Learning Practice Coevolution

Learning Practice Coevolution

让 AI 从“帮你总结书”变成“陪你练会一个东西”。

learning-practice-coevolution is not a generic summarizer. It helps you reconstruct your understanding first, then lets the agent critique, repair concepts, add examples, create transfer practice, and bring learning back into a real task.

learning-practice-coevolution 不是通用总结器。它让你先重构理解，再让 Agent 批改、修补概念、补充例子、设计迁移练习，并把学习带回真实任务。

Quick Start / 快速上手

If your agent can already use this skill, start with this:

Use $learning-practice-coevolution.

I am reading: {book / article / PDF / course note / concept}
My real purpose is: {why I am reading it}

Do not summarize first.
First ask me 1-2 necessary questions, then make me reconstruct the idea in my own words.
After that, critique my understanding and give me one transfer exercise for a real task.

中文：

使用 $learning-practice-coevolution。

我正在读：{书 / 文章 / PDF / 课程笔记 / 概念}
我的真实目的：{我为什么要读}

请不要直接总结。
先问我 1-2 个必要问题，然后让我用自己的话重构理解。
之后批改我的理解，并给我一个能用到真实任务里的迁移练习。

If you have not installed it yet, you can first paste SKILL.md or SKILL.zh-CN.md as task material. Move it into your agent's skill folder after you decide it is useful.

Can I Use It With Only An Agent?

Yes.

Minimum setup:

Agent + this Skill + user-provided material

Better setup:

Reader + Knowledge Base + Agent + this Skill

How Do I Give Material To The Agent?

Do not ask for a whole-book summary first. Give the agent material, purpose, and scope:

give material -> state purpose -> set scope -> reconstruct first -> critique -> transfer practice -> leave feedback -> continue next round

In practice, give it three things:

1. Material it can actually see: table of contents, chapter, highlights, notes, PDF pages, or your own reflection.
2. Why you are reading: concept understanding, project problem, writing preparation, or systematic learning.
3. This round's scope: if the whole book is too large, start with one chapter, one highlight set, or one concept.

Boundary:

The agent does not automatically know what is in your reading app.
If integration exists, use it and ask the agent to check access first.
If not, provide excerpts, notes, highlights, table of contents, PDF pages, or your own reading reflection.

Common Material Sources

Problems This Helps With

You finished a book and feel you understood it, but cannot use it.

The skill should not summarize for you. It should first make you explain the idea in your own words, then critique that explanation, then give you a work or project exercise.

You face a difficult concept and do not know where to begin.

Tell the agent your purpose and current level. It can split the topic into small lessons, then adjust difficulty, examples, and direction from your feedback.

You want to bring reading into a real project.

Give it both the project task and the material. The agent should treat the task as the exercise and find the most relevant parts of the material.

You want to preserve a method without freezing it too early.

Methods, SOPs, prompts, or skill ideas stay as candidates until repeated practice shows they are actually useful.

Do not use it as a generic summarizer.

Agent's First Questions

After material is provided, the agent should not summarize first. It should ask:

1. Why are you reading this? What real question does it serve?
2. What is the material scope: whole book, chapter, highlights, notes, PDF pages, or concept?
3. What output do you want: understanding check, chapter learning, transfer exercise, project material, reading card, serial lesson, or method candidate?

Copy-Ready Prompt Recipes

Chinese prompt recipes for WeRead highlights, PDF chapters, finished books, real project transfer, and serial learning are in USAGE.zh-CN.md.

Example for serial learning:

Use $learning-practice-coevolution.

I want to learn systematically: {topic}
Design a serial learning path.
Generate only one small lesson at a time.
After each lesson, give me 3 mastery checks and a feedback area.
The next lesson must read my feedback first, then adjust difficulty, examples, and direction.

Material Incompleteness Boundary

The agent should not pretend it has read the whole book.

• If only a table of contents is provided, it can make a reading map.
• If only highlights are provided, it can run understanding checks around those highlights.
• If only the title is provided, it should ask for a table of contents, chapter excerpt, notes, or user question.
• If a PDF is only partially readable, it should mark the gap and work only with visible content.

Install In Obsidian

You can install the optional plugin directly from Obsidian:

1. Open Obsidian Settings.
2. Go to Community plugins and turn off Restricted mode.
3. Click Browse.
4. Search for Learning Practice Coevolution.
5. Click Install, then Enable.

After enabling the plugin, open the command palette and search for Learning Practice. You can:

• Start learning session
• Insert reconstruction prompt
• Insert Copilot mentor prompt
• Insert transfer and misuse test
• Close as observer note

If Browse does not show the plugin yet, install it manually from GitHub:

1. Download manifest.json, main.js, and styles.css from release 0.1.2.
2. Put them under <your-vault>/.obsidian/plugins/learning-practice-coevolution/.
3. Restart Obsidian and enable Learning Practice Coevolution in Community plugins.

The Obsidian plugin is local-first. It pairs well with Copilot for Obsidian, but it does not depend on Copilot, read Copilot settings, call any AI API, collect telemetry, or make network requests.

See PRIVACY.md for the local-only privacy boundary.

What You Get In The Public Version

The public version already includes the core learning workflow: active recall, AI critique, transfer practice, serial lessons, learning feedback, and method candidates.

It does not include private BP-Wiki project paths, dashboards, private learning logs, business project examples, connector settings, internal Runtime governance, secrets, account traces, or non-public knowledge.

Files

• SKILL.md - the main skill.
• SKILL.en.md - English skill snapshot.
• SKILL.zh-CN.md - Chinese skill version.
• README.md - bilingual public onboarding entry.
• README.en.md / README.zh-CN.md - language-specific documentation.
• USAGE.zh-CN.md - Chinese usage flow and prompt recipes.
• SUPPORT.md - support and payment links.
• agents/openai.yaml - optional OpenAI-facing display metadata.
• assets/ - QR code assets for public support channels.
• manifest.json / main.js / styles.css / versions.json - Obsidian plugin assets.
• PRIVACY.md - local-only plugin privacy boundary.
• skill-package-manifest.json - skill package metadata and file hashes.
• LICENSE - MIT License.
• CHANGELOG.md - public release notes.

Project Article

• Chinese project article on WeChat: Read the article

Support

If this skill helps your reading, learning, or real-work practice loop, you can support future maintenance through the public payment links below.

These are external manual support channels. The Alipay image is a user-provided public payment QR image. This repository does not include payment passwords, API keys, bank-card data, or hidden checkout automation. See SUPPORT.md for the full payment boundary.

License

MIT License. Copyright (c) 2026 Jie Huang.

Attribution and license notice must be preserved when redistributing or adapting this skill.