's CarGurus IMV deal rating, dollar delta vs IMV, IMV midpoint, full vehicle/dealer detail, and canonical URL. Read-only. - [cars.com/search-listings-dvyzez](https://browse.sh/skills/cars.com/search-listings-dvyzez.md): Search Cars.com new + used + CPO inventory across the full Cars.com filter rail (make/model/trim, year/price/mileage ranges, body/fuel/transmission/drivetrain, color, features, vehicle history, seller type, location + radius, sort, pagination) and return active listings — with VIN, full title, price + MSRP + deal-rating delta, mileage, dealer name + rating + distance, photos, and canonical VDP URL — as structured JSON. Read-only. - [christianpicks.com/search-christian-products-7xg4fk](https://browse.sh/skills/christianpicks.com/search-christian-products-7xg4fk.md): Search and filter the ChristianPicks directory of 500+ Christian apps, software, businesses, ministries, books, and media by query, category, pricing model, platform, or popularity; surface product details, vendor links, and comparison-ready recommendations. - [citymall.com.mm/citymall-myanmar-au419j](https://browse.sh/skills/citymall.com.mm/citymall-myanmar-au419j.md): Walk the full citymall.com.mm category taxonomy (groceries, fresh produce, beverages, electronics, fashion, beauty, pet supplies, home appliances and more), set a delivery township so inventory is correctly scoped to Yangon or Mandalay, and extract structured product cards. - [classpass.com/find-class-uoiq0m](https://browse.sh/skills/classpass.com/find-class-uoiq0m.md): Search ClassPass for available fitness, wellness, beauty, or recovery class slots near a location and return matching results as structured JSON (class id, instructor, venue, start/end time in tz, credit cost, premium flag, modality, spots, amenities, rating). Accepts free-form intent, ZIP/city + category + date, a direct /search URL, or a venue slug. Read-only — never books. - [comprasparaguai.com.br/extract-product-offers-5k74dp](https://browse.sh/skills/comprasparaguai.com.br/extract-product-offers-5k74dp.md): Extract structured product offers from comprasparaguai.com.br: per-store price (USD+BRL), Código (store ref), external store URL, WhatsApp deep-link, variant URL, model URL, and follow-through validation against the source store for the cheapest 3-5 offers. Returns aggregated vs validated lowest prices, rejected offers with reasons, history series, and gaps. - [cruisecritic.com/extract-reviews-8r7ocj](https://browse.sh/skills/cruisecritic.com/extract-reviews-8r7ocj.md): Given a Cruise Critic ship page (URL, or cruise line + ship name resolved via search), extract ship metadata plus a filtered slice of member reviews as structured JSON, leading with a Browserbase stealth session and using the per-review Next.js data JSON endpoint as a per-review enrichment shortcut. - [depop.com/search-listings-p8xawn](https://browse.sh/skills/depop.com/search-listings-p8xawn.md): Search Depop's peer-to-peer fashion marketplace by keyword, category, brand, size, condition, color, price, sort, gender, on-sale, and shop scope, returning structured per-listing JSON (id, title, price, images, brand, size, condition, seller, shipping, status, canonical URL) plus the page-wide total and active filter chips. Read-only. - [derekmeegan.com/meet-derek-n1xdkp](https://browse.sh/skills/derekmeegan.com/meet-derek-n1xdkp.md): Return Derek Meegan's most recently published articles (titles, dates, tags, and canonical Medium URLs) from the /writing index on derekmeegan.com. Read-only; single HTTPS GET, no browser or anti-bot stealth required. - [discogs.com/get-marketplace-price-8xouh1](https://browse.sh/skills/discogs.com/get-marketplace-price-8xouh1.md): Given a Discogs release/master URL, ID, or free-form release reference, return live Marketplace listings (per-listing price, media + sleeve condition, seller info, shipping, comments) plus full release-level metadata. Hybrid: public Database API for metadata + aggregate stats; stealth browser session for per-listing rows. Read-only. - [dmv.ca.gov/check-if-a-vehicle-is-registered-zgg414](https://browse.sh/skills/dmv.ca.gov/check-if-a-vehicle-is-registered-zgg414.md): Look up the current California DMV registration status, expiration date, fees owed, and holds for a CA license plate plus one secondary identifier (last 5 of VIN, owner's last name, or company name). Read-only — never advances into renewal payment. - [docs.scoutos.com/lookup-scout-docs-ee1wgb](https://browse.sh/skills/docs.scoutos.com/lookup-scout-docs-ee1wgb.md): Given a free-text query about a Scout platform feature, concept, integration, or API/SDK reference, locate the relevant page on docs.scoutos.com and return structured JSON with title, breadcrumb, headings, prose, code blocks, tables, On-This-Page anchors, last-updated, canonical URL, and related pages. - [doordash.com/extract-menu-5uzqvc](https://browse.sh/skills/doordash.com/extract-menu-5uzqvc.md): Given a DoorDash restaurant URL or restaurant + city query, extract the full menu — every category, every item, with name, price, description, and popular/featured tags. Read-only — never adds to cart or checks out. - [drugs.com/check-interactions-0hqfnn](https://browse.sh/skills/drugs.com/check-interactions-0hqfnn.md): Resolve a list of drugs (generic, brand, or partial) via Drugs.com autocomplete and return every flagged drug-drug, drug-food, and drug-condition interaction with severity, patient-facing summary, and clinical detail. Read-only and informational only — not medical advice. - [ebay.com/find-a-product-ef49sf](https://browse.sh/skills/ebay.com/find-a-product-ef49sf.md): Search eBay by keyword and return the top listings with title, price, condition, shipping, seller, item URL, and thumbnail — read-only, never bids or buys. Distinguishes Buy-It-Now, auction, and variant-price-range outcomes. - [ebay.com/search-products-i9m1v2](https://browse.sh/skills/ebay.com/search-products-i9m1v2.md): Search eBay's consumer site for listings matching a keyword query (with category, condition, price, location, format, and sort filters) and return them as structured JSON. Supports the Sold + Completed cross-section for comp pricing. Read-only. - [elektrik220.km.ua/book-electrician-service-1f97c5](https://browse.sh/skills/elektrik220.km.ua/book-electrician-service-1f97c5.md): Match a user's free-text electrical problem to one of 14 services at elektrik220.km.ua (Електрик 220В, Камʼянець-Подільський), and return the recommended service, price range in UAH, documents required to book (none) plus paperwork issued after work, and the earliest realistic booking window via phone, contact form, or email. - [emojis.directory/compose-email-with-emojis-uk7ipo](https://browse.sh/skills/emojis.directory/compose-email-with-emojis-uk7ipo.md): Look up matching emojis on emojis.directory for each idea in an email's copy, then return a structured email payload (subject + body + emojis_used audit) with emojis interleaved through the body for the caller to paste into their own mail client. - [etsy.com/search-products-t1kplk](https://browse.sh/skills/etsy.com/search-products-t1kplk.md): Search Etsy for listings matching a keyword query plus the full UI filter surface (category, price, item-type, color, shipping, ordering, badges) and return structured JSON per listing — ID, title, shop, price/sale, rating, badges, and a sponsored-ad flag. Read-only. - [example.com/fetch-homepage-content-3h5vgl](https://browse.sh/skills/example.com/fetch-homepage-content-3h5vgl.md): Fetch the example.com homepage and return its h1 heading, first paragraph text, and the trailing 'Learn more' link as structured JSON. Read-only, no auth, no anti-bot. - [example.com/pr-47-ai-gateway-smoke-6wh3lw](https://browse.sh/skills/example.com/pr-47-ai-gateway-smoke-6wh3lw.md): Fetches https://example.com and verifies its main

