学习机会:一个用于刻意技能发展的 Claude Code 和 Codex 技能
打造你的专业能力,而不仅仅是你的项目。
该技能采用一种自适应的“动态教科书”方法,帮助你在进行智能体编码的同时,融入基于科学原理的专业能力构建练习。
当你完成架构性工作(新文件、模式变更、重构)后,Claude 会提供可选的、时长 10-15 分钟的学习练习,这些练习基于循证学习科学。这些练习运用了预测、生成、检索练习和间隔重复等技术,为你提供来自你自身项目工作中的半成品示例。
与 Learning-Goal 技能搭配使用效果很好,该技能通过一种名为“心理对照与执行意图”(MCII)的循证练习,引导你进行半结构化的、互动式的学习目标设定。
安装
Codex
该仓库也是一个 Codex 插件市场。要通过 GitHub 添加它:
codex plugin marketplace add https://github.com/DrCatHicks/learning-opportunities.git
如需从检出目录进行本地开发:
codex plugin marketplace add /path/to/learning-opportunities
Codex 市场包括:
- learning-opportunities — 核心学习练习技能
- learning-opportunities-auto — 可选的提交后提示钩子
- orient — 仓库导向生成器
Claude Code
该仓库是一个 Claude Code 插件市场。要安装:
-
添加市场:
/plugin marketplace add https://github.com/DrCatHicks/learning-opportunities.git -
安装插件:
/plugin install learning-opportunities@learning-opportunities -
重启 Claude Code 以激活
有关 Claude Code 插件的更多信息,请参阅插件文档。
自动提示(可选)
Linux 和 macOS 用户可以同时安装 learning-opportunities-auto 和 learning-opportunities,以便 Claude 在每次 git 提交后自动考虑提供练习。Windows 用户也可以使用——只需进行一些设置。
获取仓库导向课程(可选)
如果你正在学习一个新仓库,你可以使用 orient 技能创建一个包含建议课程的 orientation.md 文件。这种导向方法应用了关于程序理解和代码库导航的实证研究策略——包括专家开发者如何策略性地采样代码库,而不是通读所有内容。请参阅 orient 参考书目以获取完整的来源列表。
安装 orient 插件:
/plugin install orient@learning-opportunities
导航到你想要了解的仓库,然后调用 orient 技能(作为默认技能或通过其他方式)。
/orient
或者使用 Simon Willison 的展示工具
/orient showboat
然后,使用 orient 参数调用 learning-opportunities,即可获得两项课程,它们将引导你了解该仓库的核心功能。
/learning-opportunities orient
为何你可能想尝试这项技能
AI 编程工具可能带来特定风险,即通过引入低效的学习习惯,降低用户参与学习的积极性。基于若干有科学依据的基础学习原则,可以预见这些影响:
- 生成效应:接受生成的代码而减少自己编写代码,可能会跳过构建理解所需的主动处理过程。
- 流利错觉:整洁的生成代码可能让人误以为自己理解得比实际更深;同样,通过搜索轻易获取的知识,也可能助长知识错觉和心智模型更完整的错觉。
- 间隔效应:机器的速度可能推动用户持续填鸭式学习和长时间的生产会话,而缺乏有助于长期记忆的学习节奏、反思和间隔安排。
- 元认知:快速的工作流程往往没有留出空间来监控学习、发展图式表征,以及用户在使用新技术时对自身相对专业水平和知识的感知。
- 测试与检索:智能体模型倾向于给出完整答案,这可能导致用户减少从自我测试和检索新知识的具体组成部分中获益的机会,而后者恰恰能强化记忆。
SKILL.md 中的技术旨在通过重新引入以下内容来应对这些风险:
- 主动生成(预测、解释、草图)
- 检索练习(检查点、复述教学、自我测试)
- 刻意停顿(间隔安排、反思)
- 显性元认知(自我评估、识别差距)
这项技能会打断那种模式,提醒你考虑投入时间进行反思和学习。它引入了一种与 Claude 交互的“不同模式”,这种模式在感觉上会刻意不同于高度流畅、快速的智能体编码,目的是帮助你反思和探索你生成的作品。这项技能对于正在尝试使用智能体编码开发独立项目、且项目涉及多种不熟悉的语言、技术或架构模式的用户来说,可能特别有用。
工作原理
在你完成重要工作后(你可以自行定义,但我建议:创建新文件或模块、数据库模式变更、架构决策或重构、实现不熟悉的模式、以及开发过程中用户提出“为什么”问题的任何工作。关键思想是在你的个人工作流中找到一个学习机会最有利的时机),Claude 会询问:
“你想做一个关于 [主题] 的快速学习练习吗?大约需要 10-15 分钟。”
如果你接受,Claude 会引导你完成一个互动练习。一个关键的设计原则是:Claude 会暂停并等待你的输入,而不是回答它自己的问题。这可能会让人感到沮丧,但这能对抗 Claude 总是提供完整答案的默认倾向,并鼓励你付出脑力努力和学习。你可能会遇到 Claude 默认提供完整答案的情况,并需要针对此进行设计;如果你在自己的工作流中发现了你认为会普遍适用于他人的陷阱或冲突,请随时告诉我,以便我将其纳入技能中进行改进(例如,我了解到我们需要抑制提示建议)。
练习类型
- 预测 → 观察 → 反思:你预期会发生什么?现在让我们看看。有什么让你感到意外?
