On Fri, Oct 16, 2026 at 02:06 PM Sasha Levin <[email protected]> wrote: > What I had in mind was the kind of bug our scripts/coccinelle/ rules look for.
For kfree_mismatch.cocci kind, yes. Each site is its own bug. But the other example in the guide is "a missing put when leaving a loop". That's iterators/for_each_child.cocci. It was fixed one driver at a time for years until for_each_child_of_node_scoped() made the put automatic. > I agree that fixing the interface is better, but do we want to push all these > agents there? To report it, yes. The skill is read-only anyway. Whether to change the interface is for the human to decide. > So I think it should be a balance. When the skill finds the same defect > in several places, it should also ask whether the interface invites the > mistake, and if so report that the fix belongs there instead of handing > back a list for N maintainers Makes sense to me.
