重要前提
安装AI Skills的关键前提是:必须科学上网,且开启TUN模式,这一点至关重要,直接决定安装能否顺利完成,在此郑重提醒三遍:科学上网,科学上网,科学上网。查看完整安装教程 →
research-ideation by evoscientist/evoskills
npx skills add https://github.com/evoscientist/evoskills --skill research-ideation一个目标驱动的工作流,用于寻找重要的研究问题、设计新颖的解决方案,并通过结构化的论文阅读建立深入的领域理解。
按顺序遵循以下五个步骤。每一步都建立在前一步的基础上。
从一个兼具科学价值和实用价值的目标开始。目标应足够宏大,足以支撑多篇论文,但又足够具体,能够指导日常决策。
提问:"这个研究方向的最终形态是什么?如果这个问题被完全解决,世界会是什么样子?"
通过构建两个互补的树来绘制领域图谱:
完整的构建方法和四种新颖性类型,请参见 references/literature-tree.md。
寻找具有真正研究空间的任务。关键问题是:"这个问题值得解决吗,还是已经有一个成熟的解决方案占领了这个领域?"
使用成熟解决方案检查(4个级别)来决定是继续推进还是切换问题。积极寻找新的失败案例,而不是在已知基准上改进。
完整的选择框架,请参见 references/problem-selection.md。
新颖的技术是现有方法的创造性组合,而不是简单的串联。使用两种设计模式:
广告位招租
在这里展示您的产品或服务
触达数万 AI 开发者,精准高效
完整的设计方法和知识蒸馏流程,请参见 references/solution-design.md。
在代表性数据上运行实验。利用结果来完善你的理解。如果方法失败,则带着从失败中获得的新知识返回到步骤 3 或步骤 4。
产出物:研究方向摘要(问题陈述、建议方法、新颖性声明、关键风险)—— 这将成为 idea-tournament 或 paper-planning 的输入。
当实验未按预期工作时,请参见 experiment-craft 技能以进行系统性调试。
在进行常规构思之前,优先考虑这些规则:
使用论文解析树将阅读转变为结构化的问答。三个深度级别:
| 级别 | 目标 | 阅读后你能做什么 |
|---|---|---|
| 1. 技术性 | 理解所有细节和术语 | 复现方法;解释每个组件 |
| 2. 分析性 | 知道它解决了什么问题以及为何采用此方法 | 解释论文的动机和设计选择 |
| 3. 上下文性 | 知道它在文献树中的位置 | 更新你的领域图谱;产生新的研究问题 |
为你阅读的每一篇论文撰写结构化摘要。使用位于 assets/paper-summary-template.md 的模板。
完整的阅读方法和习惯养成指导,请参见 references/paper-reading.md。
当你有一个研究方向但想探索多个具体方法时,在规划之前,将其传递给 idea-tournament 进行树状结构生成和 Elo 排名。
当构思完成 —— 你有了问题、建议的解决方案和支持文献 —— 将这些产出物传递给 paper-planning:
| 产出物 | 来源步骤 | 用途 |
|---|---|---|
| 研究目标和范围 | 步骤 1 | 故事设计(任务定义) |
| 文献树(新颖性 + 挑战-洞察) | 步骤 2 | 相关工作映射,新颖性声明 |
| 问题陈述和动机 | 步骤 3 | 引言动机段落 |
| 解决方案草图和设计原理 | 步骤 4 | 方法部分规划 |
| 需要解决的关键失败案例 | 步骤 3 | 实验规划(压力测试) |
| 相关的先前工作及其局限性 | 步骤 2 | 基线选择,比较设计 |
| 主题 | 参考文件 | 何时使用 |
|---|---|---|
| 文献树构建 | literature-tree.md | 绘制研究领域图谱 |
| 问题选择 | problem-selection.md | 评估一个问题是否值得解决 |
| 解决方案设计 | solution-design.md | 设计新颖方法 |
| 论文阅读方法 | paper-reading.md | 有效阅读论文 |
| 论文摘要模板 | paper-summary-template.md | 撰写结构化论文笔记 |
每周安装量
69
代码仓库
GitHub 星标数
105
首次出现
9 天前
安全审计
安装于
cursor69
gemini-cli69
github-copilot69
amp69
cline69
codex69
A goal-driven workflow for finding important research problems, designing novel solutions, and building deep field understanding through structured paper reading.
