Skip to content

CHEN2003-CHIP/LearnLikeMe

Repository files navigation

LearnLikeMe

LearnLikeMe logo

Turn real chat history into a personalized project-learning copilot with Codex.

用真实聊天记录,为自己生成一个个性化的项目学习副驾。

OpenAI Agent Skill Personalized Learning History Driven

LearnLikeMe is a Codex-friendly OpenAI Agent Skill package that turns local chat history into a reusable, personalized project-learning assistant.

LearnLikeMe 是一个面向 Codex 的 OpenAI Agent Skill 项目,目标是把本地聊天记录转成可复用、个性化的项目学习助手。

Instead of forcing everyone through the same generic tutor flow, LearnLikeMe treats chat history as the default evidence source, asks only the minimum necessary follow-up questions, and generates a runtime skill that teaches codebases in the learner's own style.

它不会默认先做完整问卷,而是优先从历史聊天记录中提炼偏好,只在证据不足、互相矛盾或低置信度时才补问。

Why This Is Different | 为什么不一样

  • History-first personalization, not questionnaire-first personalization

  • Stable preferences and situational preferences are separated

  • Generated outputs are written to local files, not only shown in chat

  • The generated runtime skill is proactive, roadmap-driven, and Codex-friendly

  • Privacy defaults to local-first: history and generated user-specific skills stay out of Git by default

  • 先看历史聊天,再补问,而不是先做一整套问卷

  • 区分稳定偏好和场景偏好

  • 生成结果会落到本地文件,而不是只显示在聊天里

  • 生成出的 runtime skill 是主动式、路线图驱动、适合 Codex 使用的

  • 默认隐私优先:历史记录和用户专属生成结果默认不进 Git

What You Get | 你会得到什么

  • Builder skill: learn-like-me

  • Generated runtime skill: project-learning-assistant

  • Default generated skill path: generated-skills/project-learning-assistant/

  • Default generated profile path: generated-skills/learning-profile.md

  • History-first personalization from Markdown chat exports

  • Fixed runtime output contract:

    • project map
    • current understanding
    • hidden questions
    • next best step
  • builder skill:learn-like-me

  • 生成出的 runtime skill:project-learning-assistant

  • 默认 skill 输出目录:generated-skills/project-learning-assistant/

  • 默认 learning profile 输出文件:generated-skills/learning-profile.md

  • 基于 Markdown 历史聊天记录的个性化分析

  • 固定的运行时输出合同:

    • project map
    • current understanding
    • hidden questions
    • next best step

Default Workflow | 默认工作流

1. Clone the repository | 克隆仓库

git clone https://github.com/CHEN2003-CHIP/LearnLikeMe.git
cd LearnLikeMe

2. Put your materials into the input folder | 把材料放进输入目录

Drop exported Markdown chats into:

learn-like-me-input/history/

Optional files:

  • learn-like-me-input/preferences.md
  • learn-like-me-input/context.md

把导出的 Markdown 聊天记录放到:

learn-like-me-input/history/

可选补充:

  • learn-like-me-input/preferences.md
  • learn-like-me-input/context.md

3. Open the repo in Codex | 用 Codex 打开仓库

Make sure Codex can read:

  • learn-like-me/
  • learn-like-me-input/history/
  • learn-like-me-input/preferences.md
  • learn-like-me-input/context.md
  • generated-skills/

确保 Codex 能读取:

  • learn-like-me/
  • learn-like-me-input/history/
  • learn-like-me-input/preferences.md
  • learn-like-me-input/context.md
  • generated-skills/

4. Give Codex the starter prompt | 把启动提示词发给 Codex

Use CODEX-STARTER-PROMPT.md.

直接使用 CODEX-STARTER-PROMPT.md

5. Let Codex generate the profile and skill | 让 Codex 生成画像和 skill

Codex should:

  1. read history first
  2. infer preferences with confidence
  3. ask only minimal follow-up questions
  4. write the profile to generated-skills/learning-profile.md
  5. write the runtime skill to generated-skills/project-learning-assistant/

Codex 应该:

  1. 先读历史聊天
  2. 带置信度地推断偏好
  3. 只问最少的补充问题
  4. 把画像写到 generated-skills/learning-profile.md
  5. 把 runtime skill 写到 generated-skills/project-learning-assistant/

6. Use the generated runtime skill | 使用生成后的 runtime skill

After generation, tell Codex to use the generated skill directly, for example:

Use generated-skills/project-learning-assistant to help me learn this repository.
Start with a project map, then guide me step by step.

生成后,直接这样让 Codex 使用它:

请使用 generated-skills/project-learning-assistant 带我学习这个仓库。
先给我 project map,再一步一步讲。

Output Contract | 输出合同

Every runtime response must end with:

  • project map
  • current understanding
  • hidden questions
  • next best step

每次 runtime skill 的回答都必须以这四个部分结尾:

  • project map
  • current understanding
  • hidden questions
  • next best step

Repository Structure | 仓库结构

.
|- learn-like-me/
|  |- SKILL.md
|  |- agents/
|  |- references/
|  `- assets/
|- learn-like-me-input/
|  |- history/
|  |- preferences.md
|  `- context.md
|- generated-skills/
|  `- project-learning-assistant/
|- CODEX-STARTER-PROMPT.md
`- skill.zip

Privacy by Default | 默认隐私策略

User chat history under learn-like-me-input/history/ stays local by default.

User-specific generated outputs under generated-skills/ also stay local by default.

放在 learn-like-me-input/history/ 下的用户聊天记录默认只用于本地分析。

放在 generated-skills/ 下的用户专属生成结果也默认只保留在本地。

Sample vs Generated Output | 示例产物与真实产物

The sample-generated skill under learn-like-me/assets/sample-generated-skill/ is a reference artifact.

The active generated output should live under generated-skills/.

learn-like-me/assets/sample-generated-skill/ 里的内容只是示例快照。

真正的本地生成结果应该写到 generated-skills/

Core Idea | 核心理念

The best project-learning assistant should not start from zero.

It should start from how you already learn.

最好的项目学习助手,不应该从零开始。

它应该从“你本来就是怎么学习的”开始。

About

帮助用户按自己的节奏学习代码仓库、架构和模块的个性化 Agent Skill。

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors