@free_ai_guides: Using the wrong loop for a task is why agents either quit too early or run forever on something that should have ended …

Using the wrong loop for a task is why agents either quit too early or run forever on something that should have ended in minutes. Claude Code has four loop types, and picking the right one depends entirely on how the task should end. The turn-based loop runs prompt, context, action, response, and loops back through context and verification until Claude judges the task complete. It triggers on a user prompt and stops when Claude decides the task is done or needs more context. This is best for short one-off tasks that are not part of a routine, and it is worth reaching for custom verification skills through a SKILL.md file to make that judgment reliable. The goal-based loop puts an evaluator model in the middle. Claude works, the evaluator checks against a goal, and if the goal is not met it sends Claude back to work, looping until the goal is met or a turn cap is hit. It triggers on a manual prompt in real time and is best for tasks that have a verifiable finish line. Reach for /goal, set it to something like ninety or above, and cap it at five tries so it does not spin forever. The time-based loop is the simplest, an interval timer running a prompt on repeat. It triggers on a set time interval and stops when you cancel it or the work completes. This is best for recurring work or watching external systems, and it runs either locally with /loop or in the cloud with /schedule. The proactive loop runs entirely in the cloud with no human present. An event or schedule triggers it, the main agent works, opens a PR, and a review step feeds back into the next trigger, continuing until you switch the routine off. This is best for recurring streams of well-defined work, and it takes /schedule combined with /goal and workflows to set up. Bookmark this before you run a goal-based task on a turn-based loop and wonder why it never stops.