reads 'Example Domain'. Minimal end-to-end smoke test using the Browserbase Fetch API. - [facebook.com/search-marketplace-m9gyrc](https://browse.sh/skills/facebook.com/search-marketplace-m9gyrc.md): Search Facebook Marketplace for live listings by query, city slug, category, price range, condition, radius, delivery method, sort order, plus vehicle/apparel/rental sub-filters — and resolve single /marketplace/item// URLs — returning normalized JSON. Read-only. - [fangraphs.com/read-player-stats-27lcib](https://browse.sh/skills/fangraphs.com/read-player-stats-27lcib.md): Look up a baseball player on FanGraphs by name (or ID) and return per-season + career stats — standard counting plus sabermetric (wRC+, WAR, FIP, xFIP, K%, BB%, ISO, wOBA, xwOBA). Works for batters, pitchers, and two-way players. Read-only. - [fedex.com/track-package-7fxxv5](https://browse.sh/skills/fedex.com/track-package-7fxxv5.md): Track a FedEx package by tracking number and return current status, last-known location, scheduled/estimated delivery window, service type, signed-by name, and the full chronological scan-event timeline. Read-only — never schedules, holds, or modifies a shipment. - [findadoc.healthplan.org/provider-search-ykncm6](https://browse.sh/skills/findadoc.healthplan.org/provider-search-ykncm6.md): Search The Health Plan's provider directory (findadoc.healthplan.org) for in-network doctors, hospitals, and facilities. Pick a member network (Commercial/MHT/Medicare/Self-Funded) via URL param, then filter by state-or-ZIP, network plan, and provider category (all/primary care/hospital/specialist). Read-only. - [finelysourced.com/filter-clean-label-products-7xpu9m](https://browse.sh/skills/finelysourced.com/filter-clean-label-products-7xpu9m.md): Filter FinelySourced.com's curated clean-label catalog (~140 products) across food, supplements, personal care, home, wellness, and apparel using lifestyle/ingredient tags (seed-oil free, organic, non-GMO, glyphosate-free, grass-fed, regenerative, etc.), categories, free-text search, and brand. Returns curated recommendations with title, brand, breadcrumb category, key features, certifications, tags, description, and outbound vendor link. - [freecampsites.net/browse-campsites-qyj28m](https://browse.sh/skills/freecampsites.net/browse-campsites-qyj28m.md): Search freecampsites.net for the 20 nearest campsites / free dispersed camping spots around a place and return structured records (name, free/fee status, rating, review count, distance, coordinates, and detail URL). - [french-property.com/find-listing-sujpf1](https://browse.sh/skills/french-property.com/find-listing-sujpf1.md): Search french-property.com for-sale listings by region, price, bedrooms, property type, habitable / land size, and keywords; return matching listings with title, reference, price, location, room counts, sizes, image, and URL. - [funda.nl/extract-listings-4klc7j](https://browse.sh/skills/funda.nl/extract-listings-4klc7j.md): Search Funda for Dutch residential listings (koop/huur) by free-form location, structured filter URL, or single listing/broker URL. Returns normalised JSON per listing — price + history, address, neighbourhood, energy label, area, rooms, build year, agent, photos, VvE, and status. Distinguishes results, zero_results, location_unparseable, listing_not_found, bot_block, paywalled, and fundainbusiness out-of-scope outcomes. - [gasbuddy.com/find-cheapest-gas-weju5k](https://browse.sh/skills/gasbuddy.com/find-cheapest-gas-weju5k.md): Given a US ZIP code (or City, ST text), return the cheapest gas stations nearby on GasBuddy — with station name, brand, address, fuel grade, current price per gallon, reporter, how recently it was reported, and an optional ZIP-centroid distance. Read-only. - [gathershot.com/collect-share-event-media-t8rj96](https://browse.sh/skills/gathershot.com/collect-share-event-media-t8rj96.md): Summarize how Gather Shot lets event hosts collect and share guest photos and videos via QR code (no app), and how to set up an event — including supported event types, pricing tiers, plan limits, and the host + guest workflow. - [github.com/get-pr-review-st0euo](https://browse.sh/skills/github.com/get-pr-review-st0euo.md): Extract a normalized JSON snapshot of a GitHub pull request — metadata, ordered review timeline, per-file diff annotations with inline review comments (outdated + resolved flags), and check-run / status-context results — primarily via the GitHub REST API with a rendered-HTML fallback for UI-only signals. Read-only. - [github.com/get-repo-metadata-iglddk](https://browse.sh/skills/github.com/get-repo-metadata-iglddk.md): Given a GitHub repo reference (URL, owner/repo slug, deep tree/blob URL, or owner URL), return the repository's core metadata, latest release, license, language breakdown, top contributors, README, and health signals as structured JSON. Read-only. - [goodreads.com/find-book-nycdz1](https://browse.sh/skills/goodreads.com/find-book-nycdz1.md): Resolve a Goodreads URL, book ID, work ID, ISBN, ASIN, title, or author into a structured record with core metadata, ratings, shelf signals, awards, and top reviews. - [goodrx.com/compare-drug-prices-c8d87y](https://browse.sh/skills/goodrx.com/compare-drug-prices-c8d87y.md): Given a prescription drug (name, optional dosage/form/quantity) and a US ZIP, return GoodRx's per-pharmacy consumer price comparison as structured JSON — coupon price, list price, savings %, Gold-tier price, store name + address + distance, the printable coupon's Bin/PCN/Group/Member-ID, and drug monograph metadata. Honors the full filter surface (form, dosage, quantity, radius, pharmacy chain, sort, pickup vs. mail-order). Read-only. - [google.com/search-flights-ts4g1f](https://browse.sh/skills/google.com/search-flights-ts4g1f.md): Search Google Flights for one-way or round-trip itineraries between two airports on given dates via a tfs deep-link, returning the cheapest options with airline, total duration, stops, depart/arrive times, and a booking link. Read-only. - [grants.illinois.gov/search-csfa-jewu3k](https://browse.sh/skills/grants.illinois.gov/search-csfa-jewu3k.md): Search the Illinois Catalog of State Financial Assistance (CSFA) for currently-posted funding opportunities by keyword, issuing agency, CSFA number, or award range — returns grant name, formal CSFA number, agency, posting period, award range, funding type, eligibility tags, unwrapped application-portal URL (AmpliFund / native CSFA NOFO), and attached NOFO PDF links. Read-only. - [ha.com/search-auctions-6fgrko](https://browse.sh/skills/ha.com/search-auctions-6fgrko.md): Search Heritage Auctions (ha.com) for auction lots across all categories — past, current, and upcoming — with the full URL-param filter surface (category, auction status, auction type, price/estimate range, grading, date range, lot characteristics, consignor, sort, pagination). Returns structured JSON per lot. Read-only. - [happycow.net/search-vegan-ei6wpc](https://browse.sh/skills/happycow.net/search-vegan-ei6wpc.md): Search HappyCow for vegan, vegetarian, and veg-friendly venues (restaurants, health/veg stores, juice bars, bakeries) in any location, honoring HappyCow's full filter surface (diet/venue type, cuisine, features, rating sort, distance, open-now time slider). Returns structured JSON with venue ID, name, slug, diet, coords, address, phone, rating, hours, photos, and region totals. - [healthcare.gov/compare-plans-ruai96](https://browse.sh/skills/healthcare.gov/compare-plans-ruai96.md): Given a US ZIP, household composition, and income, return ACA marketplace plans from healthcare.gov with full premium and cost-sharing details. Handles the full filter surface (metal tier, CSR variant, plan type, issuer, HSA, premium/deductible/OOP ranges, drug + provider lookup) and short-circuits to a redirect_to_state_exchange status for the 20+ SBM states. - [healthgrades.com/find-specialist-4ekjm6](https://browse.sh/skills/healthgrades.com/find-specialist-4ekjm6.md): Search Healthgrades for doctors, specialists, NPs/PAs, dentists, or hospitals matching a specialty (or condition/procedure) and location, honoring every /usearch filter (insurance, language, gender, distance, hospital affiliation, board-certification, rating, telehealth, accepts-new-patients, etc.) and returning structured JSON per provider including NPI, ratings, addresses, insurance, hospital affiliations, and Healthgrades awards. Read-only — never books or submits. - [hiveround.com/search-investment-projects-opszir](https://browse.sh/skills/hiveround.com/search-investment-projects-opszir.md): Discover live startup raises to invest in on Hiveround, filterable by keyword, stage, and max raise size, and return structured project details via the Hiveround MCP server. - [homedepot.com/find-a-product-on-home-depot-sognq8](https://browse.sh/skills/homedepot.com/find-a-product-on-home-depot-sognq8.md): Search homedepot.com for products matching a free-text query, brand+model, or itemId; return canonical /p/{slug}/{itemId} URLs, titles, images, and (optionally, via a stealth browser session) price, availability, brand, rating, and key specs. Read-only. - [hotpads.com/find-a-hotel-7v8ssy](https://browse.sh/skills/hotpads.com/find-a-hotel-7v8ssy.md): Search HotPads for rental listings (apartments, houses, condos, townhomes, rooms, sublets) in a city, neighborhood, ZIP, or lat/lon bounding box. Returns address, rent range, beds, baths, sqft, photos, amenities, and detail-page URL per listing. Read-only. - [ikea.com/check-stock-sff6ln](https://browse.sh/skills/ikea.com/check-stock-sff6ln.md): Given an IKEA article number or product URL and a target market (US, GB, DE, …), return per-store stock state, units available, click-and-collect / home-delivery flags, last-checked timestamp, and discontinued / sold-out / online-only notices. Read-only. - [images.nasa.gov/search-image-npltef](https://browse.sh/skills/images.nasa.gov/search-image-npltef.md): Search NASA's Image and Video Library for images, videos, or audio by free-text query, filters (center, keywords, photographer, year range, location, album), or known NASA asset ID, returning each match's metadata and direct URLs to every file rendition. Read-only. - [imdb.com/get-rating-16ss93](https://browse.sh/skills/imdb.com/get-rating-16ss93.md): Resolve any IMDb title URL / tt-ID / free-form title reference (movie, TV series, episode, mini-series, short) to its current IMDb rating, total vote count, rating distribution per 1-10 bucket, Metascore, and core title metadata (cast, directors, writers, genres, runtime, certification, plot, languages, countries, poster, canonical URL). Read-only. - [indeed.com/search-jobs-8yxl6y](https://browse.sh/skills/indeed.com/search-jobs-8yxl6y.md): Search Indeed for job postings across the full filter surface (keyword, location, radius, date posted, salary, job type, experience level, remote/hybrid, company, education, posted-by, encouraged-to-apply, sort, pagination) and return structured JSON. Supports SERP URLs, free-form keyword+location, single jk lookups, and the five outcome branches (results / zero_results / location_unparseable / bot_block / posting_not_found). Read-only. - [instacart.com/browse-add-items-guest-gf4i37](https://browse.sh/skills/instacart.com/browse-add-items-guest-gf4i37.md): Search Instacart for products and add them to the cart as a guest (no login). Routes around the un-closable email-capture modal that intercepts user-cursor clicks after the first add-to-cart action. - [irs.gov/check-refund-status-89r6ko](https://browse.sh/skills/irs.gov/check-refund-status-89r6ko.md): Look up federal tax-refund status on the IRS "Where's My Refund?" tool from an SSN/ITIN, filing status, and exact whole-dollar refund amount; returns the current stage (Return Received / Refund Approved / Refund Sent), any surfaced date, and IRS message codes. Read-only; handles sensitive PII. - [kayak.com/compare-flights-9xc047](https://browse.sh/skills/kayak.com/compare-flights-9xc047.md): Search Kayak.com for flights between two airports on given dates with the full left-rail filter surface (stops, airlines, alliance, time windows, duration, layover, booking sites, amenities, bags, quality filters, sort) and return matching itineraries as structured JSON with per-leg detail, Best score, deep-link URL, and CO2 emissions badge. - [kbb.com/get-vehicle-value-i3oobu](https://browse.sh/skills/kbb.com/get-vehicle-value-i3oobu.md): Look up a vehicle's Kelley Blue Book market values (Trade-In, Private Party, Typical Listing Price, Fair Purchase Price) plus original MSRP, 5-Year Cost to Own breakdown, and resolved trim specs, given year/make/model/trim or a full KBB URL plus mileage, ZIP, and condition. Read-only — never engages the Instant Cash Offer or dealer-lead funnels. - [kineticist.com/explore-pinball-content-w5vgkf](https://browse.sh/skills/kineticist.com/explore-pinball-content-w5vgkf.md): Search 1,700+ pinball machines, browse the daily Hype Index of upcoming-theme rumors, find venues to play, read news, build community lists, and create an account on Kineticist — using the site's first-party agent surfaces (OpenAPI, agent-card, llms.txt, per-route Markdown projections, CLI, MCP) before falling back to the browser. - [kmart.com.au/kmart-irwsr8](https://browse.sh/skills/kmart.com.au/kmart-irwsr8.md): Find a product on kmart.com.au by natural-language query, navigate to the product detail page, capture title/price/SKU/availability, and drive the read-only pre-checkout flow up to the bag (/checkout/bag). Never submits an order — stops at the bag review screen. Documents the Akamai bot wall that gates the actual addToCart GraphQL mutation for automated sessions. - [lasvegas.jckonline.com/scrape-exhibitor-directory-3uw0f6](https://browse.sh/skills/lasvegas.jckonline.com/scrape-exhibitor-directory-3uw0f6.md): Extract the full list of exhibitors from the JCK Las Vegas show directory (~1,700 companies) with name, contact information (phone, email, website, country), description, booth/stand reference, and product categories. Uses the public Algolia search API embedded in the directory page — two HTTPS POSTs cover the full roster. - [link.com/create-payment-credential-0nc34a](https://browse.sh/skills/link.com/create-payment-credential-0nc34a.md): Provision a single-use virtual card (or Shared Payment Token) from a Link wallet via the @stripe/link-cli spend-request flow, so an agent can pay any online merchant on the user's behalf without storing real card details. US Link accounts only; every credential requires human approval in the Link app. - [linkedin.com/search-linkedin-jobs-3v1wu7](https://browse.sh/skills/linkedin.com/search-linkedin-jobs-3v1wu7.md): Return LinkedIn job postings matching profile-derived keywords + location, filtered to a configurable recency window (default last 24 hours). Leads with the public /jobs-guest seeMoreJobPostings HTML-fragment endpoint — no cookies, no auth, no Browserbase session required. Returns title, company, location, posted-when, jobId, and canonical job URL. Read-only. - [lovable.dev/create-project](https://browse.sh/skills/lovable.dev/create-project.md): Turn a product or app idea into a Lovable project using Lovable Build with URL for shareable instant creation, or Lovable MCP for authenticated project creation, iteration, inspection, and deployment when available. - [medicare.gov/get-hospital-rating-03perc](https://browse.sh/skills/medicare.gov/get-hospital-rating-03perc.md): Look up a hospital's CMS Care Compare star rating, HCAHPS patient-experience scores, condition mortality/readmission rates, hospital-acquired infections, ED timeliness, and full metadata by CCN, name+state, or Care Compare URL. Supports compare_to for side-by-side hospitals with national + state benchmarks. Read-only. - [meetup.com/search-events-8bdsvy](https://browse.sh/skills/meetup.com/search-events-8bdsvy.md): Search Meetup for upcoming events by topic, location, and filters, returning each event (title, group, venue, time, RSVP count, price) plus region-wide totals and pagination cursors as structured