Chaining Agents Needs a Spine

Most agent chains fail because they pass work around without narrowing the task, and the Ralph loop forces actual, reliable closure.