Состязательное рецензирование научной статьи

★ 7.1 · testing

kill-argument is a Claude Code skill that runs a two-thread adversarial review of a research paper: one fresh reviewer writes the single strongest ~200-word rejection memo designed to convince a senior area chair, while a second fresh reviewer defends the paper point-by-point and classifies each attack as answered_by_current_text, partially_answered, or still_unresolved. Unlike balanced score-based reviews that distribute attention evenly across weaknesses, this skill forces a commitment to the one argument most likely to kill the submission. Both reviewer threads are independent mcp__codex__codex calls with no shared context from prior review rounds, enforcing a strict reviewer-bias guard. Output is written to KILL_ARGUMENT.md and KILL_ARGUMENT.json in the paper directory, with optional HTML rendering. The skill is most valuable for theory papers with five or more theorem-class environments and for rebuttal preparation after standard review rounds have already stabilized.