Learn Anything Fast

中文说明

Turn any topic into a source-grounded, measurable learning loop.

learn-anything-fast is a portable Agent Skill for learning a topic deeply enough to use it independently. It turns a vague goal such as “learn AI agents” into a practical loop of mapping, retrieval, practice, correction, compression, teach-back, and external verification.

The skill treats “10x” as an aspiration, not a promise. Progress is demonstrated with observable evidence: working artifacts, test results, quizzes, explanations, and independent tasks.

Compatibility

The portable core is the SKILL.md file plus the relative references/ directory. The optional agents/openai.yaml file adds Codex/ChatGPT UI metadata and is ignored by hosts that do not use it. This repository has no API-key, MCP, or runtime dependency of its own.

Codex desktop builds may also expose ~/.codex/skills/; use the /skills list or the host documentation to confirm the active path. Only one Codex location is needed.

What It Does

• Builds a level ladder from a learner’s current ability to independent performance.
• Identifies the highest-leverage 20% of concepts and separates what to learn now, recognize, or defer.
• Creates time-boxed learning plans with milestones, practice tasks, deliverables, and advancement gates.
• Coaches with one question or task at a time, without revealing the answer first.
• Diagnoses knowledge, reasoning, execution, and careless errors in a compact weakness ledger.
• Produces one-page cheat sheets only from material the learner has actually encountered.
• Uses Feynman teach-backs to find vague, missing, or jargon-hidden understanding.
• Calibrates claims against primary sources, official documentation, and real artifacts.

Learning Modes

The skill selects the smallest mode that fits the request:

The Learning Loop

Map -> Focus -> Retrieve -> Practice -> Correct -> Compress -> Teach -> Verify
  ^                                                               |
  +------------------------- adapt from evidence -----------------+

The learner produces more than the AI. Familiarity alone is not treated as mastery: advancement requires a defined threshold, such as two successful retrieval rounds plus an independent implementation task.

Install

Clone this repository into the host’s skills directory. Replace the destination with the path for your host from the table above:

mkdir -p ~/.agents/skills
git clone https://github.com/zgl610329-wq/learn-anything-fast.git ~/.agents/skills/learn-anything-fast

For an existing installation, update it with:

git -C ~/.agents/skills/learn-anything-fast pull

If the host does not discover the skill immediately, reload its skills list or restart the host. Review the skill before enabling it in a trusted project, especially if future versions add scripts or tools.

Install And Use By Host

Codex

Codex’s current user-level skills directory is ~/.agents/skills/; repository skills live in .agents/skills/. Some desktop builds also expose ~/.codex/skills/, so verify with /skills.

mkdir -p ~/.agents/skills
git clone https://github.com/zgl610329-wq/learn-anything-fast.git ~/.agents/skills/learn-anything-fast

Use it explicitly:

$learn-anything-fast Help me learn Python async programming in 30 minutes per day and become able to build reliable async services.

Codex can also load it automatically when the request matches the description in SKILL.md.

Claude Code

Claude Code discovers personal skills under ~/.claude/skills/ and project skills under .claude/skills/.

mkdir -p ~/.claude/skills
git clone https://github.com/zgl610329-wq/learn-anything-fast.git ~/.claude/skills/learn-anything-fast

Use it with:

/learn-anything-fast Design a 30-minute-per-day plan to master Python async programming.

Claude Code can also activate it automatically when the request matches the skill description.

GitHub Copilot In VS Code

VS Code supports project skills in .github/skills/, .claude/skills/, and .agents/skills/, plus personal skills under ~/.copilot/skills/.

From the root of your project:

mkdir -p .github/skills
git clone https://github.com/zgl610329-wq/learn-anything-fast.git .github/skills/learn-anything-fast

Then open the Chat view and use:

/learn-anything-fast for a 30-minute-per-day Python learning plan

Or ask a matching learning question and let Copilot load the skill automatically. The /skills menu can list and manage discovered skills.

Gemini CLI

Gemini CLI discovers user skills in ~/.gemini/skills/ and project skills in .gemini/skills/.

mkdir -p ~/.gemini/skills
git clone https://github.com/zgl610329-wq/learn-anything-fast.git ~/.gemini/skills/learn-anything-fast

Inside Gemini CLI, refresh and inspect the skill list:

/skills reload
/skills list

Then use it explicitly:

Use the learn-anything-fast skill to create a source-grounded roadmap for learning RAG in 30 minutes per day.

Other Agent Skills-Compatible Hosts

Copy the repository directory into the host’s documented skills location. The portable entrypoint is always SKILL.md; references/prompts-and-templates.md is loaded when the skill asks for a roadmap, quiz ledger, cheat sheet, or teach-back review.

If a host does not support the Agent Skills format, it can still use the project manually: paste the contents of SKILL.md into its system/project instructions and provide the references/ files as supporting context. Automatic discovery, slash commands, and progressive loading will not be available in that case.

Use It

Call the skill explicitly:

使用 $learn-anything-fast 帮我学习 Python 异步编程，每天 30 分钟，目标是能独立写出可靠的异步服务。

You can also request a specific mode:

使用 $learn-anything-fast 的 Diagnose 模式，测试我对 RAG 的理解。

For a complete learning program:

使用 $learn-anything-fast 帮我从零掌握 AI Agent 应用开发，要求有项目、测验、晋级门槛和每周复盘。

Repository Layout

learn-anything-fast/
├── SKILL.md                         # Portable skill instructions
├── agents/
│   └── openai.yaml                  # Optional Codex/ChatGPT metadata
├── references/
│   └── prompts-and-templates.md      # Output templates and state formats
├── README.md                        # English documentation, default landing page
├── README.zh-CN.md                  # Chinese documentation
└── LICENSE

Design Principles

1. Evidence over feeling. “I read it” is not completion evidence.
2. Practice before explanation. Try to retrieve or build before seeing a polished answer.
3. One useful correction at a time. Repair the smallest gap that blocks progress.
4. Source-grounded claims. Prefer primary and official sources; label uncertainty and inference.
5. Adaptive scope. Stop expanding the curriculum when the target performance is demonstrated.
6. Human agency. The AI structures practice and feedback; the learner does the thinking and building.

Contributing

Useful contributions include:

• New domain-specific examples and practice tasks.
• Better rubrics for diagnosing common learning errors.
• Official-source resource paths for additional topics.
• Translations that preserve the learning-loop behavior, not just the wording.
• Bug reports with the prompt, expected behavior, actual behavior, and relevant context.

Keep templates concrete and testable. Avoid turning the skill into a generic motivational coach or a passive summarizer.

License

MIT. See LICENSE.

Official Host Documentation

• Agent Skills specification
• Codex: Build skills
• Claude Code: Extend Claude with skills
• VS Code: Use Agent Skills
• Gemini CLI: Tools and skills