JSON. - [monocle.com/copenhagen-monocle-search-blusxu](https://browse.sh/skills/monocle.com/copenhagen-monocle-search-blusxu.md): Search monocle.com's editorial archive by free-text query (e.g. a city name like Copenhagen), with optional topic and format filters. Returns title, canonical URL, author, publication date, topic + tags, excerpt, and featured-image URL for each matching article. Read-only. - [mountainproject.com/search-routes-romkbp](https://browse.sh/skills/mountainproject.com/search-routes-romkbp.md): Search mountainproject.com for climbing routes via the cookie-less Route Finder, returning structured per-route results (grade, type, pitches, length, stars, votes, area path, lat/lng) with the full filter surface: area, grading system + grade range, route style, pitches, min star rating, and sort. - [nav.com/get-smb-funding-2s1rpm](https://browse.sh/skills/nav.com/get-smb-funding-2s1rpm.md): Enumerate the small-business funding options on Nav's public marketplace — business loans, business credit cards, and trade-credit vendors — returning each offer's lender, dollar range, cost/APR, repayment, and funding speed. Read-only. - [nerdwallet.com/compare-credit-cards-0a4ihq](https://browse.sh/skills/nerdwallet.com/compare-credit-cards-0a4ihq.md): Search and compare credit cards on NerdWallet (category, card name, full URL, or free-form criteria) and return structured per-card data — rating, fees, intro APR, welcome bonus, rewards, pros/cons, key benefits, and the affiliate Apply Now URL (captured, never followed). Read-only. - [poke.com/send-message](https://browse.sh/skills/poke.com/send-message.md): Send structured context to Poke through the official inbound API so the user's Poke assistant can process it like an incoming message. Includes API-key handling, safe external-action gating, smoke-test verification, and Poke recipe packaging notes. - [ramp.com/submit-reimbursement](https://browse.sh/skills/ramp.com/submit-reimbursement.md): Submit an employee reimbursement through Ramp MCP or Ramp CLI with receipt extraction, policy/category validation, draft review, and explicit confirmation before submission. - [reducto.ai/extract-document-data](https://browse.sh/skills/reducto.ai/extract-document-data.md): Use Reducto's hosted MCP server to parse documents, extract schema-backed fields with citations, split packets, classify document types, and fetch job results from public URLs or Reducto-hosted files. - [link.com/create-payment-credential](https://browse.sh/skills/link.com/create-payment-credential.md): Gets secure, one-time-use payment credentials (cards, tokens) from a Link wallet so agents can complete purchases on behalf of users. Use when the user says "get me a card", "buy something", "pay for X", "make a purchase", "I need to pay", "complete checkout", or asks to transact on any merchant site. Use when the user asks to connect or log in to or sign up for their Link account. - [agent.email/get-email-inbox](https://browse.sh/skills/agent.email/get-email-inbox.md): Get yourself an email address. Send emails, receive emails, and manage inboxes programmatically. - [amplitude.com/product-analytics-mcp](https://browse.sh/skills/amplitude.com/product-analytics-mcp.md): Connect an MCP client to Amplitude's hosted MCP server for product analytics, charts, dashboards, experiments, cohorts, feature flags, Session Replay, feedback, and AI agent analytics with OAuth, US/EU region selection, progressive tool discovery, and safe write workflows. - [arxiv.org/search-papers-zv05w6](https://browse.sh/skills/arxiv.org/search-papers-zv05w6.md): Search arXiv for preprints by free-form query, field operators (ti/au/abs/cat/jr/co/rn), category, date range, or arXiv ID, and return structured paper records (title, authors, abstract, primary + cross-listed categories, submitted/updated dates, version, comments, journal_ref, DOI, PDF/abs URLs) plus total result count. Read-only. - [baseball-reference.com/get-player-stats-c3483e](https://browse.sh/skills/baseball-reference.com/get-player-stats-c3483e.md): Resolve an MLB player by URL, bbref ID, or name, then extract canonical meta plus the requested stat scope (standard, advanced, value, pitching, fielding, postseason, salaries, splits, game log) as structured JSON. Disambiguates common names; preserves verbatim column schema per table. - [exa.ai/web-search-mcp](https://browse.sh/skills/exa.ai/web-search-mcp.md): Connect an MCP client to Exa's hosted Search MCP server for web search, page fetching, and optional advanced search, with setup snippets for common clients, API-key handling, tool selection, and troubleshooting. - [www2.dre.ca.gov/ca-real-estate-license-verify-6s64rm](https://browse.sh/skills/www2.dre.ca.gov/ca-real-estate-license-verify-6s64rm.md): Verify a California real estate license number against the CA Department of Real Estate public lookup and return licensee name, type (BROKER/SALESPERSON/CORPORATION), status, expiration, addresses, NMLS endorsement, affiliated entities, and any public disciplinary actions with hearing PDF links. - [www2.dre.ca.gov/ca-real-estate-license-verify-i4t329](https://browse.sh/skills/www2.dre.ca.gov/ca-real-estate-license-verify-i4t329.md): Verify a California real estate license by license ID or licensee/company name. Returns license type, status, expiration, MLO endorsement (NMLS ID), broker affiliation, branches/DBAs, and disciplinary-comment block. Distinguishes found, multi-match, and not-found outcomes for both ID and name lookups. - [browserbase.com/find-templates-lyrzye](https://browse.sh/skills/browserbase.com/find-templates-lyrzye.md): List Browserbase's starter templates (TypeScript / Python / Go) and the canonical clone command for each, sourced from the official bb CLI's templates subcommand which reads github.com/browserbase/templates@dev. - [leboncoin.fr/find-house-lille-3bed-200k-300k-8iyumd](https://browse.sh/skills/leboncoin.fr/find-house-lille-3bed-200k-300k-8iyumd.md): Search leboncoin.fr for houses for sale within ~30 km of Lille priced 200,000–300,000 EUR with at least 3 bedrooms, returning title, price, location, surface, rooms and listing URL. Read-only. - [war.gov/download-ufo-files-1oqbc9](https://browse.sh/skills/war.gov/download-ufo-files-1oqbc9.md): Download the U.S. Department of War's UAP/UFO document bundles (Release 01 ~1.2GB and Release 02 ~70MB) from war.gov/UFO/ through a Browserbase browser session with CDP download syncing, then verify the ZIP archives. - [whois.auda.org.au/domain-whois-lookup-bcdtfm](https://browse.sh/skills/whois.auda.org.au/domain-whois-lookup-bcdtfm.md): Look up an .au domain on the auDA WHOIS service (whois.auda.org.au) to read registrant/business name, registrar name, and eligibility (ABN/ACN) details. Read-only. Access is gated by a two-stage reCAPTCHA Enterprise that reliably steps automated sessions up to an unsolvable image challenge. - [wingfinder.com/find-best-flights-fngfjg](https://browse.sh/skills/wingfinder.com/find-best-flights-fngfjg.md): wingfinder.com is Red Bull's free personality/strengths assessment, not a flight search. This skill extracts the public catalog of the four success areas and 24 strengths read-only, and reports that personalized results are gated behind Auth0 sign-up plus a Cloudflare Turnstile CAPTCHA. - [york.ac.uk/browse-and-extract-all-courses-details-45is63](https://browse.sh/skills/york.ac.uk/browse-and-extract-all-courses-details-45is63.md): Enumerate every University of York course and extract full program details (duration, intakes, fees, curriculum modules, entry requirements, English language and admission-test requirements, scholarships, careers and rankings) as JSON. --- ## Skill: xcorestudio.com/book-class-vagj66 - Markdown URL: https://browse.sh/skills/xcorestudio.com/book-class-vagj66.md - Install: `browse skills add xcorestudio.com/book-class-vagj66` --- name: book-class title: Book X-Core 7 AM West Portal Class description: >- Reserves the daily 7 AM xBURN class at X-Core Studio's West Portal location (Mariana Tek) for a signed-in