rebuttal-strategist

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Rebuttal Strategist

Turn real reviewer feedback into a tactical rebuttal plan, experiment plan, paper revision plan, and response draft. This skill starts after reviews arrive. It is not a pre-submission shadow review.

Use this skill for:

OpenReview review extraction and thread analysis
reviewer intent and decision-state analysis
issue board creation from real comments
deciding which experiments, analyses, proofs, or clarifications to add during rebuttal
drafting point-by-point author responses
preparing concise follow-up replies during discussion
tracking paper revisions promised in the response
saving rebuttal memory for future rounds

Pair this skill with:

```
research-project-memory
```
when real reviews, issue boards, rebuttal experiments, or promised revisions should persist across sessions
```
paper-reviewer-simulator
```
for pre-submission shadow review or for stress-testing the draft response
```
run-experiment
```
when the rebuttal plan requires new experiments
```
conference-writing-adapter
```
when accepted reviewer criticism requires paper restructuring or clearer prose
```
citation-audit
```
and
```
citation-coverage-audit
```
when reviews identify citation problems
```
camera-ready-finalizer
```
after acceptance to verify promised revisions, final claim/evidence consistency, de-anonymization, and release handoff

将真实审稿人的反馈转化为具有战术性的回复计划、实验计划、论文修订计划和回复草稿。该技能在收到评审意见后启用，并非用于提交前的模拟评审。

本技能适用于：

OpenReview评审意见提取与线程分析
审稿人意图与决策状态分析
根据真实评论创建问题看板
决定回复阶段需补充的实验、分析、证明或说明内容
草拟逐点的作者回复
准备讨论阶段的简洁跟进回复
跟踪回复中承诺的论文修订内容
保存回复记忆用于后续评审轮次

可搭配以下技能使用：

```
research-project-memory
```
：当真实评审意见、问题看板、回复实验或承诺的修订内容需要跨会话持久化时
```
paper-reviewer-simulator
```
：用于提交前的模拟评审或对回复草稿进行压力测试
```
run-experiment
```
：当回复计划需要开展新实验时
```
conference-writing-adapter
```
：当审稿人的合理批评要求调整论文结构或优化表述时
```
citation-audit
```
和
```
citation-coverage-audit
```
：当评审意见指出引用问题时
```
camera-ready-finalizer
```
：论文录用后，验证承诺的修订内容、最终论点与证据的一致性、去匿名化及交付发布

Skill Directory Layout

技能目录结构

text

<installed-skill-dir>/
├── SKILL.md
└── references/
    ├── decision-strategy.md
    ├── experiment-response-planning.md
    ├── issue-board.md
    ├── memory-model.md
    ├── openreview-protocol.md
    ├── rebuttal-writing-style.md
    ├── report-template.md
    └── review-intent-analysis.md

text

<installed-skill-dir>/
├── SKILL.md
└── references/
    ├── decision-strategy.md
    ├── experiment-response-planning.md
    ├── issue-board.md
    ├── memory-model.md
    ├── openreview-protocol.md
    ├── rebuttal-writing-style.md
    ├── report-template.md
    └── review-intent-analysis.md

Progressive Loading

渐进式加载

Always read

references/review-intent-analysis.md

references/issue-board.md

, and

references/rebuttal-writing-style.md

Read
```
references/openreview-protocol.md
```
when the user provides an OpenReview URL, forum ID, review thread, or asks to fetch review information.
Read
```
references/decision-strategy.md
```
when scores, confidence, reviewer stances, or AC/meta-review dynamics matter.
Read
```
references/experiment-response-planning.md
```
when reviews request more experiments, baselines, ablations, proofs, analyses, or details.
Read
```
references/memory-model.md
```
when saving or reusing project rebuttal state.
Use
```
references/report-template.md
```
for substantial plans or saved reports.
Verify current venue rebuttal rules, response length, deadline, anonymity constraints, and discussion policy from official sources when they affect the response.

