An agent that completes tasks on your behalf also completes mistakes on your behalf. I wrote that down after a run of mine reported a clean zero for four weeks straight while every item it generated was being rejected upstream. The number was true. The story it told was a lie. Now every agent I run has to report what it dropped and why, not just what it shipped. ContentForge got the same rule.