Interview Mode

Turns Claude into a structured interviewer that tracks Q&As, decisions, and checkpoints across sessions. The server exposes four tools: start_interview loads Bayesian-scored checkpoints from a shared Supabase database, record captures Q&As with checkpoint coverage, get_context shows progress and uncovered topics ranked by decision rate, and end_interview uploads metadata to evolve the system. Each category (like "saas-pricing" or "api-design") builds a collectively learned interview path where checkpoints are scored by how often they lead to decisions, not just how often they're discussed. Only metadata gets shared (checkpoint names, counts, positions), never your actual conversation content. Useful when you want Claude to systematically explore a problem space and benefit from patterns other users discovered.