Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› conext-writing-style
conext-writing-style
GitHub指导撰写ACM CoNEXT论文，强调首屏展示网络贡献与部署上下文，主张每个声明必须绑定真实平台测量证据，并将局限性作为核心论点前置。同时严格遵循PACMNET的页数预算规范，确保内容严谨且符合审稿预期。
Trigger Scenarios
需要撰写或修改ACM CoNEXT会议论文
优化网络系统/测量类论文的引言和论证结构
检查论文是否符合PACMNET的篇幅限制
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill conext-writing-style -g -y
SKILL.md
Frontmatter
{
"name": "conext-writing-style",
"description": "Use when shaping the prose and structure of an ACM CoNEXT paper — leading with the networking problem and deployment context, tying every claim to a measurement on the real target platform, arguing limitations rather than reciting them, and holding the acmart page budget for long (≤16) and short (≤10) papers."
}
CoNEXT Writing Style
Write to the CoNEXT reader: a networking systems/measurement expert who wants the networking contribution and the evaluation platform on the first page, evidence proportional to the claim and drawn from the real target, and a limitations posture visible from the start. Because CoNEXT papers are PACMNET journal articles judged in a two-round, one-shot-revision process, the draft must read like a manuscript whose claims are already backed, not a promising sketch.
The first-page arc
Lead with the arc the worked example demonstrates
(../../resources/worked-examples/01-introduction.md):
- The networking problem, in a real deployment context — a problem an operator, protocol designer, or systems builder recognizes, stated in the first breath.
- Why the current state is inadequate — what existing systems/measurements miss, and why the gap matters on real paths or hardware.
- The contribution, as networking claims — a mechanism, a measured phenomenon, or both, not a model or a score.
- Evidence on the real target — a testbed run, a deployment, or a trace, each claim paired with the measurement that backs it.
- What changes for networking + limitations posture — the payoff for practice, with the central threat named up front rather than deferred.
Claims tied to measurements
- State the claim as an outcome an operator or protocol cares about ("completion time for throttled short flows drops"), not as an offline metric ("accuracy is high").
- Pair each claim with its evidence and its platform: which testbed, which trace, which vantage points, which baseline. A claim with no matching measurement is the fastest path to a one-shot-revision change-list item.
- Report uncertainty. Confidence intervals, multiple runs, and effect sizes read as rigor; a single bar with no variance reads as a lab artifact to a measurement reviewer.
Limitations as argument, not boilerplate
CoNEXT reviewers reward a paper that names its own central threat and bounds it. Put the limitation where the result lives, not only in a closing section:
- If a measurement infers an outcome from a proxy signal, say so and bound it with a ground-truth subset.
- If evaluation is on one testbed or one network, scope the generalization claim explicitly.
- If a mechanism assumes hardware behavior, state what breaks if the assumption fails.
A limitations section that only lists generic caveats ("results may not generalize") wastes the one place you can pre-empt the reviewer's strongest objection.
Page-budget discipline (acmart)
- Long papers ≤16 pages text+figures (+ unlimited references + ≤4 appendix pages); short papers ≤10 pages (+ unlimited references + ≤2 appendix pages). Verify the current numbers.
- References are unlimited, so never sacrifice a citation for space; compress prose, figures, and redundant tables instead.
- The appendix budget is small and for supporting, not deciding, material — nothing a reviewer needs
to accept the paper should live only there (see
conext-supplementary). - Do not alter the
acmarttemplate to reclaim space; editorial compression is the only safe lever.
Networking-prose specifics
- Define the topology and scale early. A reader should know what network, what hardware, and what scale before the evaluation.
- Name the baselines honestly and describe how they were tuned; an untuned strawman is a soundness flag.
- Figures earn their space. A topology diagram, a CDF, or a time series that a reviewer can read in one glance is worth more than a paragraph; a decorative architecture box is not.
- Anonymize as you write — refer to your own system and prior work in the third person, and do not name testbeds, operators, or repositories (double-anonymous review).
Common CoNEXT-specific failure modes
| Failure | Why it hurts at CoNEXT | Fix |
|---|---|---|
| Simulation stands in for the real platform | Platform-realism is a core CoNEXT expectation | Add a testbed/deployment run, or scope the claim |
| Claim outruns the measurement | The two-round review catches it | Pair each claim with evidence, or soften the claim |
| Limitations deferred to one closing paragraph | Misses the chance to pre-empt objections | Argue the central threat where the result lives |
| Model/leaderboard framing | Reads as an ML paper wearing a networking title | Reframe around the networking lesson (topic-selection) |
| Over-signposted roadmap | Substitutes structure for argument | One-line roadmap; let the contributions carry it |
Output format
[First-page arc] problem -> inadequacy -> contribution -> real-target evidence -> what changes + limitations
[Claim-evidence pairs] <claim -> measurement + platform + baseline>
[Limitations posture] <central threat named where the result lives? yes/no>
[Budget] pages used (body/appendix), acmart compliant? refs unlimited
[Anonymity] third-person self-reference; no testbed/operator/repo names
Version History
- 9f86f09 Current 2026-07-19 14:56
