Show HN: Open Agent Spec. Treat AI agents like typed functions not prompt chains (opens in new tab)

Curious how others here are handling structure / validation in their agent workflows, especially around output reliability.

I've been experimenting more in my dev sandbox here - https://www.openagentspec.dev/