务必阅读

references/review-intent-analysis.md

、

references/issue-board.md

和

references/rebuttal-writing-style.md

。

当用户提供OpenReview链接、论坛ID、评审线程或要求获取评审信息时，阅读
```
references/openreview-protocol.md
```
。
当评分、置信度、审稿人立场或AC/元评审动态至关重要时，阅读
```
references/decision-strategy.md
```
。
当评审意见要求补充更多实验、基线、消融实验、证明、分析或细节时，阅读
```
references/experiment-response-planning.md
```
。
当需要保存或复用项目回复状态时，阅读
```
references/memory-model.md
```
。
如需制定完整计划或保存报告，使用
```
references/report-template.md
```
。
当回复规则、长度限制、截止日期、匿名约束和讨论政策会影响回复内容时，从官方渠道核实当前会议的相关规定。

Core Principles

核心原则

Rebuttal is strategic, not just polite. The goal is to change the decision path.
The audience is reviewers and the area chair. Write responses that help the AC see an accept path.
Prioritize persuadable reviewers and high-impact misunderstandings. Do not spend most of the budget arguing with an unmovable reject unless their objection can dominate the meta-review.
Separate real scientific weaknesses from presentation misunderstandings.
Lead with evidence. Every response should cite a result, table, figure, theorem, appendix detail, new experiment, or concrete planned revision.
Concede correct criticism cleanly. Defensive tone loses credibility.
Never promise experiments, revisions, code, or analyses that the authors cannot deliver.
Preserve anonymity and venue policy during rebuttal.

回复是具有策略性的，而非仅仅保持礼貌。目标是改变决策走向。
受众是审稿人和领域主席（AC）。撰写的回复应帮助AC找到录用的理由。
优先关注可被说服的审稿人以及影响重大的误解。除非反对意见会主导元评审，否则不要将大部分精力花在说服坚决拒绝的审稿人上。
区分真实的科学缺陷与表述上的误解。
以证据为导向。每一条回复都应引用结果、表格、图表、定理、附录细节、新实验或具体的修订计划。
坦然接受合理的批评。防御性语气会失去可信度。
绝不要承诺作者无法完成的实验、修订、代码或分析内容。
回复过程中遵守匿名规则和会议政策。

Step 1 - Define Rebuttal Context

步骤1 - 定义回复背景

Identify:

venue, year, and track
response deadline and response length or format limit
whether discussion is per-review, unified, or both
whether follow-up comments are allowed
current scores, confidence, and any meta-review/AC note
paper source files and appendix
review source: OpenReview URL, exported JSON, copied review text, PDF, or screenshots transcribed by the user
mode:
- ```
plan
```
  : strategy, issue board, experiment plan
- ```
draft
```
  : write rebuttal text
- ```
followup
```
  : answer new reviewer questions during discussion
- ```
post-mortem
```
  : update memory after final decision

Default to

plan

before drafting if no issue board exists.

明确以下信息：

会议、年份与赛道
回复截止日期及回复长度或格式限制
讨论方式是按评审单独进行、统一进行还是两者兼具
是否允许跟进评论
当前评分、置信度以及任何元评审/AC备注
论文源文件及附录
评审来源：OpenReview链接、导出的JSON、复制的评审文本、PDF或用户转录的截图
模式：
- ```
plan
```
  ：策略制定、问题看板、实验计划
- ```
draft
```
  ：撰写回复文本
- ```
followup
```
  ：在讨论阶段回答审稿人的新问题
- ```
post-mortem
```
  ：最终决策后更新记忆

如果尚未创建问题看板，默认先进入

plan

模式再开始草拟回复。

Step 2 - Collect Reviews and Paper Evidence

步骤2 - 收集评审意见与论文证据

Read or fetch:

all reviews
scores and confidence
reviewer questions
strengths and weaknesses
official comments and follow-up threads
any meta-review or AC comment
paper abstract, introduction, method, experiments, appendix, figures, tables, and related work
experiment logs or notes that may answer reviewer concerns

If using OpenReview, follow

references/openreview-protocol.md

. If fetching is blocked or login is needed, ask the user for exported review text or pasted comments.

阅读或获取：

所有评审意见
评分与置信度
审稿人的问题
优势与劣势
官方评论及跟进线程
任何元评审或AC评论
论文摘要、引言、方法、实验、附录、图表及相关工作
可能解答审稿人疑问的实验日志或笔记

如果使用OpenReview，遵循

references/openreview-protocol.md

。如果无法获取或需要登录，请向用户索要导出的评审文本或粘贴的评论内容。

Step 3 - Atomic Review Parsing

步骤3 - 原子化评审解析

Break every review into atomic issues.

