Community라이팅 & 에디팅github.com

plain-paper-writing

Master guide and operational rules for plain-language, high-impact academic writing and surgical paper revision for top-tier CS and security papers (IEEE S&P, USENIX Security, CCS, NDSS). Focuses on simple vocabulary, simple grammar, seamless flow (no mechanical First/Second or (1)(2)(3) lists), conceptual hierarchy, strict design-vs-implementation separation, zero fluff, and crystal-clear sentence/word-level diff delivery when revising manuscripts.

plain-paper-writing란 무엇인가요?

plain-paper-writing is a Antigravity agent skill that master guide and operational rules for plain-language, high-impact academic writing and surgical paper revision for top-tier CS and security papers (IEEE S&P, USENIX Security, CCS, NDSS). Focuses on simple vocabulary, simple grammar, seamless flow (no mechanical First/Second or (1)(2)(3) lists), conceptual hierarchy, strict design-vs-implementation separation, zero fluff, and crystal-clear sentence/word-level diff delivery when revising manuscripts.

지원 대상~Claude Code~Codex CLI~Cursor✓Antigravity
npx skills add https://github.com/zjy-dev/skills/tree/HEAD/skills/plain-paper-writing

Installed? Explore more 라이팅 & 에디팅 skills: steipete/notion, langchain-ai/langchain, bytedance/podcast-generation · View all 6 →

즐겨 사용하는 AI에게 물어보기

이 에이전트 스킬이 미리 로드된 새 채팅을 엽니다.

문서

Plain-Language Academic Paper Writing (Top-Tier CS & Security)

This skill operationalizes the "plain-language, reviewer-first" writing philosophy established for top-tier computer science and system security conferences (IEEE S&P, USENIX Security, ACM CCS, NDSS).


核心哲学:把审稿人当傻瓜 (The "Impatient Reviewer" Principle)

  1. 审稿人极其疲惫且时间有限:绝不能让审稿人猜你的意思或在长从句中寻找主谓宾。5 秒扫读抓主旨,1 分钟读完毫无卡顿。
  2. 词汇简单,语法更要简单:
    • 顶会论文不是 GRE 词汇竞赛。用最简明、纯正的大白话英语(高中至大一核心词汇),讲透复杂的系统、机制与因果。
    • 坚决杜绝带有从句嵌套、逗号连环的“四行大长难句”。控制句长在 15–25 词 之间,严格执行 一句话只讲一件事 (One Sentence, One Action/Claim)。
  3. 一气呵成,杜绝机械套话 (Seamless Flow):
    • 抛弃中学生式的起承转合词(如 First, Second, Next, Finally, Notably, Furthermore)。
    • 抛弃草稿感的列表分点(如 (1)(2)(3))。
    • 依靠紧密的因果逻辑链与纯正的短句推进(或强动词并列链),让文章读起来如同呼吸般顺畅。
  4. 设计与工程实现严格分离 (Strict Design vs. Implementation Separation):
    • 除 Implementation 章节外,其他任何章节(Abstract、Introduction、Motivation、Design、Evaluation 等)严禁混入工程实现细节。
    • 诸如构建检查、单元测试(unit tests)、静态 lint 检查、具体第三方库 API、CLI/RPC 胶水代码、JSON 序列化、线程池调参、正则解析细节等均属工程实现(Engineering Mechanics),必须全部下沉并收敛到 Implementation 章节。
    • 在 Introduction 与 Design 章节中,只讲核心挑战、为什么替代方案不可行(例如直接依赖模型生成可能出现幻觉与低效)、高层系统/形式化抽象(如类型状态建模与约束降维)以及机制保证。
    • 形式化措辞边界:轻量级形式化方案(如 formalizing ... in a typed-state IR、lowering into ... checkers)完全可以正面写;但若无 Coq/Isabelle 等完备数学证明,严禁使用 verify / verified / proven / fully safe 等强证明词汇,检测与校验统一用 check / test / detect / validate。

避坑与替换词表 (Negative Words vs. Plain Alternatives)

