Work marked done and work actually done diverge quietly, and the person best placed to notice is the one who wrote the instructions and has already moved on.
Checking against the instruction, not against taste
General quality feedback on delivered work is hard to act on and easy to dispute. Feedback against the acceptance criteria that were written before the work started is neither. Either the criterion is met or it is not, and the evidence is attached.
This also surfaces a second problem worth knowing about: tasks whose acceptance criteria were too vague to check. That is a finding about the instruction rather than the delivery, and it gets reported as one.
Ground rules count too
Beyond the task's own criteria, delivered work is checked against the standing rules: the ones the client set, and the ones the team holds itself to. Those are rarely restated in an individual task and are exactly the ones that get missed under time pressure.
It reviews, it does not fix
The output is feedback on the task with evidence. Correcting the work stays with whoever did it, which is the only version of this that improves anything.
Runs well with
Skills compound. These are the ones we usually install alongside it.