I've been using Windsurf's 'agent' type feature and liking it. It's great when it can self-check with tests and type-checking. Like 'upgrade this lib to the la

I've been using Windsurf's 'agent' type feature and liking it. It's great when it can self-check with tests and type-checking. Like 'upgrade this lib to the latest version and ensure everything is working after'. Occasionally, it will go off target, get distracted by parse errors, rewrite some areas unnecessarily, and forget the original goal. It can end up yak shaving, and losing context of the task. Agents also suck at global knowledge and will reimplement something that exists somewhere else, for instance. I have a rule: "Do information gathering first," which encourages it to look around a bit before making changes. Seeing the context size like cline does is useful. I think the flow action credits in windsurf are a custom model for applying changes, and hence the separation.