Show HN: Zehrava Gate – an control plane that sits between AI agents and prod

Hey HN, I built Zehrava Gate because I kept having the same problem: AI agents that could read anything worked great, but the moment I let one write - to a CRM, send an email, charge a card - I had no consistent answer to "who authorized that?" What it is: A self-hosted policy engine + approval queue for agent actions. The agent submits an intent before executing. Gate evaluates a YAML policy (deterministically - no LLM), scores risk, and either auto-approves, holds for human review, or blocks. Every decision is logged with a signed execution token. The agent can't skip it.