计算机论文草稿终审与分层重写工作流。先逻辑后语言：检查全文章节结构 → 章节内部逻辑 → 跨章节一致性 → 写作风格 → 摘要逐句功能 → 图表 → 投稿前 sanity check。问题按 P0/P1/P2 分级。适用于 NeurIPS/ICML/ICLR/KDD/CCS/USENIX/SIGMOD/VLDB/IEEE Transactions 等顶会顶刊。Use when 用户请求论文终审、逻辑诊断、分层重写、投稿前检查，或在草稿已成形阶段进行系统性修改。
Paper editing with built-in style constraints for academic writing quality, including anti-AI detection, LaTeX validation, and figure standards
Address reviewer/editor comments by number with localized, traceable edits. Use when the user provides numbered reviewer concerns, editor proof queries, or revision lists and asks to "address comment N", "fix W1-W4", "respond to editor queries", etc. Complements review-response (rebuttal drafting) and paper-edit (style enforcement).
Use when ingesting arXiv papers into a project-local Zotero library, building a fresh Zotero collection tree for a research project, the Zotero free-tier 300 MB cloud quota is full ("File would exceed quota" / 413 Request Entity Too Large on attachment upload), the user wants PDFs to live alongside project files rather than in Zotero cloud, or the user asks to "把论文加进 Zotero / 建立 Zotero 分类 / set up related_work folder with Zotero".
Write publication-ready ML/AI/Systems papers for NeurIPS, ICML, ICLR, ACL, AAAI, COLM, OSDI, NSDI, ASPLOS, SOSP. Use when drafting papers from research repos, structuring arguments, verifying citations, or preparing camera-ready submissions. Includes LaTeX templates, reviewer guidelines, and citation verification workflows.
Standardized tmux experiment monitoring with progress parsing, CSV log reading, and status reporting
从论文中提取图片，优先从arXiv源码包获取真正的论文图
Roleplay as the "Armchair Executioner" — a maximally destructive, minimally accountable peer reviewer. Use when the user asks for a peer review, wants to stress-test a paper, or invokes /mean-reviewer.