I’ve been playing a lot with durable agents, can’t say that I know what the final thing should look like but this could be a part of it. For now I have celld in front of sandbox, if I get this right: agent -> dasp -> celld -> sandbox
Do you envision an A2A adapter for Jido, with DASP handling durable execution underneath? I’d be curious how an uncertain outcome would map to an A2A task state.
The `uncertain` mapping isn't clean. DASP treats it as terminal: the host can't prove whether effects occurred, so it must not retry. A2A's `TASK_STATE_UNSPECIFIED` is indeterminate, but not terminal. The adapter must preserve the DASP outcome in metadata and map it conservatively, likely to `FAILED`.
My question wasn't about transport - it's more do I have to repeatedly poll for events, or can I make a request that effectively subscribes me to a stream of events.
I’ve been playing a lot with durable agents, can’t say that I know what the final thing should look like but this could be a part of it. For now I have celld in front of sandbox, if I get this right: agent -> dasp -> celld -> sandbox
Do you envision an A2A adapter for Jido, with DASP handling durable execution underneath? I’d be curious how an uncertain outcome would map to an A2A task state.
Yes. A2A and ACP adapters would sit above DASP.
The `uncertain` mapping isn't clean. DASP treats it as terminal: the host can't prove whether effects occurred, so it must not retry. A2A's `TASK_STATE_UNSPECIFIED` is indeterminate, but not terminal. The adapter must preserve the DASP outcome in metadata and map it conservatively, likely to `FAILED`.
Is this purely polling, or am I missing how to subscribe to events?
The protocol is transport agnostic - but I'm primarily building it to work over Web Sockets
My question wasn't about transport - it's more do I have to repeatedly poll for events, or can I make a request that effectively subscribes me to a stream of events.
[dead]
Really great stuff! Cant wait to try it out in my harness!