[topcmm]

Really cool project! Bridging the gap between software agents and the physical world is super hard. For API failure modes, it would be very helpful to clearly distinguish between "human timed out/didn't respond" and "human actively rejected the task." Keep up the good work!

[tetubrah]

Thanks — great callout. We’re implementing this split explicitly:

- `human_timeout` (accepted but no response before SLA / no one accepted in time) - `human_rejected` (a human actively declined)

We’ll expose it as structured failure codes in the API response (not just a generic failure), so agent-side retry/escalation logic can branch correctly. Appreciate the push — this is exactly the reliability boundary we care about.