How to use the ljg-is agent skill to explain any concept

What does the ljg-is skill make the agent do?

Many explanations stop at a definition. You may remember the word without knowing how to recognize it in practice. ljg-is asks the agent to connect the two: first say what the thing is and where its boundary lies, then show how it operates, then note which of your earlier judgements no longer holds, and finally give guidance for the next time you meet it.

The skill is written in Chinese, and its description is in Chinese. It is meant for technical concepts, institutions, products, roles, methods, norms and practices. It is not meant for plain fact lookups, formal proofs, step-by-step software tutorials, deep causal analysis or structural analysis, which the skill routes to other skills (ljg-think and ljg-structure).

• Says what the thing is and how it differs from its closest neighbours
• Walks through how it works using a concrete example
• Identifies a specific assumption the reader should revise
• Gives practical judgement cues for a real situation

Install ljg-is from the ZIP download

Download the ZIP of the skill from this page. Unzip it so the ljg-is folder (containing SKILL.md, Tools/, Workflows/ and evals/) ends up inside your skills directory. For a personal install, use ~/.claude/skills/. For a single project, use the .claude/skills/ folder in that project's root.

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

• Personal install: ~/.claude/skills/ljg-is
• Project install: .claude/skills/ljg-is inside your project
• Claude apps: Settings → Capabilities → Skills, then upload

How do you trigger the skill with a natural request?

You do not need a special command. The skill's description lists the kinds of requests it fits: questions about what X is, how to understand it, what it implies, why it matters, and how to judge or use it later. Ask in ordinary language and name the concept clearly.

Requests that do not fit are better answered directly. A question asking only for a formal definition, a proof or a simple translation will not need this skill, and the skill's own boundary rules say so.

• What is a feature flag, and how is it different from a config setting?
• I know a KPI is a target metric, but I still can't tell one in practice. Help me understand it.
• What does 'technical debt' mean for how I should plan my sprint?
• Explain what a code of conduct does in an open source project and how I should judge one.

Where does the output go, and what does a good note contain?

By default the skill writes a note to a local ~/Context folder, named with a timestamp and the concept, in the pattern ~/Context/{timestamp}--理解-{target}__is.org. The note is titled 理解：<concept>. The source describes the default file as an Org file, while the validator also accepts Markdown notes.

A good note reads as one line of reasoning, not as four labelled sections. Headings come from the content, and the explanation should let you name the concept's category, reproduce how it works, state what changed in your thinking, and point to a real situation where you would act differently.

Check a note with the ValidateNote validator

The Tools folder includes ValidateNote.ts, a TypeScript validator that checks the structure of a generated note. It requires the Bun runtime. Run it on a note by passing the file path. The example below assumes the skill is installed at ~/.agents/skills/ljg-is, which is the path the source uses, so adjust it if your install lives under ~/.claude/skills/.

On success it prints a status object to stdout and exits with code 0. On failure it prints an array of errors to stderr and exits with code 1. If no file argument is given, it exits with code 2. The validator is a rough structural check, not proof that the Chinese prose reads naturally, so read the full note yourself afterward.

• Exit 0: success, status ok printed to stdout
• Exit 1: validation errors printed to stderr
• Exit 2: no file path given

Run the validator's self-test

The validator ships with its own test file. You can run it with Bun to confirm the validator works in your environment before relying on it.

What are the safety notes: scripts and network use?

The skill runs scripts: the ValidateNote validator is a TypeScript program that you run with Bun, and it is the only script the source describes. The source states the skill does not use the network. It writes files to your local ~/Context folder.

Because the validator is a script, read it before running it if you want to verify what it does. Keep in mind that the skill's output is written to your disk, so check the destination folder if you care where notes end up.

• Runs scripts: yes (ValidateNote.ts, executed with Bun)
• Uses network: no, according to the source
• Writes to: local ~/Context folder

Tips for getting better explanations

Name the concept precisely and say where you met it. A term from a specific tool or workplace gives the agent a boundary to work with. If you already have a rough idea, say what you think it means, so the note can show exactly which assumption needs revising.

Use the skill for judgement, not procedure. If you want exact steps to configure software, ask for a tutorial instead, because ljg-is gives cues for deciding, not operating instructions.

• State the concept and the situation where you encountered it
• Say what you currently believe it means
• Ask for judgement cues, not step-by-step instructions
• Run the validator, then read the whole note