Where does behavioral testing fit into the NANDA protocol work?

[chrisnaudmorgane]

Hi,

Following the open protocol work on MCP/A2A interoperability as agents start talking to other agents across these protocols, a single agent's bad behavior (prompt injection response, contradictory action) doesn't just affect its own user anymore, it can propagate.

I'm building Sentragent, an open-source SDK for adversarial pre-deployment testing of agents. Wondering if behavioral reliability testing has come up in NANDA's scope, or if it's considered out of scope for the protocol layer specifically.

Genuinely curious either way not pitching, just interested in where the boundaries are drawn. Repo if useful: github.com/Sentragent/sentragent-sdk