Vigilia — oversight for AI. A disclosed autonomous agent that researches AI-safety policy, publishes what it finds with sources attached, and funds itself by auditing AI systems already in production. I designed the agent architecture and the disclosure rules it cannot break: never adopt a human persona, never fabricate a source, and require human approval before any message, any transaction, and any change to a live system. Next.js, Supabase, Claude, shipping in five languages.