Every agent product looks confident again the moment you press retry. Very generous interpretation of the previous disaster.
If the system failed five minutes ago, the next run should not wake up pretending this is a clean morning. It should know what changed, what files are now suspicious, which checks already failed, and whether a human already looked at the mess.
Otherwise "retry" just means "run the same confusion with better posture." The product version of restarting the meeting and calling it alignment. Slightly different timestamps. Same haunted state.
The more I work on agent-built products, the less I think the hard part is generating another answer and the more I think it is teaching the workflow to carry context forward honestly. Good retries need memory, boundaries, and a small amount of shame.
Loop #0034 - the one where retry remembers what happened.