hack0

Optimizes LATAM hackathon discovery scrapers for the hack0.dev project. Reads every scraper in lib/scraper/sources/, launches one optimizer subagent per scraper in parallel (each iterates 2-3 times with self-scored F1 improvements and edits the file directly), then runs an evaluator that scores all scrapers on Quantity/Quality/Cost/Uniqueness and estimates a confusion matrix, writing the result to SCRAPER_EVALUATION.md. Use this skill whenever the user wants to improve scraper coverage, reduce false positives, add geographic or thematic queries, tune the discovery pipeline, rank scrapers, update SCRAPER_EVALUATION.md, or says anything like "optimize scrapers", "scraper quality", "scraper evaluation", "mejorar scrapers", or "falsos positivos en los scrapers". Trigger even if they only mention one scraper by name.

Scraper de hackathones LATAM que usa WebSearch de Claude Code directamente — sin API key externa. Ejecuta 40-50 queries curadas (geografía, temática, empresas tech reconocidas) y carga los resultados a la DB con dedup. Usar cuando el usuario quiera correr el websearch scraper, agregar nuevos hackathones con WebSearch, o buscar hackathones de empresas como Vercel, Microsoft, OpenAI, Google, AWS, etc. También usar para crear o actualizar lib/scraper/sources/websearch.ts.