Excellent practical write-up! Making agent workflows resilient to crashes and restarts is a crucial part of moving beyond demos into production. I like the focus on durability and recoverability rather than assuming every execution will go as planned.