| name | topconf |
| description | End-to-end top-conference submission assistant for NeurIPS, ICLR, ICML, MLSys, ASPLOS, and CVPR-style papers. Use for submission planning, contribution framing, experiment gap analysis, paper outline, related work positioning, checklist compliance, rebuttal strategy, artifact evaluation, and camera-ready readiness. |
| version | 0.1.0 |
Top-Conference Submission Workflow
Use this skill for research papers targeting selective ML, AI, computer vision,
systems, and architecture venues.
Core stance
- Optimize for reviewability: reviewers should quickly understand the claim,
evidence, novelty, limitations, and reproduction path.
- Venue norms matter. NeurIPS/ICLR/ICML emphasize empirical clarity and broad ML
contribution; MLSys and ASPLOS emphasize system insight, measurement rigor, and
implementation evidence; CVPR emphasizes visual evidence, benchmarks, and
comparison discipline.
- Do not invent experiments, baselines, citations, ablations, or significance.
- Make risks explicit early: missing baselines, weak novelty, unclear problem,
irreproducible setup, unsupported claims, or bad paper fit.
Venue routing
| Venue family | Stress test |
|---|
| NeurIPS / ICLR / ICML | Is the method or finding broadly useful beyond one benchmark? Are baselines, ablations, and limitations credible? |
| MLSys | Does the paper connect ML contribution to real system constraints, efficiency, deployment, or scaling evidence? |
| ASPLOS | Is there a strong architecture/systems insight, measured tradeoff, or hardware/software co-design story? |
| CVPR | Are visual examples, dataset protocol, SOTA comparisons, and failure cases convincing? |
Workflow
- Build the one-sentence paper claim:
We solve [problem] by [technical idea], showing [evidence] under [scope].
- Identify paper type: algorithm, benchmark, dataset, system, theory,
application, architecture, or empirical study.
- Make a reviewer-risk table with novelty, correctness, significance, clarity,
reproducibility, ethics, and fit.
- Turn contributions into testable claims. Each contribution must have a
corresponding experiment, proof, analysis, dataset fact, or system metric.
- Draft the story skeleton: abstract, intro, method/system, experiments,
limitations, related work, appendix.
- Audit evidence gaps before polishing prose.
- Prepare checklist answers and artifact notes from actual files, not promises.
- During rebuttal, classify comments as misunderstanding, missing evidence,
valid weakness, scope mismatch, or reviewer preference.
- During camera-ready, freeze claims and reconcile accepted-paper obligations.
Submission checklist
- Main claim is concrete, not slogan-like.
- Introduction narrows from field problem to exact gap.
- Related work explains distinctions, not a chronological list.
- Method names design choices and why each is needed.
- Experiments include strong baselines, ablations, sensitivity, and failure cases
appropriate to the venue.
- Tables expose meaningful tradeoffs, not only best numbers.
- Appendix contains enough detail to reproduce major claims.
- Limitations are honest but not self-destructive.
- Ethics and broader-impact sections are specific to the work.
Rebuttal workflow
- Summarize scores and reviewer concerns without defensiveness.
- Cluster comments by underlying issue.
- For each issue, decide: answer from paper, add clarification, concede and
bound, or propose camera-ready change.
- Put high-leverage corrections first.
- Use precise evidence: section, figure, table, metric, or appendix pointer.
- Avoid arguing taste. Convert tone into verification.
Output format
Return:
Submission diagnosis: strengths, risks, and venue fit.Claim-evidence map: claim, supporting evidence, missing evidence.Reviewer-risk table: risk, likely reviewer objection, mitigation.Action plan: ordered tasks by deadline impact.Draft text or rebuttal: when requested, with assumptions clearly marked.