禁忌 / 装腔作势词汇 (AI & Academic Fluff)推荐大白话替换 (Plain & Direct)核心理由
demonstrate, showcase, illustrateshow简单直接,不掉书袋。
notably, remarkably, interestingly直接删掉,用 Some of... 摆事实审稿人反感空洞副词引导的自吹自擂。
are vital for mitigating memory exploitsprotect software against ... attacks动词直截了当,主谓清晰。
broken or removed(平级同义反复)weakened or broken必须体现从“部分削弱”到“完全破坏”的逻辑递进。
program corpora(NLP 跨界生造词)benchmark programs 或 source code系统与安全领域标准术语。
execution divergence / anomalous statecontinues running without crashes, yet memory is corrupted具象大白话:表面正常跑,内部状态已坏。
evade conventional CI pipelinesevade existing greybox fuzzers and static analyzers精准对准学术竞品与现有工具,不打工业界流水线稻草人。
Guided by these checkers(因果颠倒)Guided by mapped program locations / branch feedback检查器是裁判,不能指导生成;指导生成的是目标位置与执行反馈。

摘要(Abstract)两段式标准工业解剖

顶级安全会议的 Abstract 推荐严格采用两段式(Two-Paragraph Anatomy),全篇纯文本 Prose,严禁在正文打分点列表:

第 1 段:背景、机制、痛点、标杆案例与检测盲区

  • 句 1 (Context & Scope):指出核心安全机制及保护目标(精选 3 个代表,不堆砌 including...,不加 modern 等空洞修饰词)。

    例:Memory-safety mechanisms, such as bounds checking, shadow-memory tagging, and lifetime tracking, protect C and C++ software against heap overflows and use-after-free attacks.

  • 句 2 (The Core Problem):给出核心问题的大白话定义(服务继续运行且无告警或崩溃,但内存状态已损坏)。

    例:However, many memory bugs can \textbf{hide without immediate crashes}---the program continues serving requests normally, while heap metadata is quietly corrupted or freed objects are reused.

  • 句 3 (Running Example - Direct Consequence):引用经典 CVE/Bug,不解释复杂中间过程,直接给安全后果。

    例:For example, in \textbf{CVE-2015-0235}, a size calculation flaw in \texttt{glibc} allowed a crafted hostname to overflow a heap buffer and enable remote code execution.

  • 句 4 (The Detection Gap):解释为什么现有测试(纯覆盖率 Fuzzing / 常规回归测试)盲区明显。

    例:Such bugs expose a clear \textbf{detection gap}: conventional crash-oriented fuzzing only flags errors that trigger immediate faults, missing subtle heap corruption along valid execution paths.

第 2 段:系统、四个简单句推进、安全红线、成果与定论

  • 句 1 (System Introduction):亮出系统名与核心定位(无废话套话)。

    例:In this paper, we present \textbf{\textsc{MemScope}}, a lifetime-directed agentic framework for uncovering latent heap corruption bugs in C and C++ programs.

  • 句 2 (Step 1 - Extraction):说明从哪里提炼什么(简单句,主谓宾状)。

    例:\textsc{MemScope} first extracts object-lifetime constraints from source code, API contracts, and historical patch logs using a hybrid CoT and RAG pipeline.

  • 句 3 (Step 2 - Mapping):说明映射到程序具体位置(简单句)。

    例:It then maps these constraints to allocation sites and pointer dereferences, locating where objects cross ownership boundaries.

  • 句 4 (Step 3 - Oracle Formulation):先提预言机概念,再提 Checkers 作为实现手段(严防概念倒置!)。

    例:To detect violations, \textsc{MemScope} turns this mapped logic into programmatic bug oracles, implemented as static and dynamic shadow-state checkers.

  • 句 5 (Deterministic Adjudication Red Line - Independent Sentence):独立成句,明确裁决全靠确定性规则而非 LLM,彻底堵死审稿人质疑。

    例:All bug verdicts are decided strictly by these checkers, with \textbf{zero LLM involvement in bug adjudication}.

  • 句 6 (Step 4 - Directed Fuzzing to Hit Mapped Locations):体现定向 Fuzz 的真正本质——定向去打前面 map 到的位置。

    例:Guided by the mapped program locations, specialized LLM agents perform directed fuzzing to hit targeted allocation paths, synthesizing and mutating inputs to trigger lifetime violations under checker evaluation.

  • 句 7 (Evaluation - Active "We" Simple Sentence):主动简单句交代评估对象与发现数字。

    例:We evaluated \textsc{MemScope} on widely used C and C++ network libraries and parsers. \textsc{MemScope} discovered \textbf{\tbd{N} previously unknown memory bugs}, of which \textbf{\tbd{M} have been confirmed upstream}.

