com um clique
paper-proofread
LaTeX 论文审校:两阶段工作流,先检测所有问题再选择性修复。基于 ICRA/CVPR 级审稿标准。
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Menu
LaTeX 论文审校:两阶段工作流,先检测所有问题再选择性修复。基于 ICRA/CVPR 级审稿标准。
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
Coordinate multi-agent academic research workflows with sanitized Paperclip-style handoffs, verification gates, and release-safe outputs.
实现前的信心评估:检查理解和准备的充分程度
任意输入(代码、文档、论文、图片)→ 知识图谱 → 聚类社区 → HTML + JSON + 审计报告
知识库维护 — 扫描新资料、入库、健康检查、统计查询
生成 draw.io 图表(.drawio),可选导出 PNG/SVG/PDF
编排端到端自主 AI 研究项目,双循环架构:内循环快速实验迭代,外循环综合分析引导方向
| name | paper-proofread |
| description | LaTeX 论文审校:两阶段工作流,先检测所有问题再选择性修复。基于 ICRA/CVPR 级审稿标准。 |
Two-phase LaTeX paper proofreading system based on real conference review standards (ICRA, RSS, CVPR, NeurIPS, RA-L, T-RO).
Source: awesome-claudecode-paper-proofreading by Hyungtae Lim (ICRA 2025 Outstanding Reviewer).
This skill has TWO reference files containing the detailed review rules with ✅/❌ judgment examples. You MUST read both files before running any checks. The bullet points below are summaries only — the reference files contain the actual judgment criteria.
Read these files FIRST:
- references/01_latex_workspace_review.md (for workspace/full mode)
- references/02_paper_proofreading.md (for content/full mode)
If the reference files are missing, tell the user that the full review instructions are unavailable rather than proceeding with only the summaries in this file.
Invoke when the user asks to:
Ask the user for mode if not specified:
| Mode | Scope | When |
|---|---|---|
workspace | LaTeX infrastructure audit (C1-C9) | Preamble, packages, macros, refs, labels, figures |
content | Paper content proofreading (A-I) | Grammar, claims, structure, notation, captions |
full | Workspace first, then content | Recommended before submission |
.tex file (e.g., main.tex) — mandatoryBefore running any checks, silently read the full workspace:
.tex file\input{...} and \include{...} recursivelyshortcuts.tex, macros.tex, commands.tex, preamble.texpreamble_symbols.tex, preamble_packages.tex (in ../references/ if workspace uses shared folder).bib filefigures/ directory for missing assets (workspace mode)figures/ (final manuscript figures) from pics/ (raw sources — do NOT flag for missing .tex references)Never review only the top-level file unless the user explicitly narrows scope.
CRITICAL: Do NOT modify any files in Phase 1. Do NOT rewrite paragraphs. Do NOT apply style changes proactively.
[1], [2], [3]...CRITICAL | MAJOR | MINOR | STYLEAfter Phase 1 output, stop and wait for user decision.
Only edit files after explicit user approval:
| Command | Action |
|---|---|
fix safe | Only definite typos, duplicate words, clear grammar errors |
fix all critical | Only CRITICAL issues |
fix 1, 3, 5 | Specific issue numbers |
fix all | All issues |
discard 3, 7 | Skip specific issues |
When fixing:
When applying fixes in Phase 2:
—) in rewritten prose — use comma, colon, semicolon, or restructureThese are summaries. Read
references/01_latex_workspace_review.mdfor full rules with examples.
Duplicate/conflicting/unused/missing packages. cleveref [nameinlink,capitalize] options. hyperref colorlinks. caption font consistency. Math packages (amsmath, mathtools, bm). Algorithm environment conflicts.
hyperref→cleveref. amsmath→mathtools. xcolor→tikz. caption→subcaption. Flag subfig+subcaption conflict, epsfig redundancy, \usepackage{times}/pslatex.
\methodname defined once, used consistently, no hardcoded strings. \xspace on text-mode macros. \etalcite defined and used with ~ + plural verb. Subscript macros for repeated patterns. Shared symbol conflicts between local and shared preamble files. \renewcommand flagged as intentional or accidental.
\Cref{} standardization. Multi-reference grouping. Subfigure format 5(a) with \renewcommand\thesubfigure. Forward references minimized. ~ spacing before \Cref{}.
Prefixes: fig:, tab:, eq:, sec:, alg:, app:. No duplicate labels. No unreferenced labels. No dangling \ref{} calls.
Cited keys in bib. Unused bib entries. Duplicate BibTeX keys across files. arXiv formatting. Missing required fields. Venue abbreviation consistency. ~ before \cite{}.
Missing figure files. Dummy/placeholder figures. Absolute paths. Label placement (inside \caption{} or before, NOT after). Unreferenced figures/tables. Missing size specs. \hline vs booktabs. Decimal alignment (siunitx S columns).
