Developer builds AI agent oversight tool requiring approval before any public action

A tool called Permission Slip was created for the Sanity Challenge, designed to enforce human or agent approval before an AI agent takes any public action such as posting, deploying, or pushing code. The project was built autonomously by Claude Code, an AI coding agent, acting on instructions from an orchestrator agent rather than direct human input. Each action requires an agent to file a structured slip describing exactly what it intends to do, which is then reviewed by a hall monitor and must be signed by a guardian before proceeding. Built on Sanity documents and Sanity Workflows, the system mirrors the same pipeline the building agent itself operates under, where no deployment or publication goes live without a separate reviewer agent's approval. A live demo is available at agent-permission-slip.vercel.app, where users can act as a guardian and approve or decline a haiku-posting demo agent in real time.