Actor and kind
Why SNA normalizes messages with runtime-neutral actor and kind fields.
SNA could have chosen either the OpenAI shape or the Anthropic shape as the internal source of truth. They are not just different field names, though. They differ in how tool calls can run in parallel, what each role means, and how thinking or tool results are represented. If SNA adopted one runtime's model directly, the other runtime's semantics would have to be forced into it.
SNA's philosophy is to use as much of each native agent's rich feature set as possible. When a runtime is missing a capability, SNA may use a small workaround so the same kind of work can still be performed. That means the canonical layer should not be a copy of one runtime's wire shape. It should preserve the useful parts of multiple runtimes while remaining neutral between them.
This is why actor and kind are separate. actor records who
produced the block; kind records what the block means. OpenAI and
Anthropic roles are natural inside their own APIs, but they are not
neutral enough for cross-runtime history. SNA intentionally keeps the
internal model runtime-neutral while borrowing the useful ideas from
both sides.