TODO, XXX, FIXME, TBD, ???, [CITE], [REF], [FILL], \tobeupdated. Inconsistent method/dataset/metric naming. Default template content. \journalVersion{} non-empty. deprecated/ folder still \input-ted. Double spaces. Missing ~. Hard-coded \\ in prose. \vspace{-Xmm} accumulated total.
Unit formatting (\,). Thousand separators. \ie/\eg macros (define if missing). state-of-the-art noun vs adjective. Acronym first-use in abstract AND body separately. Sentences starting with "And", "But", "Or". "etc." in formal prose. Figure reference order vs document position.
These are summaries. Read
references/02_paper_proofreading.mdfor full rules with examples.
Subject-verb agreement. Articles. Prepositions. Tense consistency (present for facts/contributions, past for experiments). Related Work tense (present preferred; flag only mixing). Oxford comma. Comma splices.
Typos/spelling → CRITICAL. Duplicate words → CRITICAL. Nominalization (prefer direct verbs). Redundant expressions ("In order to"→"To"). Citation-as-noun (use Author~\etalcite{#}). Circular descriptions. Verb choice ("suggest"→"propose"/"investigate"/"present").
Every "significantly" needs statistical test or quantitative alternative. "outperform"/"superior"/"state-of-the-art" verified across all metrics. Causal logic gaps. Variables used before definition. Claims in figure captions. "Only a few works..." contradicted by citation list. Scope-limiting language without justification.
Introduction: dedicated contribution paragraph. Related Work: standalone section (CRITICAL if merged into intro), 15-25 papers for 6-8 page paper, at least one comparison to this work (MAJOR if none). Method: equation references with narrative context. Experiments: each opens with WHY/WHAT/HOW statement, claim coverage verified, most impressive first. General: no repetition across sections, no verbatim intro→conclusion copy, ablation cross-references method section.
Captions: grammatically complete, self-contained (abbreviations defined, baselines cited with \cite{}, datasets named), no body text duplication, period at end. Figures: font size consistent with caption (\footnotesize), tick labels legible, thousand separators in axis ticks, no excessive whitespace. Tables: bold/underline convention defined in caption, consistent metric names, units in headers. Reference order: sequential ascending. Quantitative consistency: text numbers match table/figure data exactly → CRITICAL if mismatch.
Thin space before units (\,). Thousand separators for integers ≥1000 (NOT decimals). Consistent figure reference style. \ie/\eg macros. et al. with period. ~ before \cite/\ref. state-of-the-art consistency.
Abstract: WHY(1-2 sentences)→PROBLEM(1 sentence)→HOW&WHAT(~3 sentences)→RESULTS(1 sentence). Acronyms expanded within abstract. Zero \cite{} → CRITICAL. Single paragraph → CRITICAL if broken. Conclusion: not verbatim abstract restatement. Limitations acknowledged. Future work specific with grounding sentence.
Symbol overload detection (build symbol table across all equations). Different symbols for same concept. Vector/matrix boldface (\mathbf{}). Greek vectors (\boldsymbol{}). Coordinate frame notation consistent subscript order. Superscript semantic consistency. Term capitalization consistency.
Rule 1: Compound adjective before noun → hyphenate (real-time system). Rule 2: -ly adverb + adjective → NEVER hyphenate (tightly coupled, NOT tightly-coupled). Common robotics/CV patterns: "end-to-end", "state-of-the-art", "deep learning" (noun phrase, not hyphenated as adjective).
## LaTeX Workspace Audit Results
**Files inspected**: [list all files read]
**Total issues**: N
### Issues by File
**`main.tex`**
[1] L.53 Description | Impact / Suggested fix
**`shortcuts.tex`**
[2] L.14 Description | Impact / Suggested fix
### Issues by Severity
CRITICAL
[N] File — brief summary
MAJOR
[N] File — brief summary
MINOR
[N] File — brief summary
STYLE
[N] File — brief summary
### Infrastructure Suggestions
[Workspace-level recommendations not tied to a specific file]
## Paper Proofreading Results
Paper quality: GOOD / NEEDS REVISION / MAJOR REVISION
| Type | Count |
|----------|-------|
| CRITICAL | |
| MAJOR | |
| MINOR | |
| STYLE | |
Most common problems:
- ...
### Issues by File
[Same format as workspace audit]
### Caption Review
| # | Figure/Table | Issue | Suggestion |
### LaTeX Formatting Patterns
| # | Pattern | Example Found | Suggested Fix |
### Optional Polishing Suggestions
[High-level structural improvements only]
After outputting ALL findings, always end with:
---
**Phase 1 complete.** All issues listed above with numbers [1], [2], [3]...
Reply with one of:
- `fix safe` — typos and clear grammar errors only
- `fix all critical` — CRITICAL issues only
- `fix 1, 3, 5` — specific issue numbers
- `fix all` — all issues
- `discard 3, 7` — skip specific issues
**No files will be modified until you confirm.**