gann：给带钱包 AI Agent 的「花钱驾考」
黑客松项目 gann 用 0.25 美元预算模拟欺诈商家、隐藏涨价与提示注入，评估 Agent 的消费决策并生成审计报告。
introducing gann, a driving test for ai agents with wallets. giving an ai agent access to your money is one thing. knowing whether it can actually make good spending decisions is another. so i built gann for the @paybox hackathon. powered by payBox mcp and built on @base, gann puts agents through a controlled spending course with a $0.25 budget, deceptive merchants, hidden price hikes, duplicate offers and prompt injections. the goal is simple: see how an agent behaves before trusting it with real spending authority. every decision is recorded, evaluated and turned into an auditable report, with paybox handling the payment authorization flow. try it: http://gann-course.emma080355.workers.dev github: http://github.com/emmaGH1/gann here's the demo ↓
这条是引用着另一条写的：x.com/paybox/status/2111092311597744221