Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› itcs-reproducibility
itcs-reproducibility
GitHub用于生成ITCS论文的可复现性数学包，确保所有核心主张有完整自包含证明、精确定义及依赖来源。适用于需要严格验证理论结果、应对无反驳期审稿的场景。
Trigger Scenarios
撰写ITCS论文附录
构建自包含数学证明
整理定理引用与假设
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill itcs-reproducibility -g -y
SKILL.md
Frontmatter
{
"name": "itcs-reproducibility",
"description": "Use to make an ITCS paper's mathematics independently checkable — complete proofs of every central claim, self-contained definitions, pinned dependencies on prior results, and a matching full version on arXiv\/ECCC\/ePrint — the pure-theory analogue of a reproducibility package."
}
ITCS Reproducibility
"Reproducibility" at a pure-theory venue means one thing: a competent, skeptical reader can verify every central claim from the paper alone. There is no code to rerun and no dataset to re-mine — the analogue of a reproducibility package is a complete, self-contained, correctly attributed proof. ITCS makes this concrete in its call: submissions must include complete proofs of all central claims (in an appendix if needed). This skill is the discipline that makes the mathematics checkable, which — since ITCS has no rebuttal — is also your only defense against a reviewer who gets stuck.
Completeness: every central claim is fully proved
- No "proof omitted," no "proof is standard," no "see the full version" for anything a reviewer must check to believe the result. Deferring a routine calculation to an appendix is fine; deferring the load-bearing lemma is a reproducibility failure and, at ITCS, a likely reject.
- Distinguish proved from assumed. Every step is either proved here or cited to a precise prior result. A theorem that quietly relies on an unproved claim is the theory analogue of a package that silently calls a missing dependency.
- Prove the anchoring result in full. For a new-model paper, the separation/possibility result that shows the model is alive is exactly what a reviewer will try hardest to break — give it the most complete treatment in the paper.
Self-containment: definitions and notation
- Define every symbol and model parameter before use. A reviewer should verify a theorem
statement without opening your prior papers — which lightweight double-blind would expose
anyway (see
itcs-related-work). - State the exact model, not "the usual one." Small variations in a definition (adaptive vs. non-adaptive, worst-case vs. average, the quantifier order) change what is true; pin them down.
- Restate borrowed results you rely on. Quote the precise statement (and its hypotheses) of an external theorem you invoke, so a reviewer sees exactly what you assume and can check the fit.
Dependency provenance: pin what you borrow
The theory analogue of "pinning versions" is precise citation of the results you build on:
- Cite each borrowed theorem to a specific venue, year, and statement number — not a vague "it is known that."
- Flag any dependency on a recent, unrefereed, or conditional result (an arXiv/ECCC preprint, a conjecture, an unpublished personal communication). A result standing on an unverified preprint should say so; its correctness is then explicitly conditional.
- Separate assumptions (P != NP, a hardness assumption, a conjecture) from theorems, and make every conditional result's hypothesis unmissable in its statement.
The full version as the durable record
Under lightweight double-blind, authors are encouraged to post the full version to arXiv, ECCC, or the IACR ePrint Archive. Treat it as the permanent, complete record:
- The full version and the submission must agree. Divergent theorem statements or proofs between the two invite a reviewer to trust the wrong one. Keep them in sync, and say in the submission that a full version exists.
- Put genuinely long proofs in the full version and an appendix the PC can read — a proof a
reviewer must consult to judge the paper cannot live only in an external preprint they are not
obligated to open (see
itcs-supplementary). - Keep the preprint's identity handling consistent with your strategy — posting is allowed and common, but the submitted PDF still omits author names.
The checkability passes (run before upload)
[Completeness] every central claim has a full proof present (appendix ok)? no "omitted"/"standard"?
[Self-contain] every symbol/model/parameter defined before use? theorem statements parseable alone?
[Dependencies] each borrowed result cited to venue+year+statement number? unrefereed deps flagged?
[Assumptions] every conditional theorem states its hypothesis in the statement?
[Consistency] abstract bound = theorem bound = proof bound; quantifier order uniform?
[Full version] submission and arXiv/ECCC/ePrint version agree? existence noted?
Common failure modes
| Symptom | Why it fails at ITCS | Fix |
|---|---|---|
| Central lemma "proof omitted" | Violates the complete-proofs requirement | Include the full proof (appendix) |
| Theorem parseable only with your prior paper | Not self-contained; anonymity risk too | Define everything in-paper |
| "It is known that ..." with no cite | Dependency unpinned; reviewer cannot check | Cite venue/year/statement |
| Relies on a preprint silently | Correctness is conditional but hidden | State the conditional dependency |
| Preprint proof differs from submission | Reviewer trusts the wrong version | Sync the two |
Output format
[ITCS checkability status] verifiable / gaps
[Completeness] central claims fully proved (list any "omitted")
[Self-containment] definitions complete? statements parseable alone? yes/no
[Dependencies] borrowed results pinned? unrefereed/conditional ones flagged? yes/no
[Full version] posted and consistent with submission? yes/no
[Fix queue] <ordered, load-bearing gaps first>
Version History
- 9f86f09 Current 2026-07-19 16:06
