A task assigned to a person and a task assigned to a coding agent are not the same object, and mixing them into one list makes both worse.

Human tasks carry judgement: decide the copy, choose between two designs, ring the client. They are done when a person says so.

Agent tasks carry an instruction precise enough to act on and a way to tell whether it worked. "Make the header better" is not one. "Header overlaps the hero below 380px — fix and keep the existing spacing above 640px" is.

Keeping them apart means the agent list can be run without a human reading each line first, and the human list is not padded with things nobody intended a person to do.

A task can move between lists. Something you thought needed judgement often turns out to be specifiable once you have looked at it.