- 生成 → 比较:在查看实现之前,先草拟你会如何处理这个问题。
- 追踪路径:一步步地走通执行过程,预测每一步的转换。
- 调试这个:这里可能会出什么问题,以及为什么?
- 反向教学:就像在指导一位新入职的开发者一样,解释这个组件。
- 检索回顾:在一次会话开始时,你从上一次记住了什么?
不会建议学习机会的情况……
目前建议设置两种抑制条件,可根据你的工作流程需求进行调整。在以下情况下,Claude 不会提示学习机会:
- 你已在本次会话中拒绝过一次练习
- 你已在本次会话中完成了 2 次练习
背后的科学原理与参考资料
这些练习借鉴了学习科学中经过充分验证的发现,以及关于典型学习者误解的大量研究。设计选择还参考了与开发者的多次定性访谈,内容涉及在快速智能体编程过程中,他们在自身学习与发展方面感到最沮丧、最担忧或最困难的方面。
请参阅 PRINCIPLES.md 文件,其中包含详细解释,可帮助你开发新的练习类型,或了解更多有助于自身学习的策略。
衡量这一点:一份让团队实验可见的轻量级操作手册
如果你正与团队一起尝试这项技能,可以加入轻量级的前/后测量,使实验在你的组织中更可见、更受重视。
MEASURE-THIS.md 是一份配套操作手册,内容包括:
- 一套精选的、经过验证的调查问卷题目,源自我们关于开发者成长与 AI 技能威胁的同行评审研究,可直接复制到 Google 表单或团队频道中使用
- 关于如何处理(以及不处理)结果的指导,包括为什么方差与平均值同样重要,以及我对这类测量方式的一些思考边界
- 一份“团队宣传”模板——一个填空式段落,用于将你的实验包装成发送给领导层的邮件,内容基于真实研究
- Claude.md 中的统计严谨性提示:如果你想使用 Claude 进行更复杂的分析,这些提示有助于防范常见的 AI 辅助统计错误
这些测量工具在 CC-BY-SA 4.0 许可下免费开放使用。如需完整的测量工具集和设计说明,请参阅 AI 技能威胁开放获取测量工具补充材料和开发者成长开放获取测量工具补充材料。
自定义
这项技能可以进行大幅度的优化和调整。你可能想要:
- 包含关于自身技术专长和现有知识的信息,以便从合适的水平开始练习(例如,已知的编程语言、学习目标)
- 提示 Claude 将学习机会中的见解纳入你的项目 Claude.md 文件中。
- 调整工作流的触发条件。
- 在练习中添加特定于项目的示例。
- 更改每次会话中练习数量的软上限。
- 添加特定领域的检索检查问题。
- 探索添加评估检查,以评估该技能在多大程度上成功执行了其指令。
背景
这项技能是基于学习科学开发的,并参考了对多位软件开发专业人士进行的定性访谈,访谈内容涉及他们对智能体编程的担忧。这是我关于开发者在 AI 辅助工作流中成长与技能发展的开放科学实证研究的一部分。在我对数以千计开发者的研究中,我还发现,强烈的学习价值观和学习投入能够预测,当开发者想象需要适应智能体编程时,他们感受到的威胁、担忧和焦虑会更少。学习文化还与团队整体效能的提升相关,而不仅仅是个人生产力的提升。
我很想知道你是否喜欢这个,以及你学到了什么!分享开放科学资源有助于像我这样的研究人员创造更多东西来帮助软件团队。我始终感谢任何公开的提及或分享,这能帮助更多人了解软件团队心理学。欢迎订阅我的通讯《为人类而战》(Fight for the Human),获取最新动态和更多关于软件团队心理学的内容。
卡特·希克斯博士(Dr. Cat Hicks)
我是一位研究软件团队与技术工作的心理科学家,同时也是一位作家、公开演讲者、研究架构师和实证干预主义者。我致力于建立激进的研究团队,为那些人人都在问、但很少有人真正收集证据来回答的问题提供答案。
- 网站:drcathicks.com
