ndss-writing-style
GitHub指导撰写和修改NDSS论文，强调首页明确能力受限对手与核心贡献，量化证据支撑声明，将伦理考量融入正文，并合理规划13页篇幅以通过初审。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill ndss-writing-style -g -y
SKILL.md
Frontmatter
{
"name": "ndss-writing-style",
"description": "Use when drafting or revising an NDSS paper's prose — building the first page around a capability-bounded networked adversary, sizing claims to evidence, weaving the ethics signal into the narrative, and fitting the argument into NDSS's 13-page body ahead of a two-round read."
}
NDSS Writing Style
An NDSS paper is read twice under different pressure. Round 1 decides, from a fast read, whether the paper continues at all — the early reject arrives with no rebuttal for the cut. Round 2 reads closely with your rebuttal in hand. Write the first page for the fast read and the rest for the close one.
The first-page contract
By the end of page one, a Round-1 reviewer should be able to fill in this table from memory:
| Question | Your first page must answer it as... |
|---|---|
| Who attacks? | A capability-bounded adversary: vantage (on/off-path, insider, tenant), budget, access |
| What breaks / what holds? | The precise property violated or guaranteed, in protocol/system terms |
| Against what? | Real systems named at the right granularity (protocol version, deployment class) |
| Why now? | The deployment shift, protocol change, or measurement gap that opens the question |
| How proven? | The evidence type: end-to-end exploit, testbed defense evaluation, population measurement |
| Is it handled responsibly? | One sentence signaling disclosure/harm posture, expanded later |
The single most common Round-1 kill is an adversary described by adjective ("powerful", "realistic") instead of by capability. Write attackers the way you would write an API contract: what they can call, what they cannot.
Sizing claims
- Bind every claim to the tested envelope: "defeats resolver-side validation as deployed in versions X-Y", never "defeats DNS validation."
- Quantify with populations, not superlatives. "Widespread" is an argument you have not made; "present in N of M sampled deployments, sampled by ⟨method⟩" is one you have.
- Negative space is credibility: a visible limitations paragraph that names the adversary who does beat your defense reads as competence, not weakness, to this PC.
- Never promise what evaluation cannot show — "completely prevents", "provably secure" (without a proof and a model), "first ever" (without a search you can defend).
The ethics thread
NDSS expects ethics to be woven through, not appended: a signal on page one, method-level handling where the risky procedure is described, and the optional Ethics Considerations section immediately before the references (excluded from the 13-page count, per the 2027 CFP, checked 2026-07-08) carrying the full protocol — disclosure timeline, harm analysis, Menlo-style reasoning, review-board interaction. Prose that describes live scanning or user data with no ethical acknowledgment reads to reviewers as a paper that has not understood the venue.
Fitting 13 pages
The 2027 limit is 13 pages excluding the ethics section, references, and appendices, in the NDSS template. Discipline that works:
- Budget pages at outline time: ~1 intro, ~1 background+threat model, ~4-5 design/attack, ~4 evaluation, ~1 related work, ~1 discussion+limitations. Adjust, but on purpose.
- One figure that explains the attack path or architecture is worth a page of prose; a figure that decorates costs a fifth of your evaluation budget.
- Compress with structure, not font tricks — template tampering is a desk-reject class offense at every security flagship, this one included.
- Move proofs, extra configurations, and protocol message details to appendices, but keep
every result the acceptance decision depends on inside the body (see
ndss-supplementaryfor the split rules).
Revision drill for an existing draft
PASS 1 (adversary): highlight every sentence that grants or assumes attacker
capability. They must all appear in the threat model — no capability
may debut inside the evaluation.
PASS 2 (claims): list every claim sentence; append its evidence pointer
(§/Fig/Table). Claims with no pointer get weakened or cut.
PASS 3 (vocabulary): replace "secure", "practical", "realistic", "large-scale"
with the number, version, or bound that justified the word.
PASS 4 (fast read): read only title, abstract, section heads, figure captions,
and the conclusion in 4 minutes — the Round-1 experience. Does the
contribution survive? If not, restructure headings, not sentences.
PASS 5 (ethics thread): page-one signal present? method-level handling present?
ethics section consistent with both?
Tone
Third person or first-person plural, past tense for what was measured, present for what the
system does. No exclamation of severity — the exploit speaks for itself or it does not.
Double-blind: strip names, funders, internal system codenames, and self-identifying phrases
like "our previous work" (cite yourself in third person; details in ndss-submission).
Output format
[First-page contract] 6/6 answered? list the misses
[Claim ledger] N claims, M without evidence pointers → actions
[Adversary discipline] capabilities that debut outside the threat model
[Ethics thread] page-1 signal / method handling / section — present or missing
[Page budget] planned vs actual per section
Version History
- 9f86f09 Current 2026-07-19 17:06
