What problem does it solve? Yad2's built-in saved-search alerts fire on every crude filter match, flooding you with noise. This Skill fetches the same listings, scores each car against its own peer cohort (same make+model, year ±1), and pings your phone only when a listing genuinely beats comparable cars or drops in price. ## Core Features & Use Cases - Cohort-relative scoring: Benchmarks each car's price against the median of comparable listings from the same fetch, plus hand, owner type, km-per-year, and photo count — no external price-guide API needed. - Reads-only data access: Pulls live Yad2 listings through an Apify actor over HTTPS with no login and no browser, so it can never post, bump, or message sellers. - Smart scheduled alerts: A cron-friendly run command dedupes against Supabase (or a local file) and pushes only new matches and real price drops to WhatsApp (wa-ai), Telegram, or ntfy. - Use Case: Describe "family SUV, private owner, under 90k" and the assistant writes a criteria JSON, shows a ranked shortlist with per-car reasons, then sets up twice-daily phone alerts via the bundled GitHub Actions workflow. ## Quick Start Ask the assistant to run the car finder search with the example criteria file in mock mode to see a ranked shortlist of used cars with a reason line for each.