- 软件团队与工程领导力咨询:catharsisinsight.com
- 即将出版的新书:《软件团队心理学》(The Psychology of Software Teams,2026年)
指导者(Orient):
迈克尔·穆拉基博士(Dr. Michael Mullarkey)
我是一名机器学习工程师,曾经是治疗师和社会科学研究人员。我一直在思考如何利用智能体 AI 来帮助人们学习技能,另一个例子可以参考 blendtutor。
- 网站
来源
- Bjork, R. A., Dunlosky, J., & Kornell, N. (2013). 自我调节学习:信念、技巧与错觉。《心理学年度评论》,64(1),417-444。
- Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). 利用有效学习技巧提升学生学习效果:来自认知与教育心理学的有前景方向。《公共利益中的心理科学》,14(1),4-58。
- Ericsson, K. A., Hoffman, R. R., & Kozbelt, A. (编). (2018). 《剑桥专长与卓越表现手册》。剑桥大学出版社。
- Giebl, S., Mena, S., Storm, B. C., Bjork, E. L., & Bjork, R. A. (2021). 先回答还是先谷歌?以增强而非损害后续所需信息记忆的方式使用互联网。《心理学教学与学习》,20(1),58-75。
- Hicks, C. M., Lee, C. S., & Foster-Marks, K. (2025年3月15日). 《新开发者:AI技能威胁、身份转变与开发者在向AI辅助软件开发过渡中的蓬勃发展》。https://doi.org/10.31234/osf.io/2gej5_v2
- Kalyuga, S. (2007). 专长反转效应及其对个性化教学的意义。《教育心理学评论》,19(4),509-539。
- Kang, S. H. (2016). 间隔重复促进高效学习:对教学的政策启示。《行为与脑科学政策洞察》,3(1),12-19。
- Kornell, N. (2009). 使用闪卡优化学习:间隔学习比突击学习更有效。《应用认知心理学:应用记忆与认知研究学会官方期刊》,23(9),1297-1317。
- Murphy, D. H., Little, J. L., & Bjork, E. L. (2023). 在教育中利用测试作为学习工具——而非仅用于评估的价值。《教育心理学评论》,35(3),89。
- Roediger III, H. L., & Karpicke, J. D. (2006). 测试记忆的力量:基础研究及其对教育实践的启示。《心理科学视角》,1(3),181-210。
- Rohrer, D., & Taylor, K. (2007). 数学问题的交错排列能提升学习效果。《教学科学》,35(6),481-498。
- Skulmowski, A., & Xu, K. M. (2022). 理解数字与在线学习中的认知负荷:外部认知负荷的新视角. Educational psychology review, 34(1), 171-196.
- Soderstrom, N. C., & Bjork, R. A. (2015). 学习与表现:一项整合性综述. Perspectives on Psychological Science, 10(2), 176-199.
- Sweller, J., & Cooper, G. A. (1985). 使用示例解题作为代数学习中问题求解的替代方法. Cognition and instruction, 2(1), 59-89.
