"give me several theories of the possible problem and the underlying falsifiable assumptions to support them, then launch a subagent for each theory to validate" which is just the other side of "what are the underlying assumptions of our current solution that appears to be misbehaving, and which are the closest match to the described unexpected behavior?"
This could be applied to e.g. design-related skills, or even a new "bugfix" skill tailored specifically towards finding an optimal solution to fixing a bug.
"give me several theories of the possible problem and the underlying falsifiable assumptions to support them, then launch a subagent for each theory to validate" which is just the other side of "what are the underlying assumptions of our current solution that appears to be misbehaving, and which are the closest match to the described unexpected behavior?"
This could be applied to e.g. design-related skills, or even a new "bugfix" skill tailored specifically towards finding an optimal solution to fixing a bug.