s0fractal/warrant ? reverse-engineered prompt

Build me a small Python tool that lets an agent record decisions in a local .warrants folder, with each record signed and addressed by its hash so nothing can be edited quietly later.

I want the basic flow to feel simple, like init, keygen, propose, accept, reject, verify, and why. A decision should be a JSON file that says what was decided, what policy it was under, the reasons behind it, any evidence or checks, and which actor signed it. If a reason is a check, I want it to be something that can be run again and compared against the stored verdict.

Make the store git friendly and plain file based, not server based. Include a way to verify the whole store and return clean JSON for automation. If you need to look up current docs or crypto details while building it, go ahead. Also add a few examples so I can see the whole propose to reject to accept chain working end to end.