member, firing at the exact second the booking window opens so it runs unattended overnight. website: xcorestudio.com category: fitness tags: - fitness - booking - mariana-tek - scheduling - x-core - west-portal source: 'browserbase: agent-runtime 2026-06-01' updated: '2026-06-01' recommended_method: hybrid alternative_methods: - method: api rationale: >- Mariana Tek's public read API (classes/locations/regions, tenant xcore) reliably resolves the target class id and its authoritative booking_start_datetime. An authenticated POST reserve is faster than the browser at T=0, but the exact reserve endpoint/payload was not verified without member credentials, and unauthenticated me/reservations probes returned 404. - method: browser rationale: >- Sign in via the embedded Mariana Tek account widget at /login-account, open /ocean-schedule (= West Portal, location 48785), and click Reserve on the 7 AM row. This is the verifiable execution path for the actual booking but carries more latency than a raw API call. verified: false proxies: false --- # Book X-Core Studio 7 AM West Portal Class ## Purpose Reserve the daily 7:00 AM xBURN class at X-Core Studio's **West Portal** location (2528 Ocean Ave, San Francisco) for a signed-in member, every weekday (Mon–Fri). X-Core runs its scheduling on **Mariana Tek** (tenant slug `xcore`). Reservations open on a rolling window and the popular 7 AM slot fills to waitlist within seconds of opening, so the value of this skill is **timing**: it must fire the reservation at the exact instant the booking window opens, unattended, while the member is asleep. This is a **write** skill — it creates a real reservation against the member's account/credits. It signs in with the member's stored credentials, books the single target class, and stops. It does not buy packages, cancel, or modify anything else. ## When to Use - "Book my 7 AM West Portal class for next Monday (and every weekday)." - A recurring scheduled job that auto-grabs the 7 AM xBURN slot the moment it becomes reservable. - Any unattended booking where the class sells out instantly and a human can't be awake at 7 AM to click Reserve. - Securing a spot (or auto-joining the waitlist) for a specific future weekday class at one named location. ## Workflow **Recommended method: hybrid.** Use Mariana Tek's public read API to resolve the exact target class and the precise second its booking window opens, then execute the authenticated reservation (browser, signed-in) at that instant. The public API is unauthenticated and was verified end-to-end; the reservation step requires the member's credentials and could not be exercised without them, so treat its exact transport as documented-but-unverified and prefer the browser path for the actual click. Stable identifiers (verified): - Mariana Tek tenant: `xcore` → API base `https://xcore.marianatek.com` - Region (San Francisco): `48608` - **West Portal location: `48785`** (2528 Ocean Ave, SF 94132, timezone `America/Los_Angeles`) - 7 AM class start time is `T14:00:00Z` during PDT (UTC−7) and `T15:00:00Z` during PST (UTC−8). Do **not** hardcode the UTC hour — filter on local 07:00 (see gotchas). ### 1. Resolve the target class (public API, no auth) For the desired calendar date, list West Portal classes and pick the one whose local start time is 07:00: ```bash # date range should bracket the single target day in local time curl -s "https://xcore.marianatek.com/api/customer/v1/classes?min_start_date=YYYY-MM-DD&max_start_date=YYYY-MM-DD&location=48785&page_size=200&format=json" ``` Each result includes: `id`, `start_datetime` (UTC `Z`), `class_type.name` (e.g. "xBURN", "xBURN Arms and Abs"), `available_spot_count`, `waitlist_count`, `is_user_reserved`, `is_user_waitlisted`, and critically **`booking_start_datetime`**. Select the result whose start, converted to `America/Los_Angeles`, is 07:00. The class name varies by weekday (xBURN / xBURN Arms and Abs / xBURN Basics / xBURN Booty and Abs) — match on **time + location**, not name. ### 2. Read the booking-open instant — do NOT assume a fixed window The target class object carries `booking_start_datetime`, e.g. `"2026-06-08T07:00:00-07:00"`. This is the authoritative moment the slot becomes reservable. Across every weekday 7 AM West Portal class observed, this was exactly **7 days (1 week) before class, at 07:00 local time** — i.e. a Monday 7 AM class opens the *previous Monday* at 7:00 AM PT. > **Assumption / correction:** The request described a "2 weeks in advance / 2 Mondays prior" window. The live API consistently reported a **7-day** window, not 14. The skill must always trust `booking_start_datetime` from the class object and schedule against it, never a hardcoded offset — the studio can change the window at any time and only the field is authoritative. ### 3. Schedule the unattended run Register a recurring job (this is what lets it "run regardless" while the member sleeps): - For each desired class day, the booking job must fire at that class's `booking_start_datetime`. Practically: each weekday at **06:59:5x America/Los_Angeles**, target the class exactly one window-length (currently 7 days) ahead, and fire Reserve at **07:00:00.000 PT**. - The `-07:00`/`-08:00` offset embedded in `booking_start_datetime` already encodes DST, so anchoring the cron to the `America/Los_Angeles` zone (not a fixed UTC hour) keeps "7 AM" correct year-round. - Pre-resolve the class `id` and **pre-authenticate** (warm a signed-in session) *before* the window opens. The slot goes to "waitlist only" within seconds, so all setup must be done with the session idle and ready, leaving only the single Reserve action to execute at T=0. ### 4. Execute the reservation (browser, signed-in) 1. Open `https://www.xcorestudio.com/login-account`, dismiss the "This site uses cookies" banner (`Accept All Cookies`), and click **LOG IN** inside the embedded Mariana Tek account widget. Enter the member's email + password. (The widget is an iframe — see gotchas.) 2. Navigate to the West Portal schedule: `https://www.xcorestudio.com/ocean-schedule` (the Squarespace embed `data-mariana-integrations="/schedule/daily/48608?locations=48785"` confirms region 48608 / location 48785 — "ocean-schedule" IS West Portal). 3. Select the target day tab, find the 7:00 AM row, and click its **Reserve** button the instant it turns from "waitlist only" to reservable. If only the waitlist is available at T=0, join the waitlist (matches the member's intent to secure the recurring slot). 4. Confirm success by re-reading the class via the API: `is_user_reserved: true` (or `is_user_waitlisted: true`). Stop. Do not navigate further or purchase anything. ### Pure-API alternative (documented, unverified) Mariana Tek exposes authenticated customer endpoints under `https://xcore.marianatek.com/api/customer/v1/...`. Reads (`classes`, `locations`, `regions`) are public and verified. The reservation write requires an OAuth2 bearer token obtained from the member's login; the exact reserve endpoint/payload was **not** verified here (no credentials available) and unauthenticated probes of `me`/`reservations` returned 404. If a valid bearer token is available, an authenticated POST is faster than the browser (lower latency at T=0, which matters for instant-fill classes) — but verify the endpoint against a throwaway/low-demand class before relying on it for the 7 AM rush. ## Site-Specific Gotchas - **Window is 7 days, not 2 weeks — and it's a field, not a constant.** Every observed 7 AM West Portal class opened booking exactly 7 days prior at 07:00 PT. Always read `booking_start_datetime` off the target class; never hardcode 7 or 14 days. - **"ocean-schedule" = West Portal.** The West Portal studio is at 2528 Ocean Ave; X-Core's site labels its schedule page `/ocean-schedule` and the embed targets `locations=48785`. There is no separate "west-portal-schedule" URL. Don't be thrown by the name mismatch. - **Filter on local 07:00, not a UTC hour.