Each issue should have:

reviewer ID
exact local concern
category: novelty, correctness, theory, experiment, baseline, ablation, related work, clarity, reproducibility, ethics, limitation, formatting, question, praise
severity
whether the reviewer is factually correct
whether the issue is answerable with existing evidence
response posture

Do not answer broad review paragraphs as a blob. One paragraph may contain several independent issues.

将每条评审意见拆解为原子化问题。

每个问题应包含：

审稿人ID
具体的局部关注点
类别：创新性、正确性、理论性、实验性、基线、消融实验、相关工作、清晰度、可复现性、伦理、局限性、格式、问题、表扬
严重程度
审稿人是否事实正确
该问题是否可用现有证据解答
回复姿态

不要将宽泛的评审段落作为整体回复。一个段落可能包含多个独立问题。

Step 4 - Infer Reviewer Intent and Decision State

步骤4 - 推断审稿人意图与决策状态

Read

references/review-intent-analysis.md

and

references/decision-strategy.md

For each reviewer, infer:

stance: champion, likely accept, borderline persuadable, skeptical but fixable, likely reject, unmovable reject
whether the reviewer likes the core idea
whether requested changes are small, medium, or fatal
whether the reviewer is asking for clarification or building a reject case
what evidence would move them

Then infer the paper-level decision path:

current decision state
main accept path
main reject path
which reviewer or issue is pivotal
what the AC is likely to care about

阅读

references/review-intent-analysis.md

和

references/decision-strategy.md

。

针对每位审稿人，推断：

立场：支持者、可能录用、可说服的边缘状态、持怀疑态度但可修正、可能拒绝、坚决拒绝
审稿人是否认可核心观点
要求的修改是小幅度、中等幅度还是致命性的
审稿人是要求澄清还是在构建拒绝理由
哪些证据可以改变其立场

然后推断论文层面的决策路径：

当前决策状态
主要录用路径
主要拒绝路径
关键的审稿人或问题
AC可能关注的内容

Step 5 - Build Issue Board

步骤5 - 构建问题看板

Read

references/issue-board.md

Rank issues:

```
must-win
```
: could decide acceptance if answered well
```
must-answer
```
: direct reviewer question or serious concern
```
quick-win
```
: easy clarification with high value
```
experiment-needed
```
: requires new experiment/analysis/proof
```
paper-revision
```
: can be fixed by promised text change
```
do-not-overinvest
```
: low impact or unmovable objection

The issue board should decide what gets response budget.

阅读

references/issue-board.md

。

对问题进行排序：

```
must-win
```
：若回答得当可能决定录用结果
```
must-answer
```
：审稿人的直接问题或严重担忧
```
quick-win
```
：易于澄清且价值高的问题
```
experiment-needed
```
：需要开展新实验/分析/证明的问题
```
paper-revision
```
：可通过承诺修改文本解决的问题
```
do-not-overinvest
```
：影响小或无法改变的反对意见

问题看板应决定回复资源的分配优先级。

Step 6 - Plan Experiments and Evidence

步骤6 - 规划实验与证据

Read

references/experiment-response-planning.md

For each issue requiring evidence:

define the smallest credible experiment, ablation, proof sketch, analysis, or table
estimate feasibility before deadline
identify required data, code, compute, metric, and owner
define success, partial success, and failure response wording
decide whether it belongs in rebuttal text, appendix, or future revision

Prefer targeted experiments that directly answer reviewer objections over broad new result hunting.

阅读

references/experiment-response-planning.md

。

针对每个需要证据的问题：

定义最具可信度的最小实验、消融实验、证明梗概、分析或表格
评估截止日期前的可行性
确定所需的数据、代码、计算资源、指标及负责人
定义成功、部分成功和失败情况下的回复措辞
决定内容应放在回复文本、附录还是未来修订中

优先选择直接回应审稿人反对意见的针对性实验，而非广泛探索新结果。

Step 7 - Choose Response Posture

步骤7 - 选择回复姿态

For every issue choose one posture:

```
accept-and-fix
```
```
clarify-misunderstanding
```
```
rebut-with-evidence
```
```
partially-concede
```
```
provide-new-result
```
```
scope-and-limit
```
```
defer-to-revision
```
```
do-not-address-directly
```

Read

references/rebuttal-writing-style.md

for wording guidance.

