新的 AI 代理原语：为什么政策需要自己的语言（以及为什么 YAML 和 Rego 不足）

AI agents have evolved beyond mere experiments; they are now capable of writing code, transferring funds, and managing infrastructure. As their autonomy increases, a critical question arises: how can their actions be safely controlled? Many teams start with system prompts and YAML configurations, while some transition to generic policy engines like OPA/Rego or Cedar. However, these approaches were not specifically designed for agents. YAML lacks native concepts such as budgets, phases, and delegation, while Rego, though powerful, is too generic and treats 'deny' as an afterthought at runtime. This gap led to the development of FPL (Faramesh Policy Language), a domain-specific language tailored for AI agent governance.