  • 句 8 (High-Severity Impact):说明安全漏洞级别,直接上定心丸。

    例:Some of these bugs were assigned high-severity CVE identifiers.

  • 句 9 (Holistic Conclusion):评价系统整体,且靶子精准对准学术竞品(使用 show,不用 demonstrate)。

    例:These results show that \textsc{MemScope} effectively uncovers deep heap-corruption bugs that evade existing greybox fuzzers and static analyzers.


章节概览(Section Overview)黄金范本

在方法论或系统设计章节开篇,用一段紧凑有力的白话建立全貌认知(示例如下):

In this section, we present the design of \textsc{MemScope}, a lifetime-directed framework for uncovering latent heap corruption bugs in C and C++ programs. Figure~\ref{fig:method-overview} illustrates its overall workflow, which consists of four core steps: constraint extraction, source-location mapping, programmatic oracle construction, and agentic directed fuzzing. First, we extract object-lifetime constraints from source code and historical bug fixes using a hybrid CoT and RAG pipeline. With these constraints established, we map them to allocation sites and pointer dereferences, identifying where objects cross ownership boundaries and which conditions govern their lifetimes. Next, we turn this mapped logic into programmatic bug oracles. We implement these oracles as static and dynamic shadow-state checkers to test whether program inputs expose memory violations, eliminating false positives from model hallucination. Finally, our agentic directed fuzzing coordinates specialized agents: a generator agent synthesizes inputs designed to stress target ownership paths, while a feedback agent analyzes execution traces to guide iterative mutations, with our programmatic oracles evaluating each candidate. We discuss each step in detail in the following subsections.

Overview 句式心法:

  1. 开篇立论:In this section, we present the design of SYSTEM, a [TYPE] for [GOAL].
  2. 图表指引:Figure X illustrates its overall workflow, which consists of four core steps: (1), (2), (3), and (4).
  3. 动作串联:
    • First, we [ACTION 1] ...
    • With these [RESULTS] established, we [ACTION 2] ..., identifying [KEY INSIGHT] ...
    • Next, we turn this [LOGIC] into [CONCEPT]. We implement these as [CHECKERS] to [PURPOSE], eliminating [RISK].
    • Finally, our [PROCESS] coordinates specialized agents: a generator agent ..., while a feedback agent ..., with our oracles evaluating each candidate.
    • We discuss each step in detail in the following subsections.

经典已发表顶会论文参考范本 (Gold Standards)

范本 1: AddressSanitizer (USENIX ATC 2012)

"Memory access bugs, including buffer overflows and uses of freed heap memory, remain a serious problem for programming languages like C and C++. Many memory error detectors exist, but most of them are either slow or detect a limited set of bugs, or both. This paper presents AddressSanitizer, a new memory error detector. Our tool finds out-of-bounds accesses to heap, stack, and global objects, as well as use-after-free bugs. It employs a specialized memory allocator and code instrumentation that is simple enough to be implemented in any compiler, binary translation system, or even in hardware. AddressSanitizer achieves efficiency without sacrificing comprehensiveness, and has found over 300 previously unknown bugs in the Chromium browser and many bugs in other software."

范本 2: Angora (IEEE S&P 2018)

"Fuzzing is a popular technique for finding software bugs. However, the performance of the state-of-the-art fuzzers leaves a lot to be desired. Fuzzers based on symbolic execution produce quality inputs but run slow, while fuzzers based on random mutation run fast but have difficulty producing quality inputs. We propose Angora, a new mutation-based fuzzer that outperforms the state-of-the-art fuzzers by a wide margin. The main goal of Angora is to increase branch coverage by solving path constraints without symbolic execution. To solve path constraints efficiently, we introduce several key techniques: scalable byte-level taint tracking, context-sensitive branch count, search based on gradient descent, and input length exploration. On the LAVA-M data set, Angora found almost all the injected bugs, found more bugs than any other fuzzer that we compared with, and found eight times as many bugs as the second-best fuzzer in the program who. Angora also found 103 bugs that the LAVA authors injected but could not trigger. We also tested Angora on eight popular, mature open source programs. Angora found 6, 52, 29, 40 and 48 new bugs in file, jhead, nm, objdump and size, respectively. We measured the coverage of Angora and evaluated how its key techniques contribute to its impressive performance."