- Tankelevitch, L., Kewenig, V., Simkute, A., Scott, A. E., Sarkar, A., Sellen, A., & Rintel, S. (2024, 5月). 生成式AI的元认知需求与机遇. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems (pp. 1-24).
- Hicks, C. (2025). AI自行车的认知头盔:第一部分. Fight for the Human. https://www.fightforthehuman.com/cognitive-helmets-for-the-ai-bicycle-part-1/
许可协议
本作品采用知识共享署名 4.0 国际许可协议进行许可。
关于
一个用于AI辅助编程中进行刻意技能培养的Claude或Codex技能
资源
许可协议
CC-BY-4.0 许可协议
星标数
2.2k
关注者
复刻数
发布版本
软件包
贡献者
语言
- Shell 100.0%
Learning Opportunities: A Claude Code and Codex Skill for Deliberate Skill Development
Build your expertise, not just your projects.
This skill uses an adaptive "dynamic textbook" approach to help you integrate science-based expertise building exercises while doing agentic coding.
When you complete architectural work (new files, schema changes, refactors), Claude offers optional 10-15 minute learning exercises grounded in evidence-based learning science. The exercises use techniques like prediction, generation, retrieval practice, and spaced repetition to provide you with semi-worked examples from across your own project work.
Pairs well with Learning-Goal, a skill that guides you through semi-structured, interactive learning goal-setting using the technique of Mental Contrasting with Implementation Intentions (MCII), an evidence-based exercise.
Installation
Codex
This repository is also a Codex plugin marketplace. To add it from GitHub:
codex plugin marketplace add https://github.com/DrCatHicks/learning-opportunities.git
For local development from a checkout:
codex plugin marketplace add /path/to/learning-opportunities
The Codex marketplace includes:
learning-opportunities— the core learning exercise skilllearning-opportunities-auto— optional post-commit prompting hookorient— repo orientation generator
Claude Code
This repository is a Claude Code plugin marketplace. To install:
-
Add the marketplace:
/plugin marketplace add https://github.com/DrCatHicks/learning-opportunities.git -
Install the plugin:
/plugin install learning-opportunities@learning-opportunities -
Restart Claude Code to activate
For more on Claude Code plugins, see the plugin documentation.
Automatic Prompting (Optional)
Linux and macOS users can install learning-opportunities-auto alongside learning-opportunities to have Claude automatically consider offering an exercise after each git commit. Windows users can use it too — a little setup is required.
Get Repo Orientation Lessons (Optional)
If you're learning a new repo you can create an orientation.md file with suggested lessons using the orient skill. The orientation approach applies strategies from empirical research on program comprehension and codebase navigation — including how expert developers sample codebases strategically rather than reading exhaustively. See the orient bibliography for the full source list.
Install the orient plugin:
/plugin install orient@learning-opportunities
Navigate to the repo you want to orient yourself to, and call the orient skill either as default
/orient
Or using Simon Willison's showboat tool
/orient showboat
Then call learning-opportunities with the orient argument to get offered two lessons that will orient you to core features of the repo
/learning-opportunities orient
Why You Might Want to Experiment with This Skill
AI coding tools can create specific risks for decreasing users' engagement in learning by introducing inefficient learning habits. These effects can be anticipated based on several foundational science-backed learning principles:
- Generation effect: Accepting generated code and decreasing generating one's own code can skip the active processing that builds understanding.
- Fluency illusion: Clean generated code can be perceived as more understood than it truly is; likewise, easily accessible knowledge from search can promote the illusion of knowledge and the illusion of more complete mental models.
- Spacing effect: Machine velocity can push users toward constant cramming and long production sessions without the cadence, reflection and spacing of learning that leads to longer-term retention.
- Metacognition: Fast workflows often don't leave room to monitor learning and develop schema representation as well as a user's sense of their own level of relative expertise and knowledge when working with novel technology.
- Testing and retrieval: Agentic models push toward giving complete answers, which could result in users taking fewer opportunities to benefit from self-testing and retrieving specific components of new knowledge, which strengthens retention.
