s0fractal/trinity ? reverse-engineered prompt

Build me a TypeScript project that lets an AI agent produce verifiable receipts for every important action it takes, so a person can later see who did what, under whose authority, and verify it from public data.

I want the core flow to include signed decisions, witness based approval for trust root changes, and a way to rederive the final proof without trusting the machine that created it. It should feel like a small command line driven coordination system, with a clean SDK for other apps to use, plus clear docs for how to run checks and verify the live court result.

Please make it practical and easy to inspect, with tests, generated status views, and a simple way to confirm that the public bytes match the claimed receipt. If you need current library docs or best practices, look them up online.