Follow these five steps in order. Each step builds on the previous one.
Start with a goal that has both scientific and practical value. The goal should be ambitious enough to sustain multiple papers, but concrete enough to guide daily decisions.
Ask: "What is the ultimate form of this research direction? What would the world look like if this problem were fully solved?"
Map the field by constructing two complementary trees:
See references/literature-tree.md for the full construction method and four types of novelty.
Find tasks with genuine research space. The key question: "Is this problem worth solving, or has a well-established solution already claimed this territory?"
Use the well-established solution check (4 levels) to decide whether to proceed or switch problems. Actively seek new failure cases rather than improving on known benchmarks.
See references/problem-selection.md for the full selection framework.
Novel techniques are creative combinations of existing methods, not simple concatenations. Use two design patterns:
See references/solution-design.md for the full design methodology and knowledge distillation pipeline.
Run experiments on representative data. Use results to refine your understanding. If the approach fails, return to Step 3 or Step 4 with updated knowledge from the failure.
Output artifacts : Research direction summary (problem statement, proposed approach, novelty claim, key risks) — this becomes the input to idea-tournament or paper-planning.
See the experiment-craft skill for systematic debugging when experiments don't work as expected.
Prioritize these rules before regular ideation:
Turn reading into structured Q&A using a paper parsing tree. Three levels of depth:
| Level | Goal | What You Can Do After |
|---|---|---|
| 1. Technical | Understand all details and terminology | Reproduce the method; explain each component |
| 2. Analytical | Know what problem it solves and why this approach | Explain the paper's motivation and design choices |
| 3. Contextual | Know its position in the literature tree | Update your field map; generate new research questions |
Write a structured summary for every paper you read. Use the template at assets/paper-summary-template.md.
See references/paper-reading.md for the full reading methodology and habit-building guidance.
When you have a research direction but want to explore multiple concrete approaches, pass to idea-tournament for tree-structured generation and Elo ranking before planning.
When ideation is complete — you have a problem, a proposed solution approach, and supporting literature — pass these artifacts to paper-planning:
| Artifact | Source Step | Used By |
|---|---|---|
| Research goal and scope | Step 1 | Story design (task definition) |
| Literature tree (novelty + challenge-insight) | Step 2 | Related work mapping, novelty claims |
| Problem statement and motivation | Step 3 | Introduction motivation paragraphs |
| Solution sketch and design rationale | Step 4 | Method section planning |
| Key failure cases to address | Step 3 | Experiment planning (stress tests) |
| Relevant prior work and their limitations | Step 2 | Baseline selection, comparison design |
| Topic | Reference File | When to Use |
|---|---|---|
| Literature tree construction | literature-tree.md | Mapping a research field |
| Problem selection | problem-selection.md | Evaluating whether a problem is worth solving |
| Solution design | solution-design.md | Designing a novel approach |
| Paper reading methodology | paper-reading.md | Reading papers effectively |
| Paper summary template | paper-summary-template.md | Writing structured paper notes |
Weekly Installs
69
Repository
GitHub Stars
105
First Seen
9 days ago
Security Audits
Gen Agent Trust HubPassSocketPassSnykWarn
Installed on
cursor69
gemini-cli69
github-copilot69
amp69
cline69
codex69
Python PDF处理教程:合并拆分、提取文本表格、创建PDF文件
66,200 周安装
小红书数据分析平台 Xinhong Data:账户增长、内容表现、达人合作、行业趋势深度洞察
62 周安装
KeyID Agent Kit MCP:为AI代理提供免费电子邮件地址和27种工具
570 周安装
Medusa 管理员用户创建指南:使用 new-user Skill 快速配置后台账户
562 周安装
数据分析验证工具 - 分享前审查分析准确性、方法论和潜在偏见
581 周安装
UI/UX设计师专家技能:精通设计系统、无障碍设计与现代工作流程
569 周安装
Grimmory 自托管图书馆管理器:支持 EPUB/PDF/MOBI/CBZ 的 Docker 部署方案
588 周安装