rebuttal

name: rebuttal description: "Reviewer 评论流水线：分类 → 找 supporting evidence → 写 rebuttal → cross-model review → submit。3 道 safety gate：no fabrication / no overpromise / full coverage。Use when user says "rebuttal", "reviews", "reviewer 来了", "反驳", "答辩", "OpenReview comment"." argument-hint: "[paper-dir-with-reviews] [--venue ICML] [--char-limit 5000]" allowed-tools: Bash(*), Read, Write, Edit, Agent, Skill, WebSearch

/rebuttal — Reviewer-Comment Reply Pipeline

借鉴 ARIS /rebuttal。Stage 9 / acceptance loop。

Overview

5 阶段：

0. Parse reviews → paper/reviews/parsed.json（per-reviewer concerns 列表）
1. Categorize 每条 concern：factual-question / request-for-experiment / clarification / subjective-criticism
2. Evidence gather 每条对应 paper/code/data 段
3. Draft rebuttal 文本（按 char limit）+ /cross-review --rounds 2
4. Safety gate：
   • 🔒 No fabrication — 每 claim 必须 link 到 paper/code/user-confirmed result
   • 🔒 No overpromise — 任何「we will add X」必须 user 批准
   • 🔒 Full coverage — 每 concern 都 addressed
5. Emit：PASTE_READY.txt（精确字符数）+ REBUTTAL_DRAFT_rich.md（rich text，留 manual edit）

Workflow

VENUE=${VENUE:-ICML}
CHAR=${CHAR_LIMIT:-5000}
python3 tools/rebuttal.py parse paper/reviews/*.md > paper/reviews/parsed.json
# loop 5 stages...

Output

• paper/reviews/parsed.json
• paper/rebuttal/draft.md
• paper/rebuttal/PASTE_READY.txt
• paper/rebuttal/REBUTTAL_DRAFT_rich.md

Composition

• 联动 /kill-argument 准备反驳武器
• 联动 /cross-review 二审
• 如有新 experiment 需求 → /experiment-bridge (未实现，placeholder)