research-project-router
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseResearch Project Router
研究项目路由工具
Use this skill first when a request spans more than one research workflow. The
goal is to route deliberately and preserve project state.
当请求涉及多个研究工作流时，请首先使用此技能。目标是精准路由并保留项目状态。
Read First
必读内容
references/repository-contract.mdreferences/output-contracts.md
references/repository-contract.mdreferences/output-contracts.md
Routing Table
路由表
| User intent | Prefer skill |
|---|---|
| Turn ideas, gaps, hunches, and vague goals into research questions, contribution claims, scope, and evidence plans | |
| Choose CS venue, track, audience, review criteria, or paper positioning | |
| Add PDFs, papers, proposals, archives, datasets, or web pages | |
| Convert PDFs, scans, DOCX, tables, or figures to Markdown/assets | |
| Normalize BibTeX, identifiers, citation keys, missing/unused citations | |
| Find related work, build SOTA, compare papers, discover gaps | |
| Run reproducible systematic/scoping review and PRISMA-style screening | |
| Check citations, claims, numbers, evidence, bibliography | |
| Design/audit CS evaluation, baselines, metrics, ablations, validity | |
| Review a manuscript/proposal as a severe peer reviewer | |
| Draft, revise, polish, or review a paper/proposal | |
| Respond to reviews, rebuttal, revision plan, camera-ready changes | |
| Prepare artifact evaluation, code/data release, reproducibility package | |
| Migrate a bad repo, archive, or notebook pile into the project structure | |
| Understand or reproduce an external research repo | |
| Run experiments, benchmark ideas, maintain run ledger | |
| Analyze datasets, statistics, tables, and figures | |
| Build research dashboards or interactive demos | |
| Assess human-subject, privacy, scraping, license, or sharing risk | |
| Fix folder structure, docs, wiki, agent instructions | |
| Onboard or audit the skill/MCP capability stack | |
| Select, configure, smoke-test, and document arXiv, OpenAlex, Zotero, Overleaf, Scholar, and scholarly metadata MCP tools | |
| Review or improve the skill package itself | |
| 用户意图 | 首选技能 |
|---|---|
| 将想法、研究空白、直觉和模糊目标转化为研究问题、贡献声明、研究范围和证据规划 | |
| 选择计算机科学（CS）会议/期刊、赛道、受众、评审标准或论文定位 | |
| 添加PDF、论文、提案、档案、数据集或网页 | |
| 将PDF、扫描件、DOCX、表格或图表转换为Markdown/资源文件 | |
| 标准化BibTeX、标识符、引用键、缺失/未使用的引用 | |
| 查找相关研究、构建SOTA、对比论文、发现研究空白 | |
| 开展可复现的系统性/范围性综述及PRISMA风格筛选 | |
| 检查引用、声明、数据、证据、参考文献 | |
| 设计/审核计算机科学（CS）评估方案、基线、指标、消融实验、有效性 | |
| 以严苛同行评审者身份评审手稿/提案 | |
| 撰写、修改、润色或评审论文/提案 | |
| 回复评审意见、准备反驳信、修订计划、终稿修改 | |
| 准备成果物评估、代码/数据发布、可复现性包 | |
| 将混乱的仓库、档案或笔记本迁移至规范项目结构 | |
| 理解或复现外部研究仓库 | |
| 运行实验、测试基准想法、维护实验记录 | |
| 分析数据集、统计数据、表格和图表 | |
| 构建研究仪表盘或交互式演示 | |
| 评估人类受试者、隐私、数据爬取、许可或共享风险 | |
| 修复文件夹结构、文档、维基、Agent指令 | |
| 接入或审核技能/MCP能力栈 | |
| 选择、配置、冒烟测试并记录arXiv、OpenAlex、Zotero、Overleaf、Scholar及学术元数据MCP工具 | |
| 评审或改进技能包本身 | |
Default Procedure
默认流程
- Identify the research artifact being touched: source, claim, data, code, experiment, paper, or project structure.
- Check whether the repository follows the contract. If not, adapt conservatively and note the gap.
- Route to the narrowest skill that can complete the task.
- If several workflows are required, order them as: design/position -> venue/frame -> ingest/convert -> synthesize/review -> evaluate/audit -> write -> rebuttal/artifact -> reproduce/experiment -> update project/wiki.
- Update or the appropriate output ledger when the task changes project knowledge.
wiki/log.md
- 确定涉及的研究成果物：来源、声明、数据、代码、实验、论文或项目结构。
- 检查仓库是否遵循约定。若未遵循，保守调整并记录差距。
- 路由至能完成任务的最细分技能。
- 若需要多个工作流，按以下顺序执行：设计/定位 -> 会议/框架 -> 导入/转换 -> 综合/评审 -> 评估/审核 -> 撰写 -> 反驳/成果物 -> 复现/实验 -> 更新项目/维基。
- 当任务改变项目知识时，更新或相应的输出记录。
wiki/log.md
Do Not
禁止事项
- Treat chat history as the project record.
- Skip ingestion and write a SOTA directly from raw PDFs.
- Let exploratory experiment outputs overwrite trusted reproduction outputs.
- Make a claim stronger than the source, dataset, or run supports.
- 将聊天历史作为项目记录。
- 跳过导入步骤直接从原始PDF生成SOTA内容。
- 让探索性实验输出覆盖可信的复现输出。
- 提出超出来源、数据集或实验结果支持力度的声明。