The techniques in SKILL.md are designed to counteract these risks by reintroducing:
- Active generation (predictions, explanations, sketches)
- Retrieval practice (check-ins, teach-it-back, self-testing)
- Deliberate pauses (spacing, reflection)
- Explicit metacognition (self-assessment, gap identification)
This skill interrupts that pattern by reminding you to consider investing in reflection and learning. It introduces a different "mode" of interacting with Claude, which will intentionally feel different than highly fluent and fast agentic coding in the service of helping you reflect and explore your generated work. This skill may be particularly useful for users who are experimenting with developing discrete projects with agentic coding that involve multiple unfamiliar languages, techniques, or architectural patterns.
How It Works
After you complete significant work (which you can self-define, but I've suggested: creating new files or modules, database schema changes, architectural decisions or refactors, implementing unfamiliar patterns, any work where the user asked "why" questions during development. The key idea is to find a moment in your personal flow where a learning opportunity is most beneficial) Claude will ask:
"Would you like to do a quick learning exercise on [topic]? About 10-15 minutes."
If you accept, Claude runs you through an interactive exercise. A key design principle: Claude pauses and waits for your input rather than answering its own questions. This can feel frustrating, but this pushes against Claude's default to always provide the full answer and encourages your own mental effort and learning. You may encounter and need to design against Claude's defaults to provide the complete answer; please feel free let me know if you find gotchas or conflicts in your own workflow that you think will generalize to others so that I can incorporate in the Skill to improve this (e.g., I learned we needed to suppress prompt suggestions).
Exercise Types
- Prediction → Observation → Reflection: What do you expect to happen? Now let's see. What surprised you?
- Generation → Comparison: Sketch how you'd approach this before seeing the implementation
- Trace the path: Walk through execution step by step, predicting each transition
- Debug this: What would go wrong here, and why?
- Teach it back: Explain this component as if onboarding a new developer
- Retrieval check-in: At the start of a session, what do you remember from last time?
Will Not Suggest Learning Opportunities When...
Two suppression conditions are currently suggested which can be adapted to your workflow needs. Claude will not prompt learning opportunities when:
- You've already declined an exercise this session
- You've completed 2 exercises this session
The Science Behind It & Resources
The exercises draw from well-established findings in learning science, along with substantive research on typical learner misconceptions. Design choices also draw from multiple qualitative interviews with developers about what aspects of rapid agentic coding they find most frustrating, worrisome, or difficult when it comes to their own learning and development.
See PRINCIPLES.md for detailed explanations which can help you develop new exercise types or simply learn more about strategies to help your own learning.
Measure This, A Lightweight Playbook for Making A Team Experiment Visible
If you're trying this skill with your team, you can layer on a lightweight pre/post measurement to make the experiment more visible and valued in your organization.
MEASURE-THIS.md is a companion playbook that includes:
- A curated set of validated survey items from our peer-reviewed research on developer thriving and AI skill threat, ready to copy into a Google Form or team channel
- Guidance on what to do (and not do) with your results including why variance matters as much as averages, and some guardrails on how I think about these types of measures
- A "team boast" template a fill-in-the-blank paragraph for packaging your experiment into an email to leadership, grounded in real research
- Claude.md nudges for statistical rigor if you want to use Claude for more complex analysis, these nudges help guard against common AI-assisted statistical mistakes
The measures are free and open access under a CC-BY-SA 4.0 license. For the full set of measures and design notes, see the AI Skill Threat open access measures supplement and the Developer Thriving open access measures supplement.
Customization
This skill can be significantly refined and adapted. You might want to:
- Include information about your own technical expertise and existing knowledge to start the exercises at the right level (e.g. known languages, learning goals)
- Prompt Claude to include insights from the learning opportunities into your project Claude.md
- Adjust trigger conditions for your workflow
- Add project-specific examples to the exercises
- Change the soft cap on exercises per session
- Add domain-specific retrieval check-in questions
- Explore adding evaluation checks to assess how successfully this skill is fulfilling its instructions
Background
This skill was developed based on learning science and informed by multiple qualitative interviews with software development professionals about their concerns around agentic coding, as part of my open science empirical evidence about developer thriving and skill development in AI-assisted workflows. In my research with thousands of developers, I've also found that a strong value and commitment to learning predicts that developers feel less threat, worry and anxiety when imagining needing to adjust to agentic coding. Learning culture also associates with increases in team effectiveness overall, not just individual productivity.
