wtf is rebuttal-skill?

Rebuttal Skill is not a standalone app but a structured skill file that teaches an AI coding assistant like Claude Code, OpenCode, or Gemini CLI how to help researchers respond to peer review comments on academic papers. It targets conferences such as NeurIPS, ICML, ACL, and CVPR, where authors get a limited window to write a rebuttal defending their paper against reviewer criticism. The skill works in stages. First, it decides whether writing a rebuttal is even worth the effort, sorting the situation into promising, borderline, or low expected return, and for weak cases it instead suggests a plan for revising and resubmitting the paper elsewhere later. If a rebuttal looks worthwhile, the skill reads through every review comment, breaks it into individual concerns, tries to figure out what the reviewer actually meant, and ranks each concern by severity from fatal problems down to minor ones. It then produces a prioritized plan of which experiments or clarifications to tackle first given limited time. Once the authors have their results, the skill helps draft the actual rebuttal text, using a pattern of directly answering the concern, presenting the evidence, and stating what will change in the revised paper. It also includes ready-made response patterns for common situations, like correcting a reviewer's misunderstanding, defending the paper's novelty, or explaining a missing baseline comparison. A stated rule throughout is that the skill must not fabricate results, present planned work as if it were already done, or hide negative findings, using clear placeholders instead when information is missing. To use it, a researcher loads the skill into a compatible AI assistant and provides the paper's abstract, the full reviews, scores, and details like the venue and deadline, and the assistant works through the stages from there.