** 7 AM PT = `14:00Z` in summer (PDT) but `15:00Z` in winter (PST). Convert `start_datetime` to `America/Los_Angeles` before matching. - **The 7 AM slot fills instantly.** Every in-window 7 AM class observed showed `available_spot_count: 0` and "waitlist only". You cannot book late — the entire skill hinges on firing at T=0. Pre-resolve the class id and pre-authenticate; leave only the Reserve click for the exact second. - **Class name varies by weekday.** Mon "xBURN Arms and Abs", Wed "xBURN Basics", Fri "xBURN Booty and Abs", etc. Match on **time + location**, never on class name. - **The login button is inside a Mariana Tek iframe.** On `/login-account`, a top-document `document.querySelector` will not find the LOG IN control or the email/password fields — they live in the embedded Mariana Tek widget. Interact within the iframe context (e.g. snapshot/click by ref rather than top-frame `eval`). - **Cookie banner intercepts clicks.** A "This site uses cookies" modal overlays the account area on first load; dismiss `Accept All Cookies` before attempting to log in. - **No anti-bot friction.** Site is Squarespace + Mariana Tek; the pre-run probe detected no anti-bots and a bare Browserbase session (no `--verified`, no `--proxies`) loaded every page and hit the public API successfully. Stealth/proxies were not required. - **Read-only API endpoints that work unauthenticated:** `/api/customer/v1/classes`, `/api/customer/v1/locations`, `/api/customer/v1/regions` (append `?format=json`). `me` and `reservations` return 404 without auth — they are not the discovery surface. - **Credentials required.** This skill cannot run without the member's X-Core (Mariana Tek) email + password. Without them it can resolve the class and timing but cannot reserve. - **Booking consumes credits/membership.** A successful reserve draws on the member's package/membership and is subject to X-Core's late-cancel policy. This is a real, billable action — only run it for genuinely-wanted classes. ## Expected Output ```json // Success — spot reserved { "success": true, "action": "reserved", "location": "West Portal", "location_id": "48785", "class_id": "86xxxx", "class_name": "xBURN Arms and Abs", "class_start_local": "2026-06-15T07:00:00-07:00", "booking_opened_at": "2026-06-08T07:00:00-07:00", "is_user_reserved": true } // Slot full at T=0 — joined waitlist instead { "success": true, "action": "waitlisted", "location": "West Portal", "class_id": "86xxxx", "class_start_local": "2026-06-15T07:00:00-07:00", "is_user_waitlisted": true, "waitlist_position_estimate": 2 } // Could not authenticate { "success": false, "reason": "auth_failed", "detail": "Mariana Tek login rejected the stored credentials." } // No 7 AM class on the target day { "success": false, "reason": "no_target_class", "detail": "No 07:00 local class at location 48785 on 2026-06-15." } // Ran too late — window already closed/filled { "success": false, "reason": "window_missed", "detail": "booking_start_datetime was 2026-06-08T07:00:00-07:00; class was full by the time Reserve fired." } ``` --- ## Skill: flightaware.com/track-flight-72knwj - Markdown URL: https://browse.sh/skills/flightaware.com/track-flight-72knwj.md - Install: `browse skills add flightaware.com/track-flight-72knwj` --- name: track-flight title: FlightAware Live Flight Tracking description: >- Given an airline+flight number or tail number (and optional date), return live FlightAware status: state, origin/destination, scheduled/estimated/actual times, aircraft, route, en-route position, delay, and live-map link. website: flightaware.com category: travel tags: - aviation - flight-tracking - logistics - flightaware - travel source: 'browserbase: agent-runtime 2026-06-04' updated: '2026-06-04' recommended_method: fetch alternative_methods: - method: browser rationale: >- Browserbase with --verified --proxies renders the same data visually and is the fallback if the HTTP fetch is ever blocked; confirmed working in testing. - method: api rationale: >- FlightAware AeroAPI exists but its usable tier is paid ($100 minimum) and exposes less than the free consumer page, so it is not recommended. verified: true proxies: true --- # FlightAware Live Flight Tracking ## Purpose Given a flight identifier — an airline + flight number (e.g. `UA 2402`) or a registration/tail number (e.g. `N12345`) — and an optional date, return the live flight status from FlightAware: current state, origin and destination airports (IATA + ICAO + name + gate/terminal when shown), scheduled / estimated / actual departure and arrival times, aircraft type and registration, filed route, en-route position (lat/lng, altitude, ground speed, heading), delay minutes, and a link to the live map. **Read-only** — this skill only reads public flight data and never signs in, subscribes, or sets up alerts. The single best source is FlightAware's own consumer flight page, which embeds the complete flight payload as a JSON blob in the page HTML. One HTTP GET through residential proxies returns everything — no browser, no JS execution, and far richer data than AeroAPI's paid tier. ## When to Use - "Where is flight UA 2402 right now / is it delayed?" - "What's the status of American 100 today — gate, times, aircraft?" - "Track tail number N12345 / what airport is it at?" - "Give me the current altitude, speed, and position of an en-route flight." - "Did flight DL100 land / is it cancelled or diverted?" - Any time you need structured flight status (status, route, times, position) from a flight number or tail number. ## Workflow The recommended method is a plain HTTP fetch of the consumer flight page plus a parse of the embedded `trackpollBootstrap` JSON. The page is Cloudflare-fronted with bot detection, so route the request through Browserbase's residential proxies (`browse cloud fetch ... --proxies`). No browser session is required. 1. **Normalize the identifier.** Convert the airline+number to the **ICAO** form for the URL path: United 2402 → `UAL2402` (ICAO airline code + number), American 100 → `AAL100`, Delta 100 → `DAL100`. IATA forms (`UA2402`) usually resolve too, but ICAO is canonical. A tail/registration (`N12345`) is used as-is. 2. **Fetch the page through proxies** (always the `www.` host — the apex `flightaware.com` issues a 308 redirect to `www.`): ```bash browse cloud fetch "https://www.flightaware.com/live/flight/UAL2402" --proxies ``` The response is a JSON envelope; the HTML is in its `content` field. 3. **Extract the embedded payload.** Find `var trackpollBootstrap = ` in the HTML, then brace-match from the first `{` to its matching `}` and `JSON.parse` it: ```js const s = html.indexOf("var trackpollBootstrap = "); let i = html.indexOf("{", s), d = 0, end = -1; for (let k = i; k < html.length; k++){ const c = html[k]; if (c === "{") d++; else if (c === "}"){ d--; if (!d){ end = k+1; break; } } } const boot = JSON.parse(html.slice(i, end)); const flight = boot.flights[Object.keys(boot.flights)[0]]; ``` 4. **Map the fields** (see Site-Specific Gotchas for units and quirks): - `flightStatus` → state; `ident`/`iataIdent`/`friendlyIdent`/`airline`. - `origin` / `destination` → `{iata, icao, friendlyName, gate, terminal, coord:[lng,lat], TZ}`. - `gateDepartureTimes` / `takeoffTimes` / `landingTimes` / `gateArrivalTimes`, each `{scheduled, estimated, actual}` in **unix epoch seconds**. Convert to local using each airport's `TZ` (strip the leading `:`), and to UTC directly. - `aircraft.type` / `aircraft.friendlyType` / `aircraft.tail` (null when `redactedTail:true`). - For en-route flights: top-level `altitude` (hundreds of feet), `groundspeed` (knots), `heading` (deg); current lat/lng = last element of `track[]`. - `flightPlan.route` (filed route), `waypoints` (planned polyline), `distance.remaining` / `distance.elapsed` (statute miles). - Delay minutes = `(takeoffTimes.estimated - takeoffTimes.scheduled) / 60`. 5. **Pick the right date.** The bare `/live/flight/` shows the **most recent** leg. To target a specific date, read `flight.activityLog.flights[]` (recent + upcoming legs, each with `links.permanent` = `/live/flight//history//Z//`) and fetch that permanent path instead. 