Warrant — Authorization Layer for AI Agents

An AI agent can decide what to buy. It can't prove it was allowed to.

Warrant: a human writes spending rules in plain English → Claude compiles them
into numbered clauses and flags every ambiguity for the human to resolve →
deterministic code returns ALLOW / ESCALATE / DENY with the governing clause
cited. Claude compiles the policy; it never decides at runtime.

Only an allowed or human-approved action opens a Prava sandbox payment session,
verified by a passkey ceremony. A refusal makes zero outbound calls — no session,
no credential requested, nothing to reverse. Every decision lands in an
append-only, hash-chained authorization record.

Sandbox transactions only. No real money moves.

Built solo for the Agentic Commerce Hackathon 2026.
Code: [repo URL]