针对每个问题选择一种姿态：

```
accept-and-fix
```
：接受并修正
```
clarify-misunderstanding
```
：澄清误解
```
rebut-with-evidence
```
：用证据反驳
```
partially-concede
```
：部分让步
```
provide-new-result
```
：提供新结果
```
scope-and-limit
```
：明确范围与局限性
```
defer-to-revision
```
：推迟至修订阶段
```
do-not-address-directly
```
：不直接回应

回复措辞指导请阅读

references/rebuttal-writing-style.md

。

Step 8 - Draft Rebuttal

步骤8 - 草拟回复

Draft in the required format:

per-reviewer response
unified response
short OpenReview author comment
follow-up reply
camera-ready response summary, if relevant

Default structure:

one-sentence appreciation and thesis
answer pivotal concerns first
group repeated concerns across reviewers
provide new results and concrete evidence
state paper revisions promised
answer remaining reviewer-specific questions

Use concise, non-defensive language. Do not waste budget thanking every reviewer separately unless format requires it.

按照要求的格式草拟：

针对每位审稿人的回复
统一回复
简短的OpenReview作者评论
跟进回复
若相关，终稿回复摘要

默认结构：

一句话致谢与核心论点
优先回答关键问题
整合多位审稿人的重复问题
提供新结果与具体证据
说明承诺的论文修订内容
回答剩余的审稿人特定问题

使用简洁、非防御性的语言。除非格式要求，否则不要单独感谢每位审稿人，以免浪费篇幅。

Step 9 - Stress-Test the Draft

步骤9 - 压力测试草稿

Before finalizing, check:

Does the draft answer the strongest reject path?
Does it give the AC an accept path?
Are promised revisions feasible?
Are new results stated with enough detail?
Are reviewer misunderstandings corrected without sounding combative?
Are correct criticisms acknowledged?
Are repeated issues consolidated?
Are venue length and anonymity constraints respected?

If the draft fails, revise once and report remaining risks.

最终定稿前，检查：

草稿是否回应了最可能导致拒绝的路径？
是否为AC提供了录用的理由？
承诺的修订是否可行？
新结果的表述是否足够详细？
是否在不显得针锋相对的情况下纠正了审稿人的误解？
是否认可了合理的批评？
是否整合了重复问题？
是否遵守了会议的长度限制和匿名约束？

如果草稿未通过测试，修订一次并报告剩余风险。

Step 10 - Update Memory

步骤10 - 更新记忆

Read

references/memory-model.md

When reviews were parsed or a strategy was created, update project-local memory under:

text

.agent/rebuttal-strategy/

Track:

original reviews and parsed issues
reviewer intent map
experiment plan
response drafts
promised paper revisions
follow-up discussion state
final decision, if known

If the project uses

research-project-memory

, also update:

```
memory/risk-board.md
```
: real reviewer risks and issue severity, using certainty
```
observed
```
for review text and
```
inferred
```
for intent
```
memory/action-board.md
```
: rebuttal experiments, response drafting tasks, promised revisions, and post-rebuttal follow-ups
```
memory/evidence-board.md
```
: new rebuttal experiments, proof sketches, analyses, or tables
```
memory/claim-board.md
```
: claims reviewers challenged, weakened, clarified, or supported
```
rebuttal/.agent/rebuttal-status.md
```
: issue board, reviewer intent map, response plan, promised revisions, and discussion state

Never mark a promised revision as done until the paper/code change exists. Link promises to actions.

阅读

references/memory-model.md

。

当评审意见被解析或策略制定完成后，更新项目本地记忆至：

text

.agent/rebuttal-strategy/

跟踪内容包括：

原始评审意见与解析后的问题
审稿人意图映射
实验计划
回复草稿
承诺的论文修订内容
跟进讨论状态
最终决策（若已知）

如果项目使用

research-project-memory

，还需更新：

```
memory/risk-board.md
```
：真实的审稿人风险与问题严重程度，评审文本使用确定性标记
```
observed
```
，意图使用
```
inferred
```
```
memory/action-board.md
```
：回复实验、回复撰写任务、承诺的修订内容及回复后的跟进事项
```
memory/evidence-board.md
```
：新的回复实验、证明梗概、分析或表格
```
memory/claim-board.md
```
：审稿人质疑、弱化、澄清或支持的论点
```
rebuttal/.agent/rebuttal-status.md
```
：问题看板、审稿人意图映射、回复计划、承诺的修订内容及讨论状态

