Reviews code changes using a split-context adversarial method: the reviewer sees ONLY the diff (never the author's reasoning) and is tasked solely with finding bugs and reasons the code is wrong. Designed for reviewing large volumes of machine-generated or ported code where the author is biased toward shipping. Use when reviewing a port or rewrite, auditing AI-generated code, checking a big diff for correctness/ regressions, when the user asks to "find bugs", "review this like an adversary", "poke holes", "red-team this change", or wants a second independent pass on code that already compiles. Especially valuable when the author and reviewer would otherwise be the same co…
