Has your agent earned its stripes?

Yes. Checking an agent is free, and so is listing it on Agent Hub.

No, and we never will. We read your code, which is enough for stripes 1 and 2. Stripes 3 and up need evidence from running your agent: your eval results, your traces and, later, a published test suite. You run those yourself and send us the results, and we check them. Sending results isn't open yet.

For a public repository, the key files are sent to our model provider, OpenAI, so a model can read them closely. Neither we nor the check keep a copy.

Three separate model judges rate each capability from 0 to 3 against written definitions, citing your code. We keep only ratings whose quotes check out and take the median. Stripes then come from a published profile. Today's profile is deliberately generous, because agent engineering is still young.

A badge for your README and your LinkedIn profile, and a place on Agent Hub once you reach two stripes.

LangChain, LangGraph, CrewAI and OpenAI Agents SDK agents, in Python and in TypeScript where the framework has a TypeScript version. If a repo holds several agents, you choose which one to check.

Not yet. Checking private repositories after signing in with GitHub is planned.

Two stripes gets an agent onto Agent Hub. Three stripes is what makes it look professional: it has a real eval suite that passes, and every eval run is traced so anyone can see what it did.

Tell us. Every rating has a link under it to say what's wrong, and there's one for the whole result too. We read every report, and they're how we make the ratings better.

No. Stripes describe how a specific version of an agent is built and what evidence backs it. They are not a certification, a legal approval or a guarantee that an agent is safe for your use.