Comment by bulba4aur

To clarify, each failing test triggers "review" agent, to determine "why" the test fails, and again, it can be improved with better heuristics probably, more in depth static analysis than the source code, but it is how it works in the current version.