论文修改交付规范 (Revision Delivery Protocol)

在修改已有论文(尤其是单行整段的 LaTeX .tex 源码)时,行级 git diff 会因整段挤在一行而彻底失效;而在对话中只贴一整段加粗定稿,会让作者无法分辨到底删了哪些原词、增了哪些新词、哪些原句未动。

因此,凡是提议修改或交付论文修改结果,必须严格按以下三要素输出每一处改动(未改动的无关句子严禁整段刷屏):

1. 交付卡片三要素模板

  1. 定位与基线(Location & Baseline):指明章节段落、回应的审稿/批注点,以及当前对比的基准版本(如:对比原稿 PDF,或对比当前工作区 main.tex)。
  2. 句级拆行 Diff 块(Sentence-Split diff):将修改处拆成一句一行放入 ```diff 代码块:
    • 仅保留修改处前 1 句、后 1 句不变原文作为上下文锚点(以两个空格 开头);
    • 被删除或被替换的原句以 - 开头(渲染为红底);
    • 新增或替换后的新句以 + 开头(渲染为绿底),并在句末标注词数如 [19w] 以自检 15–25 词规范。
  3. 句内词级修订行(Inline Word Track-Changes):若某句属于局部改词/增删短语(而非整句纯新增),在代码块下方补一行词级追踪:
    • 用 ~~删除的原词~~ 表示删除,用 新增的词 表示新增,普通正文字体表示未动原文,省去人眼在整句红绿行之间“找不同”。

2. 标准交付示例

[Intro 第 2 段首句] 具象化失效动作并统一术语(回应 [R2-C1],对比基线:原稿 PDF)

- When a parser handles invalid pointers improperly, it often leads to \textbf{latent heap corruption}: the service continues running normally, yet memory state is quietly broken.
+ When a parser frees an object before clearing its alias or overflows a buffer, it often causes \textbf{latent heap corruption}: the service continues running without warnings, yet memory state is silently broken. [34w -> 可按需拆句]
  Because normal test inputs rarely touch the corrupted region, standard functional suites pass while the server remains exposed to exploitation.
  • 词级追踪:When a parser handles invalid pointers improperly frees an object before clearing its alias or overflows a buffer, it often leads to causes latent heap corruption: the service continues running normally without warnings, yet memory state is quietly silently broken.

3. LaTeX 源码句级换行建议(One-Sentence-Per-Line,可选)

  • 在 LaTeX 中,段内单次换行(\n)与空格完全等价,空行(\n\n)才表示分段。
  • 若作者同意对修改过的 .tex 段落启用一句一行(Semantic Line Breaks),可在完全不改变 PDF 排版的前提下,让 IDE 与 git diff 永久恢复逐句对比能力。未经作者确认前,默认保持原文件换行习惯,仅在对话交付中强制执行上述句级 + 词级对比格式。

写作前检查清单 (Pre-Flight Self-Checklist)

在输出或修改任何论文段落前,必须在心中快速核验以下 9 项:

  • 句子长度是否都在 15–25 词内?有没有连续超过 30 词的喘不过气的长难句?
  • 有没有机械的 (1)(2)(3) 或中学生连词(First, Second, Next, Finally, Notably)?
  • 概念层级是否正确?(是否先提抽象概念/预言机 Oracle,再讲具体检查器或实现机制?)
  • 定向引导或核心机制是否写清了因果方向?(例如“利用前面定位的目标位置指导生成”,而不是用裁判指导生成)
  • 是否明确交代了确定性裁决机制(不依赖运行时 LLM 主观判读)?
  • 实验评估是否使用了主动的简单句(We evaluated...)?是否交代了真实安全/系统影响?
  • 最后的对比对象是否精准锁定为学术竞品(如 existing greybox fuzzers and static analyzers)而非泛泛工业流水线?
  • 【设计 vs. 实现隔离】 除 Implementation 章节外,是否彻底剔除了 unit tests、构建校验、底层库/工具链胶水等工程实现细节?在使用形式化建模表述时,是否避开了无证明支撑的 verify / verified / proven?
  • 【修改交付必查】 修改已有文稿时,是否严格使用了一句一行 diff 块(含前后各 1 句锚点)+ 句内 ~~删~~ / **增** 词级追踪,杜绝只贴整段定稿让作者盲找差异?

Individual skills in this repo

This repo contains 1 individual skill — each has its own dedicated page.

관련 스킬