A recent paper from your field. Find its biggest problem, then try to fix it.
You know the field. The agent does the reading; you do the judging.
What’s the main claim here?
What’s the biggest methodological problem?
Can you fix it?
Interviews, not regressions? “Which claims rest on the fewest interviews?” “What else would explain the same quotes?”
A fix takes 5–10 minutes: it writes and runs code. If it asks to install Python, say yes.