I'd love to know if you enjoy this and what you learn! Sharing open science resources helps researchers like me create more things to help software teams. I always appreciate a shout-out or a share in public, which helps more people learn about the psychology of software teams. Get updates and access to more of the psychology of software teams at my newsletter: Fight for the Human
Authors
Learning-Opportunities:
Dr. Cat Hicks
I'm a psychological scientist studying software teams and technology work, an author, a public speaker, a research architect, and an empirical interventionist who builds radical research teams that put answers behind questions everyone is asking but few people are gathering real evidence about.
- Website: drcathicks.com
- Software Team & Eng Leadership Consulting: catharsisinsight.com
- Upcoming Book: The Psychology of Software Teams (2026)
Orient:
Dr. Michael Mullarkey
I'm a machine learning engineer who used to be a therapist + social science researcher. I'm thinking a lot about how to leverage agentic AI to help people learn skills, see blendtutor for another example.
Sources
- Bjork, R. A., Dunlosky, J., & Kornell, N. (2013). Self-regulated learning: Beliefs, techniques, and illusions. Annual review of psychology, 64(1), 417-444.
- Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students’ learning with effective learning techniques: Promising directions from cognitive and educational psychology. Psychological Science in the Public interest, 14(1), 4-58.
- Ericsson, K. A., Hoffman, R. R., & Kozbelt, A. (Eds.). (2018). The Cambridge handbook of expertise and expert performance. Cambridge University Press.
- Giebl, S., Mena, S., Storm, B. C., Bjork, E. L., & Bjork, R. A. (2021). Answer first or Google first? Using the Internet in ways that enhance, not impair, one’s subsequent retention of needed information. Psychology Learning & Teaching, 20(1), 58-75.
- Hicks, C. M., Lee, C. S., & Foster-Marks, K. (2025, March 15). The New Developer: AI Skill Threat, Identity Change & Developer Thriving in the Transition to AI-Assisted Software Development. https://doi.org/10.31234/osf.io/2gej5_v2
- Kalyuga, S. (2007). Expertise reversal effect and its implications for learner-tailored instruction. Educational psychology review, 19(4), 509-539.
- Kang, S. H. (2016). Spaced repetition promotes efficient and effective learning: Policy implications for instruction. Policy Insights from the Behavioral and Brain Sciences, 3(1), 12-19.
- Kornell, N. (2009). Optimising learning using flashcards: Spacing is more effective than cramming. Applied Cognitive Psychology: The Official Journal of the Society for Applied Research in Memory and Cognition, 23(9), 1297-1317.
- Murphy, D. H., Little, J. L., & Bjork, E. L. (2023). The value of using tests in education as tools for learning—not just for assessment. Educational Psychology Review, 35(3), 89.
- Roediger III, H. L., & Karpicke, J. D. (2006). The power of testing memory: Basic research and implications for educational practice. Perspectives on psychological science, 1(3), 181-210.
- Rohrer, D., & Taylor, K. (2007). The shuffling of mathematics problems improves learning. Instructional Science, 35(6), 481-498.
- Skulmowski, A., & Xu, K. M. (2022). Understanding cognitive load in digital and online learning: A new perspective on extraneous cognitive load. Educational psychology review, 34(1), 171-196.
- Soderstrom, N. C., & Bjork, R. A. (2015). Learning versus performance: An integrative review. Perspectives on Psychological Science, 10(2), 176-199.
- Sweller, J., & Cooper, G. A. (1985). The use of worked examples as a substitute for problem solving in learning algebra. Cognition and instruction, 2(1), 59-89.
- Tankelevitch, L., Kewenig, V., Simkute, A., Scott, A. E., Sarkar, A., Sellen, A., & Rintel, S. (2024, May). The metacognitive demands and opportunities of generative AI. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems (pp. 1-24).
- Hicks, C. (2025). Cognitive helmets for the AI bicycle: Part 1. Fight for the Human. https://www.fightforthehuman.com/cognitive-helmets-for-the-ai-bicycle-part-1/
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
About
A Claude or Codex skill for deliberate skill development during AI-assisted coding
Resources
License
CC-BY-4.0 license
Stars
2.2k