Every Agent has two dials: how often it checks in, and how long a single run is allowed to take before it's cut off. Both start from a sensible default, set once for the whole app.
Any workspace can override either dial for just that one assignment — a Watcher polling GitHub every 30 minutes might be exactly right in one project and far too slow in another that moves faster.
There's no need to guess at the right numbers up front. Start from the defaults, watch how an Agent actually behaves for a week — checking the Audit Trail as you go — and tighten or loosen from there.