Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› aamas-author-response
aamas-author-response
GitHub用于起草AAMAS会议的反驳信，指导在双盲、无新结果等约束下，针对博弈论、MARL及系统类审稿人的共性质疑进行高效回应，并提供匿名回复策略与多视角反驳模板。
触发场景
需要撰写AAMAS论文 rebuttal
收到AAMAS初步评审意见
处理多领域审稿人（博弈/MARL/系统）的混合反馈
安装
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill aamas-author-response -g -y
SKILL.md
Frontmatter
{
"name": "aamas-author-response",
"description": "Use when drafting an AAMAS rebuttal to preliminary OpenReview reviews, covering the double-blind constraint, the no-new-results and no-revised-paper norms, how to answer game-theory, MARL, and systems reviewers at once, and how to give the area chair a clean acceptance rationale within a short response window."
}
AAMAS Author Response
Use this after AAMAS preliminary reviews are released. Reopen the current OpenReview instructions and the rebuttal-window rules before drafting; response mechanics are cycle-specific and AAMAS reviews and decisions become public afterward.
Triage
- Answer only what can move the decision: correctness of the game model or proofs, the reality of the interaction claim, baseline/opponent fairness, reproducibility, and scope.
- Use evidence already in the submission: sections, appendices, the supplement, theorem statements, learning curves, and opponent-set definitions. AAMAS rebuttals do not add new results and typically do not upload a revised paper.
- Keep the reply anonymous: no institution, no named system, no private URLs or repository owners.
- Correct outright factual errors first, then address requests for missing comparisons, ablations, equilibrium analysis, or seeds.
Drafting pattern
- State the decision-critical correction or concession in the first line.
- Point to the exact submitted evidence (section, figure, appendix, or theorem number).
- Explain the strategic or empirical consequence for the reviewer's concern.
- Promise a camera-ready wording fix only if it adds no unsupported new claim.
Multi-reviewer pushback patterns
AAMAS pools game theorists, MARL researchers, and systems-minded reviewers on one paper, so a single rebuttal often answers three different lenses at once.
| Pushback | What it signals | AAMAS-ready fix |
|---|---|---|
| "This is really a single-agent method" | The interaction claim did not land | Point to where the result depends on other agents co-adapting; cite the ablation that freezes them |
| "Which equilibrium do you converge to?" | A game theorist read the solution concept | Name it (Nash / correlated / coarse / no-regret) and cite the proof or convergence plot |
| "Baselines are self-play only" | Opponent set looks weak | Anchor to the held-out opponents or population in the appendix, or concede a camera-ready addition |
| "Seeds and variance are missing" | An empirical reviewer checked rigor | Reference the seeded runs and standard errors; do not promise new experiments as fact |
Response micro-example
Reviewer objection: the cooperation result may be an artifact of a single random seed. Reply skeleton:
- Note that Figure 3 already aggregates 30 seeds with standard errors.
- Quote the exact aggregate so the reviewer need not re-open the PDF.
- Point to Appendix C for the per-seed spread and the opponent set used.
- Offer one camera-ready sentence stating the seed count in the caption.
Discussion-window calibration
- One decision-critical point per reviewer beats an exhaustive reply; the area chair reads for whether the central interaction or correctness objection was actually resolved.
- Never paste a fresh proof or a new game variant into the response box; sketch the argument and anchor it in submitted material.
- Reply early - AAMAS rebuttal windows are short and reviewers who engage once rarely return.
- Remember the reviews and your response may become public with the paper; write as if they will.
Output format
[Priority issue] <reviewer concern>
[Decision dimension] correctness / interaction-reality / significance / reproducibility / scope
[Draft response] <AAMAS-ready anonymous text>
[Evidence anchor] <section/appendix/figure/theorem>
[Forbidden content removed] <identity leaks, new results, revised-paper promises>
版本历史
- 9f86f09 当前 2026-07-19 14:15
