How to use the ljg-teach agent skill: install and learn by scenario

What the skill makes the agent do

Once installed, the agent stops giving you a definition first. The skill follows a loop: scenario, choice, surprise, structure, naming, then a transfer scenario on a different surface. You make a decision in a concrete situation and explain why. The agent then advances the situation one step so that your old model no longer predicts the outcome.

The concept is named only after you can see the structure. Then the agent checks whether your understanding holds in a new scenario that looks different on the surface. The skill is written to run as a conversation in Chinese.

• Starts with a scenario that has a role, a goal, a few constraints and enough information to decide.
• Asks one question at a time and waits for your answer.
• Names the concept only after your reasoning has been mirrored back to you.
• Ends a round when you explain the reasoning correctly in two different surfaces, or when you say stop.

Install ljg-teach from the ZIP download

Download the ZIP for the skill from the page you are reading. Unzip it so that the skill folder, which contains the SKILL.md file, sits inside your skills directory. For a personal install, use ~/.claude/skills/. For a single project, use the .claude/skills/ folder inside that project.

If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the skill there.

• Personal install: ~/.claude/skills/ (available across your projects).
• Project install: .claude/skills/ inside the project folder.
• Claude apps: Settings → Capabilities → Skills, then upload.
• The package lists one file: SKILL.md. No other files are required.

Trigger the skill with a command or a natural request

You can invoke the skill explicitly with /ljg-teach followed by the concept, for example a concept name or a claim you want to test. You can also trigger it with everyday phrases the skill is written to recognise, such as 用情境教我 (teach me with a situation), 别先讲答案 (don't give the answer first), or 通过案例让我学会 (help me learn through cases).

The skill also applies when you simply want to understand an idea through choices and feedback, even if you do not use those exact words.

• Explicit: /ljg-teach 机会成本 (opportunity cost).
• Natural: 用情境教我 plus the idea you want to learn.
• Natural: 别先讲答案, if you want to skip the definition.
• Claim testing: put a belief in quotes, for example 「透明总会带来信任」.

Example sessions

Learning a concept: you ask for opportunity cost. The agent gives you a resource trade-off and asks you to choose and explain. After you answer, it shows the real cost of the option you gave up, mirrors the rule you were using (counting only money already spent), and only then names the concept. A new scenario with time allocation tests whether you can apply it without a hint.

Testing an opinion: you ask whether transparency always builds trust. The agent presents a team decision record and asks how transparent it should be. The surprise shows how visibility can change participants' incentives to perform. The agent separates factual predictions, value weights and institutional constraints rather than declaring transparency right or wrong.

What it needs and how it runs

The skill lists no required tools or dependencies. It does not run scripts and does not use the network, according to its package description. Its only file is SKILL.md, which contains the instructions the agent follows.

Because the skill is a set of instructions, the work happens in the conversation itself. Expect the agent to ask you questions and wait for replies rather than produce a finished document in one message.

• Requirements: none listed.
• Scripts: none.
• Network access: none.
• Files: SKILL.md.

Safety notes: scenarios may be simplified

The scenarios are teaching tools and may be invented simplified hypothetical cases. Do not treat a scenario as a verified real-world fact. If an empirical fact is needed to create a surprise, the skill is meant to verify its source first, and if it cannot, to say that the case is a simplified hypothetical.

Do not use this skill if you need a verified factual answer quickly. Scenarios can also be a poor fit for fact-heavy subjects where you need the full explanation right away.

• Treat scenarios as hypothetical unless the agent cites a source.
• Say 换局面 (change the scenario) or 揭晓 (reveal) if you want a different path or the answer now.
• You can say 停止 (stop) at any time to end the round with a short summary.

Tips for getting more from a session

Answer with your actual reasoning, even if you are unsure. The agent treats your reason as evidence of your current model, so a short honest reason helps more than a guess at the expected answer.

Ask for a hint if you are stuck. A hint reduces only one degree of freedom, not the whole problem. Keep one concept per session; several frameworks at once make it hard to see which variable changed your prediction.

• Give your reasoning in one or two sentences.
• Use 给一点提示 (give a small hint) instead of asking for the answer.
• Use one concept per session.
• Use 继续 (continue) to move on, or 总结 (summarise) to close the round.