Maybe that's why the agent chose 30 mins instead of a higher frequency poll.

I think this might be some of the variability we see with our current automations approach. I wonder if there's a more deterministic, configurable automation we can setup: • A pre-built script that takes in key-value config values from the user ◦ repo to monitor: acmeco/foo ◦ review style: [balanced, roasted] ◦ posting mode: [comment, no comment] • Script polls GH APIs every X minutes checking for new PRs etc ◦ Once it sees a new PR to review, it fires off a Conversation ◦ Else, no conversation run, saving you LLM tokens

@hundreds-twilight-32833 and I were discussing this exact scenario today where it seems like Automations are still very non-deterministic in how they get setup. Not anyone's fault, mostly a byproduct of how we've designed the system so far. But I think it could make for some unpredictable user experiences.