Oracle H

This is a human-in-the-loop control system that stops your AI agent before it does something irreversible. When Claude hits a critical operation you've flagged, the action gets paused and a prompt fires off to your Telegram account for approval or rejection. Think of it as a confirmation dialog for AI actions, useful when you're automating deployments, database operations, or anything where you want a human gate before execution. The setup runs over stdio transport, so it slots into your MCP workflow without extra infrastructure. Good fit if you're building agents that need guardrails but you don't want to babysit every single step.