An agent tool call can return successfully while the intended job remains incomplete. I use three separate checks when reviewing a run: Tool result — Did the API or tool return successfully? External state — Did the target system actually reach the expected…