for this case, is that prompt + polling like OpenClaw/Hermes cron, or do you filter events on your side before waking the agent?
Mostly asking to understand the cost model for watches like this. Does the user pay for repeated agent runs, or only when a filtered match is delivered?