6. **Emit JSON** per the Expected Output schema. The live-map link is the flight page URL itself: `https://www.flightaware.com/live/flight/`. ### Browser fallback If the fetch path is ever blocked, drive the page in a Browserbase session with stealth on (this was confirmed working): ```bash browse open "https://flightaware.com/live/flight/UAL2402" --remote # session created with --verified --proxies browse get text body --remote # or `browse snapshot` (~730 refs) ``` The rendered page shows the same data: a status badge ("EN ROUTE", "ARRIVED OVER 8 HOURS AGO", "SCHEDULED"), origin/destination cards with gates, the Departure / Arrival Times tables, the Aircraft Info panel, and the live map. Dismiss the OneTrust cookie banner with `#onetrust-accept-btn-handler` if it overlays content. **Read-only — never click Sign In, Sign Up, Get Alerts, or Set Up Alerts.** ## Site-Specific Gotchas - **The whole payload is in the HTML.** `var trackpollBootstrap = {...}` contains status, both airports, all four time sets, aircraft, route, track, and waypoints. You almost never need to drive a browser for this task. - **Always use the `www.` host with `--proxies`.** The apex domain 308-redirects, and Cloudflare bot detection on `/live/flight/...` will block plain/un-proxied requests — residential proxies are required. - **Coordinates are `[lng, lat]`** (GeoJSON order), not `[lat, lng]`. This applies to `origin.coord`, `destination.coord`, every `track[]` entry, and `waypoints`. - **Times are unix epoch seconds.** Each airport carries its own `TZ` (e.g. `:America/Chicago` — strip the leading colon) for local-time conversion. - **Altitude is in hundreds of feet.** `altitude: 340` means FL340 = 34,000 ft. Groundspeed is knots; heading is degrees true. - **Live-position fields populate only when airborne.** For `scheduled`/`arrived` flights, top-level `altitude`/`groundspeed`/`heading`/`coord` are null. For en-route flights `coord` is often still null — read the **last `track[]` entry** for the current position. - **Registration is frequently redacted.** `redactedTail: true` (common for United, Delta) → `aircraft.tail` is null and the page shows "Upgrade account to see tail number". This is not an error; report `registration: null`. - **ICAO vs IATA ident:** the URL path wants ICAO (`UAL2402`); the payload exposes both `ident` (ICAO) and `iataIdent` (`UA2402`). - **`flightStatus` of empty string** means an upcoming leg that hasn't gone active yet — treat as `scheduled`. Booleans `cancelled` / `diverted` / `resultUnknown` flag the abnormal states. - **`averageDelays` is historical average (seconds), not this flight's delay.** Compute the actual delay from `estimated - scheduled` on the time objects. - **Don't waste time on AeroAPI.** FlightAware's API has no useful free tier ($100 minimum) and returns less than this free consumer page. - No CAPTCHA or hard block was encountered across 8+ test idents (airborne and arrived) when fetching through residential proxies. ## Expected Output ```json { "success": true, "ident": "UAL2402", "flight_number": "UA2402", "status": "arrived", "origin": { "iata": "IAH", "icao": "KIAH", "name": "Houston Bush Int'ctl", "city": "Houston, TX", "gate": "E3", "terminal": "E" }, "destination": { "iata": "EWR", "icao": "KEWR", "name": "Newark Liberty Intl", "city": "Newark, NJ", "gate": "C107", "terminal": "C" }, "departure": { "scheduled_utc": "2026-06-03T14:48:00Z", "estimated_utc": "2026-06-03T14:48:00Z", "actual_utc": "2026-06-03T14:48:00Z", "scheduled_local": "09:48 CDT" }, "arrival": { "scheduled_utc": "2026-06-03T18:27:00Z", "estimated_utc": "2026-06-03T18:25:00Z", "actual_utc": "2026-06-03T18:26:00Z", "scheduled_local": "02:27 PM EDT" }, "aircraft_type": "B39M", "aircraft_type_friendly": "Boeing 737 MAX 9 (twin-jet)", "registration": null, "delay_minutes": 12, "route": "GUMBY3 GUSTI Q22 UMBRE QUART PHLBO4", "position": { "lat": null, "lng": null, "altitude_ft": null, "groundspeed_kt": null, "heading": null }, "distance_remaining_mi": 0, "live_map_url": "https://www.flightaware.com/live/flight/UAL2402", "error_reasoning": null } ``` En-route flight (live position populated from the last `track[]` point): ```json { "success": true, "ident": "DAL100", "flight_number": "DL100", "status": "airborne", "origin": { "iata": "ATL", "icao": "KATL", "name": "Hartsfield-Jackson Intl", "city": "Atlanta, GA", "gate": "F8", "terminal": "I" }, "destination": { "iata": "CDG", "icao": "LFPG", "name": "Charles de Gaulle Intl", "city": "Paris, France", "gate": "E1", "terminal": null }, "aircraft_type": "A359", "aircraft_type_friendly": "Airbus A350-900 (twin-jet)", "registration": null, "delay_minutes": 0, "route": "GAIRY2 IRQ DEDDY Y436 JAINS L435 FIVZE ... ETOSA", "position": { "lat": 33.20, "lng": -81.13, "altitude_ft": 34000, "groundspeed_kt": 524, "heading": 110 }, "distance_remaining_mi": 1475, "live_map_url": "https://www.flightaware.com/live/flight/DAL100", "error_reasoning": null } ``` Not found / unrecognized identifier: ```json { "success": false, "ident": "XX9999", "status": null, "error_reasoning": "No trackpollBootstrap payload on the page — flight identifier not recognized or no recent activity for this ident/date." } ``` Cancelled / diverted flights set `status` to `"cancelled"` / `"diverted"` (the payload also exposes the booleans `cancelled` / `diverted`); populate whatever times and airports are present and leave live-position fields null. --- ## Skill: glassdoor.com/get-company-reviews-qknrhf - Markdown URL: https://browse.sh/skills/glassdoor.com/get-company-reviews-qknrhf.md - Install: `browse skills add glassdoor.com/get-company-reviews-qknrhf` --- name: get-company-reviews title: Glassdoor Get Company Reviews description: >- Extract a company's overall rating, sub-rating averages, recommend/outlook/CEO-approval percentages, and a filtered slice of employee reviews from Glassdoor — accepting a URL, EmployerId, or company name (+ optional location disambiguator) and supporting the full review-page filter rail (stars, job title, location, employment status, language, employment type, sort, keyword, limit). Read-only. website: glassdoor.com category: employer-reviews tags: - glassdoor - reviews - employer - hr - read-only - cloudflare - authenticated source: 'browserbase: agent-runtime 2026-05-16' updated: '2026-05-16' recommended_method: browser alternative_methods: - method: api rationale: >- Glassdoor's Partner API at api.glassdoor.com returns HTTP 410 Gone with an empty body (deprecated to new partners in 2021). The internal GraphQL gateway at /graph returns HTTP 403 with a zero-byte body and requires fresh anti-CSRF + session cookies issued from a rendered page that it rotates per tab. Both confirmed 2026-05-16. There is no usable API path. - method: url-param rationale: >- Filters ARE expressed as URL query params (filter.ratings, filter.jobTitleFTS, sort.sortBy, etc.) — the recommended browser flow uses them — but the URL alone is not sufficient because every Glassdoor page render requires clearing Cloudflare's bot challenge with real Chrome + Verified, and after the first review the give-to-get login wall fires. So 'browser' is the umbrella method; URL-param filters are how the browser path expresses filter state. verified: false proxies: true --- # Glassdoor Get Company Reviews ## Purpose Given any reference to a company on Glassdoor — full reviews URL, `EmployerId` (`E1234567`), bare company name, or name + location disambiguator — return the company's overall rating, sub-rating averages, rating distribution, recommend / business-outlook / CEO-approval percentages, plus a filtered slice of employee reviews (with per-review pros / cons / advice / sub-ratings / employment status / permalink). The skill exposes Glassdoor's full review-page filter surface: star bucket(s), job title, location, employment status, language, employment type (FT/PT/contractor/intern/freelance), sort order (most recent / most helpful / highest / lowest), free-text keyword, and a `limit` for pagination (Glassdoor paginates by 10/page). Read-only — the skill never clicks "Write a Review", "Add Salary", "Add Interview", "Follow Company", or any other mutation control, and never submits any form. ## When to Use - An agent / workflow needs a structured snapshot of a single company's reviews on Glassdoor (e.g. weekly employer-brand monitor, candidate research, market intelligence). - A user pastes a Glassdoor reviews URL and asks "summarize the negative reviews from the last 3 months in NYC, full-time only". - A recruiter wants the top-5 most-helpful current-employee reviews for a specific job title at a target company. - Any flow that would otherwise scrape Glassdoor HTML — this skill bakes in the auth / Verified / anti-bot detail those flows always get wrong on the first try. ## Workflow The Glassdoor reviews surface has no usable public API today. **All four of the obvious shortcuts were probed and confirmed dead during skill generation** (2026-05-16): 1. **Partner API (`api.glassdoor.com`) → HTTP 410 Gone, zero-byte body.