Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› ieeesp-author-response
ieeesp-author-response
GitHub用于撰写IEEE S&P论文的第二轮反驳或修改回复。涵盖审阅前准备、快速分类审阅意见、应对适应性攻击及伦理质疑，并提供写作模板与策略，帮助作者在规定期限内高效完成高质量回应。
触发场景
需要撰写IEEE S&P论文的反驳信
收到修改决定需撰写修订回复
处理关于适应性攻击或威胁模型的质疑
整理伦理审查文档
安装
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill ieeesp-author-response -g -y
SKILL.md
Frontmatter
{
"name": "ieeesp-author-response",
"description": "Use when drafting the IEEE S&P (Oakland) rebuttal for second-round papers or the response plan for a Revise decision, including triaging reviews under the ~5-day window, answering adaptive-attack and threat-model objections, handling ethics questions with documentation, and writing against the Revise expectations summary."
}
IEEE S&P Author Response
Use this in two situations that S&P runs differently from most venues: the short rebuttal offered only to papers that survive the first round, and the Revise response, a full re-submission against a written expectations summary. The 2027 Cycle 2 rebuttal window was five days — reviews out Feb 11, rebuttal due Feb 16, 2027 (checked 2026-07-08) — so the work is mostly done before reviews arrive or it does not get done.
Pre-draft before reviews land
Between the early-reject notice and review release, write the skeleton:
- The three objections your own red-team read would raise (usually: an adaptive-attacker gap, a threat-model realism question, a measurement-validity question).
- The exact section/figure references answering each.
- The ethics documentation block: disclosure dates, vendor responses, IRB determination ID — assembled, dated, ready to paste.
Triage the review set in the first hour
| Objection type | Response strategy | Win probability |
|---|---|---|
| Misreading of mechanism or threat model | Quote the submission, cite section, restate in two lines | High |
| Missing experiment you actually have | Point to appendix/artifact location | High |
| Adaptive-attack gap that is real | Concede scope precisely; state what the evaluated adversary covers | Medium — honesty is scored |
| Threat-model taste ("attacker unrealistic") | Ground in incidents/measurements, not adjectives | Medium |
| Ethics record question | Dates, IDs, named approvals — documentation only | Binary: docs exist or not |
| Fundamental novelty dispute | One paragraph on the delta; do not relitigate | Low — spend words elsewhere |
Spend the budget in proportion to win probability times decision weight, and answer every reviewer at least once — an ignored reviewer in the post-rebuttal discussion is a vote you conceded.
Rules of engagement in the rebuttal
- Answer the objection actually written, not the friendlier one nearby.
- New experimental results: follow the current cycle's instructions on whether new numbers are allowed (待核实 each cycle); when unclear, phrase as verification of existing claims rather than new contributions.
- Never argue that a security concern "is out of scope" without saying whose scope — restate the threat model's boundary and why it is the right one.
- Keep the anonymous voice; disclosure specifics stay venue-neutral.
- No links out of HotCRP unless the CFP explicitly permits them.
Rebuttal skeleton
Common response (all reviewers):
C1. Threat model boundary: <2 lines + §ref>
C2. Adaptive-adversary coverage: <what was evaluated, §ref>
C3. Ethics record: disclosed <date>, IRB <id/waiver>, data handling <§ref>
Reviewer A:
A1 [misreading] The attack does not require <X>; see §3.2 ¶2.
A2 [evidence] The <n>-target evaluation is in Table 4; per-target
breakdown in Appendix C.
Reviewer B:
B1 [concession] Correct: we do not evaluate <adversary class>. Our claims
are scoped to <boundary> (§2); extending is future work.
→ ≤ <current cycle's length cap 待核实>; every point ends at a §/Table/date.
Responding to a Revise decision
The Revise summary of expectations is a contract, not a suggestion list:
- Extract each expectation into a numbered ledger; get co-author agreement on the interpretation of each before doing work.
- Deliver every item or explicitly negotiate the ones you contest — silence on an expectation reads as failure to meet it.
- Write a change document mapping expectation → change → location, so the returning reviewers can verify in minutes.
- Do not add major unrequested material; new claims reopen review scope.
- Confirm on the live cycle pages which cycle receives the revision and its deadline (待核实 — the 2027 mechanics were not published at check time).
Failure modes seen at this venue
- Spending the five days on the novelty dispute instead of the fixable evidence objections.
- Answering an ethics question with philosophy instead of a timeline.
- Treating Revise as "minor revisions plus a cover letter" — it is re-reviewed as a major revision against the written expectations.
- Missing that the rebuttal's real audience is the reviewer discussion that follows it: write lines a sympathetic reviewer can quote.
Output format
[Mode] rebuttal (5-day) / revise response
[Ledger] <n> objections: misreading <n> · evidence <n> · taste <n> · ethics <n>
[Concessions] <what is honestly out of evaluated scope>
[Documentation block] disclosure dates ✓/✗ · IRB id ✓/✗ · artifact refs ✓/✗
[Word budget plan] <allocation across reviewers>
[Open 待核实] new-results policy / length cap / revise target cycle
版本历史
- 9f86f09 当前 2026-07-19 15:59