除非论文/代码修改已完成，否则不要标记承诺的修订为已完成。将承诺与行动关联起来。

Output Modes

输出模式

Strategy Plan

策略计划

markdown

undefined

markdown

undefined

Rebuttal Strategy

回复策略

Situation Summary

情况总结

Reviewer Intent Map

审稿人意图映射

Decision Path

决策路径

Issue Board

问题看板

Experiment / Evidence Plan

实验/证据计划

Response Posture

回复姿态

Draft Outline

草稿大纲

Follow-up Strategy

跟进策略

undefined

undefined

Draft Response

回复草稿

markdown

undefined

markdown

undefined

Rebuttal Draft

回复草稿

Response Strategy Notes

回复策略说明

Draft

草稿内容

Claims Requiring Verification

需要验证的论点

Promised Paper Revisions

承诺的论文修订内容

Remaining Risks

剩余风险

undefined

undefined

Follow-Up Reply

跟进回复

markdown

undefined

markdown

undefined

Follow-Up Reply

跟进回复

New Reviewer Comment

新审稿人评论

Intent Assessment

意图评估

Recommended Reply

Risk If Ignored

忽略的风险

undefined

undefined

Final Sanity Check

最终检查

Before finalizing:

real reviews are separated from inferred intent
all scores/confidence are recorded when available
the pivotal reviewer/issue is identified
every must-win issue has evidence or a clear fallback
response text follows venue limits and policy
no impossible promises are made
experiment plan has feasible deadlines
paper revision promises are tracked
memory updates are written if the user is working in a project repo

最终定稿前确认：

真实评审意见与推断意图已区分
所有可用的评分/置信度已记录
已识别关键审稿人/问题
每个must-win问题都有证据或明确的备选方案
回复文本符合会议限制与政策
未做出无法实现的承诺
实验计划的截止日期可行
论文修订承诺已被跟踪
若用户在项目仓库中工作，已完成记忆更新

rebuttal-strategist

Original

Translation

Rebuttal Strategist

Rebuttal Strategist

Skill Directory Layout

技能目录结构

Progressive Loading

渐进式加载

Core Principles

核心原则

Step 1 - Define Rebuttal Context

步骤1 - 定义回复背景

Step 2 - Collect Reviews and Paper Evidence

步骤2 - 收集评审意见与论文证据

Step 3 - Atomic Review Parsing

步骤3 - 原子化评审解析

Step 4 - Infer Reviewer Intent and Decision State

步骤4 - 推断审稿人意图与决策状态

Step 5 - Build Issue Board

步骤5 - 构建问题看板

Step 6 - Plan Experiments and Evidence

步骤6 - 规划实验与证据

Step 7 - Choose Response Posture

步骤7 - 选择回复姿态

Step 8 - Draft Rebuttal

步骤8 - 草拟回复

Step 9 - Stress-Test the Draft

步骤9 - 压力测试草稿

Step 10 - Update Memory

步骤10 - 更新记忆

Output Modes

输出模式

Strategy Plan

策略计划

Rebuttal Strategy

回复策略

Situation Summary

情况总结

Reviewer Intent Map

审稿人意图映射

Decision Path

决策路径

Issue Board

问题看板

Experiment / Evidence Plan

实验/证据计划

Response Posture

回复姿态

Draft Outline

草稿大纲

Follow-up Strategy

跟进策略

Draft Response

回复草稿

Rebuttal Draft

回复草稿

Response Strategy Notes

回复策略说明

Draft

草稿内容

Claims Requiring Verification

需要验证的论点

Promised Paper Revisions

承诺的论文修订内容

Remaining Risks

剩余风险

Follow-Up Reply

跟进回复

Follow-Up Reply

跟进回复

New Reviewer Comment

新审稿人评论

Intent Assessment

意图评估

Recommended Reply

推荐回复内容

Risk If Ignored

忽略的风险

Final Sanity Check