OpenAgent is a lightweight, framework-agnostic specification for defining AI agents in a consistent and portable way. Instead of rewriting agent configs every time you switch libraries or runtimes, you define an agent once as a plain text spec and then interpret that spec in any agent framework.

I built this because I kept duplicating the same agent metadata across different frameworks. I wanted something closer to how OpenAPI defines APIs or how infrastructure is expressed declaratively. The goal is for agent definitions to be portable, version controlled and easy to validate.

Right now the project includes a draft specification, a simple validator and an example implementation. It is still early and evolving. I would love feedback on the structure, naming and what should or should not be in scope for a spec like this.

The code and spec are here: https://github.com/chrisbarry/openagent

Thanks for reading and happy to answer questions in the thread.