emilia-protocol

Why we included this project

Letting an autonomous agent touch money, code, or infrastructure means accepting that it might do more than it was asked. EMILIA is built around that problem: at a protected boundary it checks that the owner's authority actually covers the exact action before admitting it, then issues a portable receipt of what was allowed and later observed. It works alongside the Model Context Protocol and existing identity, approval, and policy systems instead of replacing them, so it fits into stacks that already have governance. The protocol itself is implementation-independent and published as an IETF draft, which matters when you want evidence that holds up to independent verification. Teams that need a verifiable consequence boundary around agentic workloads, rather than another agent framework, will find this worth a look.