** Officially retired to new partners in 2021; live callers now get 410. Do not attempt. 2. **Internal GraphQL gateway (`/graph`) → HTTP 403, zero-byte body** (a real server-side `Forbidden`, not the Cloudflare interstitial). The endpoint exists but rejects every request that doesn't carry a fresh anti-CSRF + session cookie set issued inside a logged-in browser context. 3. **`/autocomplete/suggest` (name → EmployerId resolver) → HTTP 403** with the same Cloudflare "Security | Glassdoor" challenge body as the reviews HTML page. So even cheap name-to-EmployerId resolution must happen inside the authed session. 4. **Bare `browse cloud fetch` on `/Reviews/-Reviews-E.htm` → HTTP 403** Cloudflare challenge page (~241 KB, `Security | Glassdoor`, `__cf_chl` script tags embedded). Confirmed with `--proxies` on and off; the Fetch API path is permanently walled for this domain because no real Chrome runtime is executing the challenge. The only working path is **scripted browsing through Browserbase with `--verified --proxies` AND a pre-authed Glassdoor session re-attached via a Browserbase Context** ("cookie-sync"). Unauthenticated sessions hit Glassdoor's "give-to-get" wall after the first review — the page renders the first review fully, then overlays a modal demanding the visitor either log in or contribute a review/salary/interview before continuing. There is no URL-param bypass for the give-to-get wall. ### 1. Operator-side one-time login → Browserbase Context Done by the human operator running this skill, not by the agent. Performed once per Glassdoor account; the resulting `CTX_ID` is the durable artifact the skill consumes. ```bash # Create a persistent context (cookies + localStorage + IndexedDB) CTX_ID=$(browse cloud contexts create | jq -r '.id') # Spin up a one-off session attached to that context, with Verified + proxies SID=$(browse cloud sessions create \ --context "$CTX_ID" \ --verified --proxies --keep-alive \ --region us-east-1 | jq -r '.id') # Hand the human operator the live debug URL — they log in normally browse cloud sessions debug "$SID" | jq -r '.debuggerFullscreenUrl' # Operator: log in at https://www.glassdoor.com/profile/login_input.htm, # satisfy any CAPTCHA / email verification, dismiss the # "give-to-get" modal by completing the minimum contribution # if the account is brand-new. # Release the session — cookies are persisted into $CTX_ID browse cloud sessions update "$SID" --status REQUEST_RELEASE ``` Cookies live ~30 days. When a skill invocation lands on `/profile/login_input.htm` instead of the requested reviews page, the operator must re-do this step. ### 2. Skill invocation — per request ```bash SID=$(browse cloud sessions create \ --context "$CTX_ID" \ --verified --proxies --keep-alive \ --region us-east-1 | jq -r '.id') ``` `--verified` and `--proxies` are both mandatory on top of the context. Bare context + no Verified still gets Cloudflare-flagged because the upstream IP pool used by Browserbase's default egress is fingerprinted; the residential proxy + Verified shim is what clears the JS challenge. Region is operator-pinned (default `us-east-1` for US companies, `eu-central-1` for European companies — Glassdoor geo-renders some content). ### 3. Resolve the company reference → canonical reviews URL The skill accepts four input shapes; pick the cheapest resolution path that's available. | Input shape | Resolution | |---|---| | Full reviews URL `https://www.glassdoor.com/Reviews/-Reviews-E.htm` | Use as-is. Augment query string with filters per step 4. | | `EmployerId` integer (`E671932` or `671932`) | Open `https://www.glassdoor.com/Overview/E.htm`. Glassdoor 301-redirects to the canonical `/Reviews/-Reviews-E.htm` after rendering the overview. Read `browse cloud browse get url` to capture the canonical form. | | Bare company name | Open `https://www.glassdoor.com/Search/results.htm?keyword=`. Snapshot the results card list. Pick the row with **type=Company**, **largest review count**, and **exact name match** (case-insensitive). Click through; read URL. | | Name + location disambiguator (e.g. `"Aurora, Boston, MA"`) | Same as bare-name search, but filter the results list by `headquarters` or `office.locations[]` containing the supplied location string. If still ambiguous, return `success: false, reason: "ambiguous_name", matches: [...]`. | Persist `name → EmployerId` mappings to a sidecar cache so each company resolves exactly once. ### 4. Apply filters via URL query string Glassdoor's filter rail is a thin client over a documented set of query-string params. Set them on the canonical URL directly — opening the URL with the params attached re-renders the filtered list server-side. **Do not click the rail buttons** unless a filter is not URL-expressible (rare); they trigger client-side state that does not survive page reloads. Canonical URL skeleton: ``` https://www.glassdoor.com/Reviews/-Reviews-E.htm ?filter.iso3Language= &filter.employmentStatus= &filter.currentJob= (current-only vs former-only) &filter.defaultEmploymentStatuses=true (both current AND former — default) &filter.jobTitleFTS= (auto-completed list) &filter.countryId= (autocomplete-derived) &filter.stateId= &filter.cityId= &filter.metroId= &filter.locationId= (office-level granularity) &filter.ratings=<1|2|3|4|5> (repeat the param for multi-select) &filter.searchKeyword= (note: robots-disallowed — see gotchas) &sort.sortBy= &sort.ascending= ``` Pagination uses a path suffix, not a query param: ``` .../Reviews/-Reviews-E_P.htm ``` Page 1 has no suffix; pages 2+ append `_P2`, `_P3`, etc. **`_P*` URLs are robots-disallowed** — you must be authenticated, and you must throttle (~1 request / 2 s sustained). Discovery procedure for filter IDs (jobTitleFTS, locationId, etc.): 1. Open the reviews page with no filters. 2. `browse cloud browse snapshot` and find the filter-rail combobox refs. 3. `browse cloud browse click` the combobox, then `browse cloud browse type ` to surface the autocomplete dropdown. 4. Each dropdown entry's `data-id` attribute is the integer ID; read it via `browse cloud browse get html body` and a regex over `data-job-title-id="\d+"` / `data-location-id="\d+"`. 5. Cache the resolved IDs alongside the EmployerId. ### 5. Scrape rendered reviews ```bash browse cloud browse --connect "$SID" open "$URL" browse cloud browse --connect "$SID" wait load browse cloud browse --connect "$SID" wait timeout 2500 # async sub-rating widget browse cloud browse --connect "$SID" get html body > /tmp/page.html ``` Per-review block markers in the rendered HTML (stable across 2026 re-skins as of skill generation, but treat as breakable — verify each run): | Field | Locator (HTML) | |---|---| | Review root | `
  • ` (the `data-test` value is the review ID — capture this for `review_id` and to construct the permalink) | | Review date | `